Booklet Text.indd

A bi-annual, refereed journal published byALHOSN University - Abu Dhabi - UAE

Volume 5 Number 1 January 2013

ChairmanProf. Abdulla Ismail

EditorProf. Abdul Rahim Sabouni

Associate EditorProf. Raymond Tennant

MembersDr. Abdellatif QamhaiehDr. Adel KheliDr. George MarkouDr. Mama ChachaDr. Marko Savic

Managing EditorDr. Al Haj Salim Mustafa

Address:P.O. Box : 38772Abu Dhabi - UAETel. : +971 2 4070700Fax : +971 2 4070799E-mail : [email protected] : www.alhosnu.ae

ISSN 2076-8516

CONTENTS

ARTICLES

LINEARITY AND NONLINEARITY IN HYDROLOGIC RESPONSE IN ARID AND SEMIARID WATERSHEDS 5Rafik Al-Sakkaf

EMBEDDED REINFORCEMENT MESH GENERATION METHOD FOR LARGE-SCALE RC SIMULATIONS: CASE STUDY 23George Markou

AN ENHANCED MODEL FOR NUMERICAL INVESTIGATION OF MASS TRANSPORT IN AN OPERATIONAL PEM FUEL CELL 43Rihab Jaralla, Jun Cao, Mama Chacha, Tawfiq J. Jaber

UAE TRAFFIC ACCIDENTS TRENDS AND INSIGHTS OVER THE PAST DECADE 65Sharaf A. Alkheder, Reem Sabouni, Hany El Naggar

A GIS-BASED PLANNING STUDY FOR A GEO-ENVIRONMENTAL DISTRICT SELECTION IN NORTH JORDAN 87Sharaf A. Alkheder, Ahmad A. Al-Malabeh, Mohammad N. Sada

AHU J. of Engineering & Applied Sciences 5 (1) 2013© 2013 ALHOSN University

5


LINEARITY AND NONLINEARITY IN HYDROLOGIC RESPONSE IN ARID AND SEMIARID WATERSHEDS

Rafik Al-Sakkaf

Department of Civil Engineering, ALHOSN University, P. O. Box 38772, Abu Dhabi, UAE. E-mail: [email protected]

ABSTRACT: The issue of linearity and nonlinearity of watershed response has been raised by hydrological researchers in relation to watershed scale. Hydrologic response is usually assumed to approach linearity with increasing watershed scale. To investigate the validity of the assumption in arid and semiarid environments, a lumped model based on the Nash cascade of linear reservoirs was applied to a number of sub-watersheds in the semiarid Walnut Gulch Experimental Watershed in southern Arizona. By using the Nash cascade of linear reservoirs, the rainfall-runoff relation was modelled. The Nash model was parameterized by the method of moments. Net rainfall was calculated by using three loss models. The rainfall-runoff model was found to be applicable to watersheds at particular scales. Results showed that as the watershed scale decreased, the hydrologic response approached linearity. In arid and semiarid watersheds, variability in space and time is intrinsic in the convective storm events which yield runoff that may even vanish through infiltration in the channel bed before reaching the watershed outlet. Two main sources of nonlinearity might be transmission losses and spatial variability of rainfall. In arid and semiarid watersheds, nonlinearity of watershed response in larger areas might require a distributed modelling approach at a warranted complexity level. The model should allow for spatial variability of rainfall and provide for transmission losses effects on the hydrograph.

KEYWORDS: Hydrologic response, Rainfall-runoff modelling, Lumped model, Linear reservoir, Nash model, Arid regions, Semiarid regions, Watershed scale

1. INTRODUCTION

Hydrologic response is usually assumed to approach linearity with increasing watershed scale [28]. The issue of linearity and nonlinearity of watershed response has been raised by hydrological researchers in literature [28, 25, 9]. In more than on publication the issue was discussed in relation to semiarid zones. Minshal [16] and Wang et al [30] showed that nonlinearty in watershed response decreased with increasing watershed area in humid regions. Wang et al [30] studied nonlinearity in watershed response in humid zones using a transfer function. Goodrich et al [9] approached the problem by invoking geomorphological models and statistical properties of watersheds. Goodrich et al [9] showed that linearity of watershed response increased with decreasing watershed area in arid and semiarid regions, which Sivapalan et al [28] suggested to be defined as scaling relationship. In contrast, they suggest that the conventional definition for nonlinearity be kept for the dynamic relationship in rainfall-runoff response. Vivoni et al [29] discussed nonlinearities in the rainfall-runoff transformation and its scale-dependence.

Runoff hydrographs of single events are characterised by the steep, almost perpendicular, rising limb of the hydrograph contrasted with a longer gradually fading falling limb. Therefore, time to peak flow is very short relative to storm duration. It may be in the order of a few minutes. It simply reflects the

RAFIK AL-SAKKAF

6

high intensity of rainfall events, thus dictating short time steps for simulation, so that peak flow is not missed. Simulating event hydrographs entails accounting for losses in the way they progress during the event. However, that is very difficult to represent in a working form. Another dimension of the difficulty is the size of the watershed where a model can be used. Cantón et al [2] provided an overview of the key factors and processes that influence runoff generation in semiarid environments including watershed scale.

Modelling response to rainfall stimulus in arid and semiarid watersheds is generally elusive. The usual response is in the form of flash floods, which are initiated by localised high-intensity-short-duration rainfall storm events. Variability in space and time is intrinsic in the convective storm events which yield runoff that may even vanish through infiltration in the channel bed before reaching the watershed outlet [32]. Flow in the channels is ephemeral, leaving the bed dry for prolonged periods. For a given event, the whole watershed may not be activated, and that is mostly the case. At the event scale, infiltration losses are always substantial relative to yield volume. Initial and transmission losses (continuing abstractions along the path of the flood wave) are two significant processes in such watersheds. Owing to these losses, only a small fraction of rainfall is transformed into runoff.

Complex models have been tested in semiarid regions and did not prove to be significantly superior to simpler ones, considering the data, parameter and computational requirements [14, 4, 5]. But the question remains as to what is the appropriate watershed size that a model can represent and simulate the hydrograph. Thus, simple conceptual models should be tested in a complex environment. Then possibly a watershed size, or a maximum size, may be defined such that the lumped model works. Elaborating a model by parameterisation may require further calibration.

Another advantage is the ability of a conceptual model to describe a group of physical processes in a simplified approximate manner, yet it establishes a cause-and-effect relationship between input and output. Thus, a degree of empiricism is also involved, which is justified by presently missing knowledge on a given physical attribute of a watershed. By studying a range of values of a given parameter in a conceptual model, one can come to a generalisation that a given model is applicable to conditions that prevail when such a range is replicated.

One widely-used conceptual watershed model, a Nash cascade of linear reservoirs [27], is based on the concept of linking a number, n, of linear reservoirs in series, whereby the outflow from the first is the inflow to the second and the outflow from the second is the inflow to the third and so on. Finally, the outflow from the last reservoir is the watershed outflow. A linear reservoir is an element whose outflow is directly proportional to storage. Principally, a linear reservoir provides a certain amount of attenuation with an associated lag. Subsequently, each reservoir in the series is assumed to have the same lag, k, and provides the same amount of attenuation. A cascade of linear reservoirs is a conceptual model with two parameters, n and k. n is the number of linear reservoirs that a given volume of water has been routed through and k the time scale (time lag) of each reservoir; thus nk is the watershed lag. n and k are estimated by the method of moments. The analytical solution of the cascade of linear reservoirs is the Nash model, which has been used in this analysis.

The objective of this paper is to establish a relationship between watershed size and the dynamic relationship between rainfall and runoff in semiarid regions by using a transfer function, namely the IUH, as derived by Nash [18, 19] for a cascade of linear reservoirs. In this paper analysis of the characteristics of runoff hydrographs in a semiarid watershed was performed. By using the Nash cascade of linear reservoirs, the rainfall-runoff relation was modelled. The Nash model was parameterised by the method of moments. Net rainfall was calculated by using three loss models, for the sake of comparison.


7

The model was applied at the Walnut Gulch Experimental Watershed in Arizona, USA. The model was applied in the main watershed (WG01) and in three of its sub-watersheds (WG11, and WG04, WG111).The model was found to be applicable to watersheds at particular scales. Results showed that as the watershed scale decreased, the hydrologic response approached linearity. Sources of nonlinearity were identified.

2. MATHEMATICAL MODELLING

2.1 The Cascade of Linear Reservoirs

Representing the watershed by an activated reservoir series is the underlying assumption in conceptualising the response of the watershed to rainfall stimulation. For the case at hand, the Nash cascade of linear reservoirs is adopted to describe the rainfall-runoff relation.

A linear reservoir has a storage that is related to its output linearly by a storage constant (time scale, lag time) (K)

KQS = (1)where:

S: the storage termK: is the storage constant (time scale, lag time)Q: is discharge

QIdtdS

−=

(2)

Substituting for S from Equation 1 and integrating Equation 2 gives,

( ) ckt

QI +−=−ln

(3)

When S(0)=1 and I(t) =0 for t>0, we obtain IUH denoted by u(0,t), which is the impulse response function for a linear reservoir,

( ) k

t

ek

tu−

=1

,0

(4)

A watershed can be represented by an n-series of identical linear reservoirs that have the same time scale (k); that is to say a cascade of linear reservoirs, also known as the Nash Cascade. When a unit volume of water is routed through the n-linear reservoirs, the instantaneous unit hydrograph (IUH) can be derived [6].

Discharge from the first reservoir in the cascade would be the inflow to the following and so on. By applying the convolution integral to Equation 4, the outflow of the 1-th reservoirs is given by:

dte

kq

t

kt

∫−

=0

11

(5)

RAFIK AL-SAKKAF

8

If the time since the beginning of an event until the input (q1) began is estimated and q1 is the input function, then the outflow from the second reservoir is defined, and so on. Convoluting successively through n reservoirs, or, alternatively, routing the unit volume through n reservoirs the IUH (qn) for the cascade is obtained as:

ekt

q k

t

n

n kn−

−

⎟⎠

⎞⎜⎝

⎛Γ

=

)1(

)(

1 (6)

Equation 6 has two parameters; k and n. �(n) is the gamma function. In other words, the IUH is a gamma distribution function. To estimate the parameters n and k, first and second moments of net rainfall and discharge about the origin are computed.

2.2 Estimating the Parameters n and k

As mentioned above, a Nash cascade is composed of a series of similar linear reservoirs that can assume a value of any real number (n), with the same lag time (k). The lag time (time scale) for the whole watershed, designated by K in Equation 1, is equivalent to the product term (nk). In this case nk is the first moment of the IUH about the origin. In discrete time intervals, the first and the second moments of the IUH (a unit volume of flow) about the origin are as follows [6, 17]:

MIUH1 = nk (7)

MIUH2 = n(n+1)k2 (8)

Further it can be shown that:

MQ1 - MI1 = nk (9)

MQ2 - MI2 = n(n+1)k2 + 2nk MI1 (10)

where:MI1: first moment of net rainfall hyetograph about the originMI2: second moment of net rainfall hyetograph about the originMQ1: first moment of runoff hydrograph about the originMQ2: second moment of runoff hydrograph about the origin

The values of MI1, MI2, MQ1, and MQ2 were estimated from the rules of mechanics by using the discrete time interval runoff hydrographs and hyetographs. By substituting the estimated moments in Equations 7 and 8 the parameters n and k could be calculated. The product nk is the difference of moments that is equal to the distance between the centriods of rainfall hyetograph and runoff hydrograph. After n and k were determined, the IUH was constructed by applying Equation 6. Further, the IUH was integrated to determine the TUH. Finally, by convolution of the IUH and the effective rainfall the runoff hydrograph was calculated using the following convolution integral.

0

( ) ( ) ( )t

Q t I t q t dτ τ= −∫

(11)


9

I(t): Intensityq(t��): IUH function�: elapsed timet: time

2.3 Modeling Approach

2.3.1 Lumped Linear Model

The principal advantage of lumped modelling is the minimal data requirement. It is easier to estimate the parameters, but then accuracy of representation might be sacrificed. In semiarid watersheds, spatial and temporal variability of rainfall is always important. Heterogeneity of watershed characteristics has also a significant effect on hydrologic parameters [1, 32]. Nonetheless, lumped modelling should not be excluded; rather it should be investigated to define the limitations.

2.3.2 Modelling Assumptions

It was assumed that the response of the watershed to the rainfall stimulation was linear. A Nash cascade was formulated for the watershed by assuming that the lag time (nk) was the difference between the centroid of effective rainfall heytograph and the centroid of runoff hydrograph. The effective rainfall volume was found by equating rainfall volume to the total runoff volume, after subtracting applicable losses. Effective rainfall distribution depended on the assumed loss model as discussed in section 2.3.3.

2.3.3 Loss Models

Calculation of losses relied on balancing effective rainfall volume with discharge volume. Initial losses were calibrated to yield effective rainfall volumes equal to discharge volumes. Effective rainfall was calculated by invoking three loss models: a constant loss model, proportional loss model and a continuing loss model after subtracting an initial loss [21, 13]. Proportional losses were estimated by a volumetric runoff coefficient. Proportional loss model accounts also for the translation as the rainfall starts before the initiation of the hydrograph.

In the proportional loss model one uses a runoff coefficient that is calculated as the ratio of input and output. As the intensity of rainfall was assumed constant during each time interval, according to this loss model runoff must be produced from each interval. Nonetheless, the proportional loss model did not err by much as compared to the continuing loss model.

Effective rainfall did not always result from the highest intensities in the storm. Nonetheless, for the constant loss model, the high intensities would generate runoff. In the case of the other two loss models, the net rainfall was always the volume left after subtracting initial losses.

3. CASE STUDY WATERSHED

Located in south eastern Arizona, USA, the area of Walnut Gulch Experimental Watershed is about 148 km2. It is approximately located around 110o W, 31o45’N. Topography is comprised of rolling hills and some steep terrain, ranging in elevation between 1190 and 2150 m+msl. Cattle grazing and recreational activities are the major land uses, although some urbanization exists in and around the town of Tombstone. Vegetation within the watershed is composed primarily of grassland and shrub-

RAFIK AL-SAKKAF

10

steppe rangeland vegetation. The geology is dominated by a thick alluvial fan (about 400 m in depth) contributing to the San Pedro River with abundant groundwater at levels ranging from a few meters to 100m in depth [23, 24].

Stream channels within the Watershed have been recently incised and are ephemeral [12]. Rainfall is bimodal; a summer and a winter mode. The latter mode is a result of frontal storms in the winter months. These frontal storms have low intensities (< 25 mm/hr) and do not usually cause runoff [10]. However, most runoff events occur as a result of high intensity, short duration summer storms. In the summer the storms are convective and are limited spatially with intensities higher than 25 mm/hr [15, 26]. With an average annual rainfall of 324 mm and an annual mean temperature of 17.6 0C in the city of Tombstone, the area’s climate is classified as semiarid [24].

4. IMPLEMENTING THE MODEL

4.1 Data Preparation

The original records had values of rainfall or runoff at irregular intervals. Rainfall and runoff records were interpolated to produce fixed-time-interval records for each storm. The events were interpolated using time intervals starting at one minute and increasing by unity up till five minutes. The reason for small time intervals was to make sure that the peak flow was not missed. The model was tested for all five time intervals for accuracy and it was proved to be insensitive to the time intervals.

Isohyetal maps of storm events were prepared using GIS. The purpose was to analyze the spatial variations to help in storm selection (See Figure 1 and Figure 2). By investigating the isohyetal maps it can be seen that there is a pattern that defines the shape of the hydrograph, depending on the location of the storm. The storm distribution and the size of storm peak determine the size and the time of peak flow.Since storms in arid and semiarid watersheds are localized, it was difficult to find storms that covered all considered sub-watersheds simultaneously.

Data files were prepared for the whole Walnut Gulch Watershed draining at Flume 1 (WG01) (148 km2) and for three of its sub-watersheds draining into Flume 4 (WG04) (2.3 km2), Flume 11 (WG11) (8.4 km2) and Flume 111 (WG111) (0.6 km2). The number of storms analysed was 29 for WG01, 25 for WG04 and 26 for WG11 and 12 for WG111. WG111 has a shorter measurement record than the other ones.

Storm 27/08/1982

5

15

10

2025

30

65

5

50

1025

6035

35

5

40

50

15

40

20

30

0

4555

Fig. 1: Rainfall depth isohyets for storm 27/08/1982


11

Storm 27/08/1996

15 20

2510

30

35

40

4550

55

6065

45

2535

50

20

5

0

2025

0

0

30

40

30

10

15

40

15

25

1035

45

5

Fig. 2: Rainfall depth isohyets for storm 27/08/1996

4.2 Criteria for Storm Selection

Criteria of storm selection were dictated by the pattern of rainfall and runoff in the case study watershed. It was necessary to select standard events that represent the usual hydrologic response of the watershed. An example of a standard event is shown in Figure 3. These criteria are outlined as follows.

• mostly individual peak hydrograph structure,

• continuous distribution of rainfall over the time span of the storm with discontinuities in some cases,

• direct runoff peak were in the order of a few cubic meters per second, so that the flow is mainly on the surface

4.3 Coding the Model

A computer programme to apply the Nash linear model procedure was prepared by using MATLAB. A script and a user interface for input and output were prepared for the model. Output was in the form of graphs and data files.

The model computer code calculated losses from total rainfall to produce effective rainfall. The value of effective rainfall was calibrated to the volume of discharge by trial and error. Trial and error attempts were based on fitting the calculated and observed hydrographs. By running the model several times the calculated hydrograph was fitted to the observed one. Three error measures were used. Root mean square error (RMSE), mean absolute error (MAE) and the coefficient of efficiency (CE) [20]. The formulas for the three measures are as follows.

RMSE=((Oi- Pi)2/ n)0.5 (1)

MAE =|Oi- Pi|/ n (2)

CE =1-((Oi- Pi)/(Oi- )2) (3)

where:O: observedP: calculated

:average observed valuesi: index

RAFIK AL-SAKKAF

12

Storm of 27/08/1982

0

20

40

60

80

100

120

0 60 120 180 240 300 360 420 480

minutes

m3/s

ec

Fig. 3: Runoff hydrographs of 27/08/1982 as measured in three consecutive flumes (WG06, WG02, and WG01) along a reach length of 10 km, in Walnut Gulch Experimental Watershed

5. DISCUSSION OF RESULTS

5.1 Model Performance with Respect to Scale

Acceptable model results were obtained in the smaller sub-watersheds. While in the main watershed the model was for the most part unsuccessful. Unacceptable results were possibly due to unexplained routing effects and the assumption that hydrologic response was linear. Model output for some events is shown in a Table 1 for each watershed area.

Characteristics of the hydrograph of the main watershed differed slightly from those in the sub-watersheds. However, it seems that the model performed better in the sub-watersheds. Rainfall uniformity might be higher in the smaller areas as shown by the uniformity coefficients (Figure 4). The uniformity coefficient (CU) was calculated according to Christiansen [7] as follows.

r

r

−= ∑

∑avgP P

PCU

(4)

Pr: raingage depth,Pavg: average rainfall depth,r: index

Such uniformity and scale effects might be the reason for minimising the errors of averaging and lumping in small scales. The coefficient of efficiency was higher in the smaller areas (Figure 5). Obviously, the linear response assumption at small scale might be valid as the model has demonstrated better results at that scale.


13

Table1: Linear model output

Event Date

(yymmdd)

Rainfall Intensity (mm/hr)

Average Q

(mm/hr)

nk (min)

Observed Peak Flow

(m3/sec)

Calculated Peak Flow (m3/sec)

Time to

Peak %

Initial Loss

(mm/hr)MAE

RMSE (m3/sec)

Coefficient of

Efficiency

Linear model output for WG01 (148 km2)

900801 3.3 0.1 112 12.7 10.1 13 24 0.2 0.7 0.91

910827 3.6 0.1 70 19.0 12.7 13 12 0.2 1.1 0.91

Linear model output (WG11) (8.4 km2)

940829 5.8 0.4 17 3.9 3.6 8 35 0.03 0.22 0.92

960710 8.9 0.6 15 7.4 5.6 8 55 0.04 0.23 0.96


940825 2.9 0.9 40 1.9 1.8 0 79 0.02 0.08 0.98

940906 9.1 0.6 15 1.6 1.5 11 41 0.01 0.06 0.98


930805 11.9 1.5 16 0.8 0.6 13 55 0.00 0.03 0.98

930819 26.5 9.5 20 8.1 6.8 0 130 0.03 0.19 0.99

Note: MAE : mean absolute error RMSE : root mean square error

In the small watershed areas, the method of moments might be more suitable to single peak events that result from high intensity rainfall events. It is limited by the temporal and spatial distribution of rainfall intensity [3]. Thus, the cases of high intensity short duration were more suitable to the method. In the case when the method failed it might be due to location of effective rainfall hyetograph, which might intersect with runoff hydrograph. Zoccatelli et al [33] indicated the influence of the spatial rainfall variability on runoff modelling for a small catchment size.

In some cases the total volume of the hydrograph was reasonably matched with the observed one in the large area. However, the peak and the time to peak did not match well in the large watershed that might be due to the variability of rainfall. It might be that the variability of rainfall limits the applicability of the method of moments. The method of moments calculates the watershed lag from an average rainfall hyetograph. This hyetograph is averaged over a large area, which is not necessarily activated by the storm. On the contrary in the small areas more rainfall uniformity was observed, leading to better results (Figure 6).

Calculation of losses in all areas assumed only infiltration losses. The loss models did not provide for transmission losses. In the case of the small areas these losses were not significant. Transmission losses are more significant at the large watersheds (See Figure 3). Although sub-watershed WG11 was not very large, high transmission losses were observed there [9]. The runoff hydrograph in a large area travels a considerable distance and loses water to channel infiltration. The linear Nash model routes the hydrographs without considering transmission losses. That might be the other reason for the better performance of the model in small areas.

RAFIK AL-SAKKAF

14

The transmission losses have less effect on the response in an arid catchment as the size diminishes [11]. Variability of rainfall may be present even at the smaller catchment scale in arid zones, although lessened by the size of the rain cells [31, 1, 32].

Uniformity Coefficients

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15 20 25 30

Storm Number

%

WG111 WG04 WG11 WG01

Fig. 4: Uniformity coefficients

Coefficient of Efficiency

0.70

0.80

0.90

1.00

0 5 10 15 20 25 30

Storm Number

WG111 WG04 WG11 WG01

Fig. 5: Coefficients of efficiency


15

Efficiency Coefficient versus Uniformity Coefficient

0.6

0.8

1

0 20 40 60 80 100

Uniformity Coefficient %

Eff

icie

ncy

Co

effi

cien

t WG111

WG01

Fig. 6: Efficiency coefficient variation with uniformity coefficient

5.2 Peak Flow

In the successful cases in the main watershed area, the calculated hydrographs were dominated by underestimated peak flow and the lag in response. The underestimation of peak might be ascribed to the routing effect of the linear reservoir [22]. A linear reservoir translates and attenuates the hydrographs during routing. Hydrographs of selected events at the modelled watershed areas are shown in Figure 7 through Figure 10. Comparison of calculated and observed peak flows at all watershed areas are shown in Figure 11 through Figure 14.

25/07/1983 (WG01)

2.0

1.5

1.0

0.5

0.00 60 120 180 240 300 360

minutes

mm

0.02.04.06.08.010.012.014.0

m3/s

ec

Losses Effective Rainfall

Observed Calculated

15/09/1983 (WG11)

5.0

4.0

3.0

2.0

1.0

0.00 30 60 90 120 150

minutes

mm

0.0

2.0

4.0

6.0

8.0

m3/s

ec


Observed Calculated

Fig. 7: Calculated and observed hydrographs Fig. 8: Calculated and observed hydrographs

19/08/1993 (WG111)

18.0

15.0

12.0

9.0

6.0

3.0

0.00 30 60 90 120 150

minutes

mm

0.0

2.0

4.0

6.0

8.0

10.0

m3/s

ec


Observed Calculated

01/09/1984 (WG04)

12.0

10.0

8.0

6.0

4.0

2.0

0.00 30 60 90 120 150 180

minutes

mm

0.00.51.01.52.02.53.03.54.0

m3/s

ec


Observed Calculated

Fig. 9: Calculated and observed hydrographs Fig. 10: Calculated and observed hydrographs

RAFIK AL-SAKKAF

16

In general, the peak flow values were more accurately calculated in the smaller areas. In the case of WG04 (2.3 km2), the calculated matched observed peaks more accurately that the other cases. The assumption of linearity might hold at a certain size of watershed area. Alternatively, it may be fairly accurate such that it reciprocates little error for large simplicity.

5.3 Time to Peak

In the small watersheds the time to peak was simulated with greater accuracy than the peak flow. For the model to compensate for simulating the rising limb linearly, continuity was satisfied by deducting mass from the peak flow. Physically the watershed storage does not build up gradually; it is rather augmented in a short time by the flash flood. The magnitude of the “short time” is in the order of a few minutes. As mentioned earlier the storms were usually localized. In the case of the large areas the time of travel of the hydrograph would be longer, thus the routing effects and the transmission losses would be more significant. The flow wave gets abstracted at a high rate as transmission losses at the beginning of flow. The model did not provide for this process. In addition the routing effects might also be the reason for the symmetrical calculated hydrographs in WG01.

In Figure 7 the uniform rising limb of the calculated hydrograph at WG01 almost matches the uniform decline. In arid watersheds, most storm events produce the sudden steep rising limb and the gradual decay of the falling limb. The linear reservoir effect is clear on the hydrograph.

5.4 Watershed Lag Time

The values of k and n resulted from the assumptions of the loss calculations. So, the k and n values would reflect all the errors in assumptions. In some cases, when the rainfall continued beyond the hydrograph time base, the calculations resulted in effective rainfall that occurred after the centroid of the hydrograph. In turn that resulted in negative net moment (thus negative values of nk). In other words, MI1

and MI2 exceeded MQ1 and MQ2. In that case a negative watershed lag is calculated which is not acceptable. This is one limitation in the method of moments. It might be worthwhile to analyse the differences when k and n are adjusted by trial and error calibration. Such approach might be feasible with single peak storms.

Calculated Peak Flow (Qcal) vs Observed Peak Flow (Qobs) (WG01)

020406080

100120

0 20 40 60 80 100 120

Qobs (m3/sec)

Qca

l (m

3/s

ec)

1:1


02468

1012

0 5 10 15

Qobs (m3/sec)

Qca

l (m

3/s

ec)

1:1

Fig. 11: Calculated versus observed peaks Fig. 13: Calculated versus observed peaks


17


0

2

4

6

8

10

0 2 4 6 8 10

Qobs (m3/sec)

Qca

l (m

3/s

ec)

1:1


0

5

10

15

20

25

0 5 10 15 20 25

Qobs (m3/sec)

Qca

l (m

3/s

ec)

1:1

Fig. 12: Calculated versus observed peaks Fig. 14: Calculated versus observed peaks

For some storms in the large areas results were not reasonable. The method of moments calculates the difference in time between the occurrence of the centriods of the effective rainfall hyetograph and the hydrograph. For the method to yield reasonable results the centroid of rainfall must occur prior to the centroid of runoff. It describes a cause-and-effect relationship, which is in accordance with the assumptions of the linear system theory. In some cases the difference of time/moments was negative, which is not in accordance with the model assumptions. That outcome might be due to possible errors in data and to errors in associating rainfall-runoff events. It seems that parameter estimation by the method of moments was sensitive to synchronization errors.In general the variations in watershed lag time value with peak flow are lower in the smaller watersheds, WG11, WG04 and WG111. The watershed lag time (nk) is plotted versus the peak flows as shown in Figure 15 through Figure 18. A declining trend is more apparent in WG01 than the other three smaller areas. It might also imply that linearity might be assumed at smaller scales.

Lag (nk) vs Calculated Peak Flow (Qcal) (WG111)

0

10

20

30

40

50

0 2 4 6 8 10

Qcal (m3/sec)

nk

(min

ute

s)


0

10

20

30

40

50

0 2 4 6 8

Qcal (m3/sec)

nk

(min

ute

s)

Fig. 15: Lag versus calculated peaks Fig. 17: Lag versus calculated peaks


0

20

40

60

0 5 10 15 20

Qcal (m3/sec)

nk

(min

ute

s)


0

50

100

150

200

250

0 20 40 60 80

Qcal (m3/sec)

nk

(min

ute

s)

Fig. 16: Lag versus calculated peaks Fig. 18: Lag versus calculated peaks

RAFIK AL-SAKKAF

18

5.5 Uncertainty in Estimation of Losses

Regarding loss models, none of them proved to be better in improving the fit of the hydrograph. In most cases, the continuing loss model yielded slightly better results than the other two and mostly when the initial loss rate was more than the rainfall intensity. It is clear that the losses were high, which is expected in such cases. It may not be always true that the highest intensity of rainfall would cause the runoff. The soil profile may have not reached its saturated conditions at that point in time. At that high intensity, the soil infiltration rate may not have been exceeded.

In the case of the proportional loss model only some storms in WG01 yielded usable results. The rest did not work due to the extended duration of the rainfall relative to the runoff.

Transmission losses were not provided for in the invoked loss models. As shown in the example in Figure 3, and in many other cases, it is clear that the hydrograph was almost halved during transmission between the three stations, although the area between the two stations received part of the storm.

None of the loss models was significantly superior to the other in the case of the large area. The similarity in performance might be explained by the short duration of storms. The distribution of losses over a short time span did not change the difference in the second moments of effective rainfall hyetograph and runoff hydrograph by an apparent magnitude. This was valid for the successful cases, while for the cases that did not succeed; the events might have produced non-physical negative moment difference.

5.6 Classification of Model Output

Responses can be assembled in three groups, depending on the value of the watershed lag parameter: nk. In one group one finds the relatively high value nk, exceeding 200 min. In the other comes the group with nk at values less than a 100, and at the third the group of values at about 50.

For each group, certain traits are observed. In the first group the simulated hydrograph barely fits the observed one. The reasons might be the excessive second moment (MI2 and MQ2) that leads to negative values of n. In other words, moment variations are magnified by the square of the second moment, which arise from large storm lag. While the second group comes with a lower nk value, it calculates a low peak flow as compared to the third one. The peak is mostly more than 30% in error of the measured one.The third group was close to simulating the hydrograph, with a value of nk less than 100. In the case of the double peak, it was invariably not possible to fit both because of the break of rainfall within events that lead to runoff; in other words; in the case of non-continuous storms. In the case of storms with two rainfall peaks, the simulated hydrograph was relatively close to the measured one.

6. CONCLUSION

The watershed was modelled as a cascade of linear reservoirs using the Nash approach. The single-event modelling approach is used to model the hydrograph of flash floods in Walnut Gulch Experimental Watershed. The model has been used for the Walnut Gulch main watershed and some of the sub-watersheds. Applicability of the model at the smaller scale watersheds was found more accurate. The main finding was that the model performed better in the small areas of the watershed rather than the large areas. It may indicate that linearity increases at smaller watershed sizes.

Furthermore, the variations in the spatial distribution of rainfall are sufficiently extreme. Lumping might lead to deleting the effect of rain peaks. Runoff producing areas are different for every storm, due to


19

spatial variability of rainfall. Transmission losses reduced the volume of the hydrograph in the large basins. Based on those observations in arid and semiarid regions, linearity of response might be assumed at smaller basins, rather than large basins.

Peak flow value and time were difficult to simulate. In the large area, error in the time to peak was not less than the error of magnitude. The start of the hydrograph rising limbs did not match in most cases even with the initial-continuing loss model.

Both peak flow and the shape of the hydrograph are related to the location of the raincells within the storm event upstream. It was observed that the location of the storm in the watershed influenced the shape of the hydrograph. In other words, the distribution of rainfall has an effect on the timing and shape of the hydrograph in the large watershed. In the smaller areas the rainfall distribution was sufficiently uniform not to affect the hydrograph.

Errors were also caused by the underlying lumped linear assumptions of watershed response. It is assumed in the linear theory that every impulse must have a response. Net rainfall was the impulse to cause the response. In the calculation of losses it was assumed that the highest intensities were the cause of runoff. It might not be the case always, especially in the large areas. Furthermore, in arid watersheds that might not be the case since initial and transmission losses have different effects on runoff generation. In this model, losses are considered as one single process. On the contrary transmission losses occur in the channel bed after initial infiltration losses are satisfied and runoff has accumulated.

Two sources of nonlinearity might be identified as transmission losses and spatial variability of rainfall. When routing the hydrographs, transmission losses should be accounted for in arid and semiarid watersheds. Rainfall spatial variability is significant in such areas. All these processes might be the causes of nonlinearity in watershed response. Modelling an arid watershed by the linear lumped approach may not properly describe the watershed response. Therefore, a distributed approach may be the viable alternative, at a warranted complexity level. An alternative modelling approach is required. The model shouldprovide for transmission losses effects on the hydrograph, allow for spatial distribution of rainfall input and route excess rainfall non-linearly.

7. ACKNOWLEDGEMENT

The author acknowledges the cooperation of the staff members of the USDA-ARS Southwest Watershed Research Center in Tucson, Arizona, USA for providing the data.

8. REFERENCES

[1] Bahat, Y., Grodek, T., Lekach, J., and Morin, E. (2009). Rainfall-runoff modeling in a small hyper-arid catchment, Journal of Hydrology, Vol, 373, pp 204–217.

[2] Cantón , Y., A. Solé-Benet , J. de Vente , C. Boix-Fayos , A. Calvo-Cases , C. Asensio , J. Puigdefábregas (2011). A review of runoff generation and soil erosion across scales in semiarid south-eastern Spain, Journal of Arid Environments, Vol. 75, (12), pp 1254–1261.

[3] Chang, C. L. (2007): Influence of moving rainstorms on watershed responses, Environ. Eng. Sci., Vol. 24, pp 1353–1360.

RAFIK AL-SAKKAF

20

[4] Chiew, F., M.J. Stewardson, T.A. McMahon (1993). Comparison of Six Rainfall Runoff Modelling Approaches, Journal of Hydrology, Vol. 147, pp 1-36.

[5] Chiew, F.H.S., P.H. Whetton, T.A. McMahon, and A.B. Pittock, (1995). Simulation of the impacts of climate change on runoff and soil moisture in Australian catchments, Journal of Hydrology, Vol. 167, pp 121-147.

[6] Chow, V. T., D. R. Maidment, and L. Mays (1988). Applied Hydrology, McGraw-Hill, New York.

[7] Christiansen, J.E.(1941). The uniformity of application of water by sprinkler system, Agricultural Engineering Journal, Vol. 22, pp 89-92.

[8] Goodrich, D. C. (1990). Geometric Simplification of a Distributed Rainfall-Runoff Model Over a Range of Basin Scales. PhD. Dissertation, University of Arizona, Tucson.

[9] Goodrich, D.C., Leonard J. Lane, Rose M. Shillito, and Scott N. Miller, Kamran H. Syed, David A. Whoolhiser (1997). Linearity of Basin Response as a Function of Scale in a Semiarid Watershed, Water Resources Research, Vol. 33, (12), PP. 2951-2965.

[10] Kennedy, Jeffrey; Keefer Tim, Paige Ginger, Barnes Frank (2003). Evaluation of Dielectric Constant-Based Soil Moisture Sensors in a Semiarid Rangeland, In Renard K.G., Kenneth G., McElroy, Stephen A., Gburek, William J., Canfield, H. Evan and Scott, Russell L., eds. 2003. First Interagency Conference on Research in the Watersheds, October 27-30, 2003. U.S. Department of Agriculture, Agricultural Research Service

[11] Knighton, A. David and Gerald C. Nanson (2001). An event-based approach to the hydrology of arid zone rivers in the Channel Country of Australia, Journal of Hydrology, Vol. 254, (1-4), pp.102-123.

[12] Lane, L.J. (1982). A Distributed Model for Small Semi-arid Watersheds, Journal of the Hydraulics Division,ASCE 108: 1114-1131.

[13] Laurenson, E.M.and Mein, R.G. (1997). RORB-Version 4, User Manual. Department of Civil Engineering, Monash University, Clayton, Victoria, Australia.

[14] Michaud, J. and Sorooshian, S. (1994). Comparison of Simple Versus Complex distributed Runoff Models on a Semi-arid Watershed, Water Resources Research, Vol. 30(3): pp 593-605.

[15] Miller, S.N., Hernandez, M. and Lane, L.J. (1997). GIS Application in the Spatial Extrapolation of Hydrological Data from Experimental Watersheds, SWRC Publication, Tombstone, Arizona.

[16] Minshall, N.E. (1960). Predicting Storm Runoff on Small Experimental Watersheds, Journal of the Hydraulics Division, Vol. 86, (8), pp 17-38.

[17] Mutreja, K.N. (1986). Applied Hydrology, TATA McGraw-Hill.

[18] Nash, J. E.(1957). The Form of the Instantaneous Unit Hydrograph, International Association of Scientific Hydrology Publication, Vol. 45(3), pp 114-121.

[19] Nash J. E.(1960). A unit hydrograph study, with particular reference to British catchements,


21

Proceedings, Institution of Civil Engineers, Vol. 17, pp.249-282.

[20] Nash, J.E. and J.E. Sutcliffe (1970). River flow forecasting through conceptual models, Part 1-A discussion of principles, Journal of Hydrology, Vol .10(3): pp 282-290.

[21] Pilgrim D.H. and I. Cordery (1993). Flood Runoff in Maidment, D.R (ed.) Handbook of Hydrology, McGraw-Hill, Inc.

[22] Ponce, Victor Miguel (1989). Engineering Hydrology; Principles and Practices, Prentice-Hall Englewood Cliffs.

[23] Renard, K.G. and Nichols, M.H. (2003). History of Small Watershed Research in Non-Forested Watersheds in Arizona and New Mexico, Proceedings of the First Interagency Conference on Research in the Watersheds, 27-30 October 2003, Benson, Arizona.

[24] Renard, K.G., Lane, L.J., Simanton, J.R., Emmerich, W. E., Stone, J.J., Weltz, M. A., Goodrich, D. C. and Yakowitz, D.S. (1993). Agricultural Impacts in an Arid Environment: Walnut Gulch Studies, Hydrologic Sciences and Technology. Vol. 9 (1- 4): pp 145-190.

[25] Savenije, H.H.G. (2001). Equifinality, a Blessing in Disguise?, Hydrological Processes, 15: 2835-2838.

[26] Scott, R.L., W.J. Shuttleworth, T.O. Keefer, and A.W. Warrick. (2000). Modeling multiyear observations of soil moisture recharge in the semiarid American Southwest. Water Resources Research 36(8):2233-2247.

[27] Singh, V.P. (1988). Hydrologic Systems Vol. 1. Rainfall-runoff Modeling, Prentice-Hall, Englewood Cliffs.

[28] Sivapalan, M., C. Jothityangkoon, and M. Menabde (2002) Linearity and nonlinearity of basin response as a function of scale: Discussion of alternative definitions, Water Resources Research Vol. 38(2): pp 4-1 to 4-5.

[29] Vivoni, E.R. D. Entekhabi, R.L. Bras, and V.Y. Ivanov (2007). Controls on runoff generation and scale-dependence in a distributed hydrologic model, Hydrol. Earth Syst. Sci., Vol. 11, pp 1683-1701.

[30] Wang, C.T., C.V. Gupta and E. Waymire (1981). A Geomorphologic Synthesis of Nonlinearity in Surface Runoff.Water Resources Research Vol. 17(3): pp 545-554.

[31] Woolhiser, D. A., and D. C. Goodrich (1988). Effect of storm rainfall intensity patterns on surface runoff, Journal of Hydrology. Vol. 102: pp 335-354.

[32] Yakir, H. and E. Morin, (2011). Hydrologic response of a semi-arid watershed to spatial and temporal characteristics of convective rain cells, Hydrol. Earth Syst. Sci., Vol. 15, pp 393-404.

[33] Zoccatelli, D., Borga, M., Zanon, F., Antonescu, B., and Stancalie, G. (2010). Which rainfall spatial information for flash flood response modelling? A numerical investigation based on data from the Carpathian range, Romania, Journal of Hydrology, Vol. 394(1–2).

23


EMBEDDED REINFORCEMENT MESH GENERATION METHOD FOR LARGE-SCALE RC SIMULATIONS: CASE

STUDY

George Markou

Department Of Civil Engineering, ALHOSN University, P.O. Box 38772, Abu Dhabi, UAE.E-mail: [email protected]

ABSTRACT: Modeling of reinforced concrete structures through the use of 3D detailed models derives significant numerical issues especially when dealing with large-scale meshes that incorporate large numbers of reinforcement bars embedded in the hexahedral mesh. In 3D detailed reinforced concrete simulations, mapping the reinforcement grid inside the concrete hexahedral finite elements is performed using the end-point coordinates of the rebar macro-elements. This procedure is computationally demanding when dealing with large-scale models, where the required computational time for the reinforcement mesh generation can be excessive. This research work investigates the numerical robustness and computational efficiency of the embedded rebar mesh generation method proposed by Markou [14] that was an extension of the Markou and Papadrakakis [8] research work. The under study embedded rebar mesh generation method foresees the automatic allocation and generation of embedded steel reinforcement inside hexahedral finite elements for 32-bit and 64-bit windows based applications. In order to further investigate the numerical and computational performance of the embedded rebar mesh generation method, a full-scale model of the RC frame of the ALHOSN University Campus in Abu Dhabi of the United Arab Emirates is constructed and used so as to allocate and generate the embedded rebar finite elements. The numerical results illustrate the computational efficiency and robustness of the under study method. Finally, the numerical results that derived from the ReConAn FEA solver for the at hand numerical implementation are briefly presented.

KEYWORDS: Embedded Reinforcement, Mesh Generation, Large-Scale Models.

1. INTRODUCTION

Modeling of reinforced concrete (RC) structures with the use of 3D detailed models is usually performed by research teams [1-5] or by large consultancy companies [6-7] that foresee the thorough investigation of the mechanical behavior of geometrically complicated RC structures. Researchers have been using this modeling type so as to verify experimental results or develop new constitutive models in an attempt to derive a numerically objective modeling method that will eventually provide the ability of performing assessment analysis for any type of RC structure design.

As it was described in [8], when modeling three-dimensional RC structures with the finite element (FE) method, three main approaches are available for the simulation of reinforcement: smeared, discrete, and embedded [9-11]. The smeared and discrete formulations have been found to be unsuitable for complicated reinforcement grid geometries thus undergo several restrictions when implemented. On the other hand, the embedded reinforcement formulation provides the ability of representing the grid’s geometry in an exact manner without the need of modifying the actual arrangement of rebars to conform with the concrete FE mesh [12-13].

GEORGE MARKOU

24

The approach of allocating and generating the embedded rebars inside hexahedral elements that was proposed by Barzegar and Maddipudi [9], which is an extension of the work of Elwi and Hrudey [10], has the advantage of allowing arbitrary positions for the rebars inside the concrete elements and a free geometry for each hexahedral element. In order to optimize the performance of the above embedded mesh generation method, Markou and Papadrakakis [8] proposed the introduction of a geometrical constraint in order to decrease the computational effort for generating the input data of the embedded rebar elements, particularly when dealing with relatively large-scale numerical models. Their proposed method (Fig. 1) was incorporated in ReConAn FEA [5] which was developed and built to run in a 32-bit operating system.

Fig. 1. Flow chart of the embedded rebar mesh generation method [8]

EMBEDDED REINFORCEMENT MESH GENERATION METHOD FOR LARGE-SCALE RC SIMULATIONS

25

The purpose of this research work, is to investigate deeper the numerical performance of the embedded mesh generation method proposed by Markou [14] that was an extension of the Markou and Papadrakakis [8] research work. So as to achieve this task, the updated Markou and Papadrakakis method is used [14] in order to generate the embedded rebar finite elements inside the hexahedral element mesh of the RC frame of the ALHOSN University’s Campus located at Abu Dhabi, UAE. The method, which is incorporated in ReConAn FEA [5], was built to run in both 32-bit and 64-bit windows based operating systems.

2. GENERATING REINFORCEMENT INSIDE HEXAHEDRAL ELEMENTS

The under study embedded mesh generation method considers arbitrary positioning of the rebars inside the concrete elements [8], as shown in Fig. 2, while avoiding a nonlinear search procedure for the calculation of the natural coordinates of the embedded reinforcement nodes in the corresponding prismatic hexahedral elements. By separating the generation algorithm into to two main parts (Fig. 1), the geometry of each hexahedral element is categorized (prismatic or non-prismatic) and accordingly treated in order to compute the natural coordinates of its containing embedded rebar elements.

Fig. 2. Embedded reinforcement rebars inside hexahedral elements [8].

Through a pre-processing software code, the initial mesh generation is performed consisting of the concrete elements and the rebar macro-elements defined by their end nodes. Then the coordinates of the macro-elements’ end-nodes are used to generate the numerical model of the embedded rebar elements inside each concrete hexahedral FE. The generation of embedded rebar elements is performed for each rebar macro-element separately. This requires an independent search for each rebar macro-element, which is performed in order to detect all intersections of its straight part with the surrounding solid elements. In order to decrease the required computations of the embedded rebar generation procedure, a geometrical constraint is introduced, aiming at locating faster the embedded rebar elements inside the corresponding concrete hexahedral elements. For the detailed description of this method, refer to the relative reference [8].

GEORGE MARKOU

26

Fig. 3. Flow chart of the updated embedded rebar element mesh generation method [14].


27

As it was mentioned above, the embedded reinforcement mesh generation method proposed in [8], was integrated and built in a 64-bit operating system so as to overcome the problem that rises related to the physical memory allocation issue, which in the case of the 32-bit architecture is limited to 2 Gb. In addition to that the method was integrated with a filtering algorithm [14] that provides the ability to allocate relatively short embedded rebar finite elements, during the mesh allocation procedure. This filtering procedure provides the ability to control the mesh quality of the embedded rebar mesh thus avoiding the numerical phenomenon of the local stiffness concentration due to the relatively very small embedded rebar finite element length. The flowchart diagram of the integrated embedded reinforcement mesh generation algorithm can be seen in Fig. 3.

3. NUMERICAL IMPLEMENTATION

In this section, a parametric investigation and a numerical verification of the efficiency of the integrated embedded reinforcement mesh generation method is presented. In order to achieve this task, the ALHOSN University’s Male Campus is used for constructing the 3D detailed mesh. Finally, the results from the embedded mesh generation procedure are discussed and the numerical results that derived from the solution of the model will be presented. It must be noted that all numerical tests were performed on a 64-bit windows operating system (3.3GHz processor).

3.1 Geometrical Features and Reinforcement Details of the RC Frame

The RC building that was used so as to construct the 3D mesh is shown in Fig. 4, while the geometry of its RC frame is given in Fig. 5. As it can be seen, the building has a total width of 20.45m and a 25.6m length. The total height of the building is 13.2m (Ground floor, 1st-3rd floor), while the basement of the structure is also accounted for in the under study model (Fig. 6).

Fig. 4. ALHOSN University’s Male Campus in Abu Dhabi, UAE.

In the UAE, the most common framing system used is that of the flat slabs, which is also the framing system adopted in this research work. As it can be depicted in Figs. 5 & 7, the shear wall of the elevator is located in the center of the structure, while a total of 16 columns are placed symmetrically about the structure’s core. The geometry of the columns’ sections are 40x40cm and 60x60cm. The thickness of the slab is assumed to be equal to 30cm while the reinforcement details are shown in Fig. 7. In Fig. 8 the reinforcement details of the columns and the shear wall of the core of the structure can be seen. The

GEORGE MARKOU

28

40x40cm columns are reinforced with 8Ø20mm longitudinal rebars and two hoops Ø8mm@10cm as stirrups, while the 60x60cm columns are reinforced with 12Ø22mm longitudinal rebars and three hoops Ø10mm@10cm as stirrups. For the case of the shear wall, 72Ø18mm longitudinal rebars are used and Ø8mm@10cm as stirrups (Fig. 8).

Fig. 5. Plan View of the ground floor.

Fig. 6. Section a-a of the building.


29

Fig. 7. Reinforcement details of the slab (Ground – 3rd Floor).

Regarding the foundation type, it is assumed that the building is based on a general foundation slab which is 80cm thick. The foundation slab is reinforced with 14mm rebars every 10cm along the X and Y directions, while the foundation soil is also included in the finite element mesh in an attempt to make the numerical model as realistic as possible. A retaining wall (Fig. 6) is foreseen at the perimeter of the basement which has a thickness of 20cm. The retaining wall is reinforced with 12mm rebars at every 15cm as the main reinforcement and 8mm rebars at every 15cm as the secondary reinforcement. It must be noted here that the stair case is not included in the technical drawing of Fig. 7 and it will not be considered in the numerical model (Fig. 9).

3.2 Construction and Management of the Hexahedral Mesh

ReConAn FEA uses Femap [15] through which the initial mesh is constructed, while the input file is exported into a text file (.neu à neutral file) that is used to generate the FE numerical model during the analysis of the numerical problem. For controlling the mesh quality the Analysis by Parts approach is used, as it was presented in [14]. Fig. 9 shows the final mesh of the 75,080 hexahedral elements (8-noded) as it resulted from the mesh construction phase. The details related to the mesh are given in Table 1 and as it can be seen, the total number of concrete elements is 43,250, while the total number of nodes (excluding the embedded rebar macro-elements) is 119,232. The hexahedral edge size that was used to construct the concrete FE mesh of the RC frame was 20-40 cm.

GEORGE MARKOU

30

Fig. 8. Reinforcement details of the columns and shear wall.

Fig. 9. FE mesh of the 8-noded hexahedral elements.


31

Table 1. FE mesh data related to the 8-noded hexahedral mesh

a/a Description Value

1 Hexa8 total number of Soil elements 31,830

2 Hexa8 Concrete elements

2.1 Foundation Slab 4,810

2.2 Retaining Wall 3,630

2.3 Slabs 23,410

2.4 Columns 40x40 2,400

2.5 Columns 60x60 5,400

2.6 Shear Wall 3,600

Total 43,250

3 Total number of Hexa8 elements 75,080

4 Total number of Hexa8 nodes 119,232

The hexahedral mesh of the building was divided into 6 groups where each group was assigned with the corresponding Layers (a total of 25 layers). Fig. 10 shows the groups of Layers used during the hexahedral mesh construction procedure so as to manage graphically each part of the model. After the completion of the hexahedral mesh construction, the model was analyzed for the self-weight loads so as to ensure that the resulted numerical model was ready to be processed to the next stage which is the embedded rebar macro-element mesh construction. Fig. 11 shows the deformed shape and the von Mises stress distribution as it resulted from the Convergence Analysis stage (linear elastic analysis for the self-weight of the structure).

Fig. 10. Layer organization chart of the hexahedral mesh.

GEORGE MARKOU

32

Fig. 11. Deformed shape and von Mises stress distribution of the hexahedral mesh as it resulted from the Convergence Analysis stage.

3.3 Constructing, Managing and Verifying the Embedded Reinforcement FE Mesh

As it is shown in Fig. 12, the embedded rebar mesh was divided into 22 Layers according to the RC frame’s geometrical features. The 22 Layers were divided into 5 main groups so as to optimize the viewing procedure and the positioning of each reinforcement arrangement according to its location inside the structure. Furthermore, after the completion of the construction of the embedded rebar macro-element mesh for each structural member of the basement’s frame, a convergence analysis was performed (foundation slab, basement columns/shear wall/retaining wall/slab, see Figs. 13-14) so as to assess the derived FE mesh. After the successful completion of the convergence analysis, the embedded rebar macro-elements of the basement’s frame were replicated to the rest of the floors deriving the final embedded rebar macro-element mesh (see Table 2 & Fig. 18). Figs 14-19 show the embedded reinforcement macro-elements as they resulted from the mesh construction procedure.

Table 2. Embedded rebar macro-elements that derived from the mesh construction procedure.

a/a Structural Member c Macro-Elements

1 Foundation Slab 20 9,712

2 Retaining Wall 3 14,304

3 Columns 40x40 3 13,824

4 Columns 60x60 3 20,736

5 Shear Wall 3 18,180

6 Slabs 20 53,503

Total: 130,259

i Basement with Foundation Slab - 46,750

ii Ground Floor 20,990

ii 1st Floor 20,990

iii 2nd Floor 20,990

iv 3rd Floor 20,539


33

Fig. 12. Layer organization chart of the embedded rebar macro-element mesh.

Fig. 13. Macro-element rebar mesh construction phase (Basement and Foundation Slab).

GEORGE MARKOU

34

3.4 Solution of the Complete Model

At this stage the complete model (Fig. 19) is going to be analyzed in order to allocate and generate the final embedded rebar mesh through the use of the under study embedded rebar mesh generation method [14]. It must be noted at this point that the geometric constraint c for the embedded rebar mesh generation procedure inside the vertical and horizontal structural members was set equal to 3 and 20, respectively, as it can be seen in Table 2. For the columns and shear walls the geometric constraint parameter c was set to 3 and for the rest of the frame equal to 20. This choice is directly controlled by the discretization approach used when constructing the macro-element rebar elements. According to the number of penetrated hexahedral elements by each macro-element, the geometric constraint parameter is defined [8]. Table 3 shows the details of the resulted FE mesh, the total required time for generating the embedded rebar elements and the numerical details related to the solution of the FE model.

Fig. 14. Macro-element rebar mesh of the Basement and the Foundation Slab (46,750 macro-elements).

Fig. 15. Macro-element rebar mesh of the Ground floor (20,990 macro-elements).


35

Fig. 16. Macro-element rebar mesh of the 1st floor (20,990 macro-elements).

Fig. 17. Macro-element rebar mesh of the 2nd floor (20,990 macro-elements).

Fig. 18. Macro-element rebar mesh of the 3rd floor (20,539 macro-elements).

GEORGE MARKOU

36

Fig. 19. Macro-element rebar mesh of the RC frame of the building (130,259 macro-elements).

As it can be seen from Table 3 and Fig. 20, the total required time for the embedded mesh generation procedure was 75 minutes. The total number of generated embedded rebar FEs was 532,311 while the total number of short embedded rebar FEs that were discarded due to their small length was 5,408 (short length rebar criterion lmin = 5mm). The deformed shape of the embedded rebar mesh resulted by applying only the self-weight of the structure (Figs. 21 & 22). In order to solve the numerical model for a single load increment, ReConAn required 52 minutes, while the total computational time was 324 minutes. The required embedded rebar mesh generation time represents 23.15% of the total computational time. The resulted computational ratio is relatively small given that the at hand numerical problem was solved for a single load increment, thus for the case of a nonlinear solution procedure with several load increments and internal iterations, the computational time of the actual nonlinear solution procedure would have been significantly longer. The requirements in RAM according to Table 3, verified that the allocation of the stiffness matrix requires the largest amount of physical memory than any other matrix used during the solution procedure. As it resulted the total required RAM for allocating the stiffness matrix for the at hand model was 10.73 Gb and to solve this numerical implementation a total of 22.73 Gb RAM were required.

Fig. 21 shows the deformed shape of the embedded rebar mesh as it resulted from the analysis. The displacements are scaled so as to graphically represent the deformed shape of the frame (scale factors used in Fig. 20: x200-x2000). The deformed shape of the embedded rebar elements illustrate the robustness of the proposed embedded mesh generation method that manages to successfully allocate and generate more than half a million embedded rebar elements that are used to simulate the complete reinforcement grid of the under study RC structure. The deformed shape of the embedded rebar elements shows that their displacements, which are controlled by the hexahedral nodes’ displacements, follow the concrete element mesh deformed shape.

Fig. 22 shows the deformed shape and the von Mises strain distribution for the hexahedral elements as they resulted from the analysis procedure that was executed for the complete FE model. As it was expected, the superstructure undergoes a larger deformation in relation to the basement and the


37

foundation soil that was assumed to be sandstone did not develop significant deformations (Fig. 22). Strain concentrations were mainly observed at the column-slab joints of the flat-slab framing system of the RC structure.

Fig. 20. Information window of the ReConAn software.

GEORGE MARKOU

38

4 CONCLUSIONS

The integrated embedded mesh generation method presented by Markou [14], which is an extension of the Markou and Papadrakakis method [8], was used to generate the embedded rebar mesh of the ALHOSN University’s Male Campus. The parametric investigation performed for the required computational time in generating the embedded rebar elements, in a 64-bit operating system, verifies the computational robustness of the method in generating embedded rebar finite elements inside hexahedral meshes.

The FE mesh that was used consisted of 75,080 hexahedral elements and 130,259 embedded rebar macro-elements. Prior to the initiation of the mesh construction, for both hexahedral and embedded rebar macro-elements, a set of 25 and 22 Layers were defined, respectively, in order to manage and control the numerical model during the mesh construction phase. This topological sorting of the mesh provided the ability of controlling the frame’s mesh while it made it possible to visualize each part of the structural frame separately during the pre- and post-processing stages. After performing the analysis for this numerical implementation it derived that the embedded rebar mesh generation method managed to allocate 537,719 embedded rebar elements from which 5,804 had a length shorter than 5 mm and were excluded from the analysis procedure. The mesh generation procedure required 75 minutes.

Table 3. Numerical and computational results derived after the solution of the complete FE model.

a/a Description Value

1 Number of Hexahedral Elements 75,080

2 Number of Nodes (hexa8 only) 119,232

3 Number of Macro-Elements 130,259

4 Total Number of Embedded Rebar FEs Generated 532,311

5 Total Number of Short Embedded Rebar FEs that were Discarded by the Filter Algorithm

5,408

6 Required Embedded Mesh Generation Time 75 min.

7 Required RAM for the Stiffness Matrix 10.73 Gb

8 Number of Stiffness Matrix Elements 1,440,509,266

9 Total Required RAM for the Solution Procedure 22.73 Gb

10 Computational Time for Solving 1 Load Increment 52 min.

11 Computational Time for Writing the Output Data 172 min.

12

Total Computational Time:i. Read/Initialize Problemii. Generate Embedded Meshiii. Solve the System of Equations for 1 Load Increment /1 Internal Iterationiv. Write Output Data (out.txt file size: 645 Mb)v. Other

324 min.

5 ACKNOWLEDGEMENTS

The author would like to acknowledge the financial support from the ALHOSN University of Abu Dhabi and the Vice Chancellor Prof. Abdul Rahim Sabouni for his support throughout this research work.


39

Fig. 21. Deformed shape of the embedded rebar FE mesh.

GEORGE MARKOU

40

Fig. 22. Deformed shape and von Mises strain distribution of the concrete elements.

6 REFERENCES

[1] Jendele L. and Cervenka J. (2009), “On the solution of multi-point constraints – Application to FE analysis of reinforced concrete structures”, Computers and Structures, Vol. 87, pp. 970–980.

[2] Kotsovos M.D. and Pavlovic M.N. (1995), “Structural concrete. Finite Element Analysis for Limit State Design”, Thomas Telford: London.

[3] Spiliopoulos K.V. and Lykidis G. (2006), “An efficient three-dimensional solid finite element dynamic analysis of reinforced concrete structures”, Earthquake Engng Struct. Dyn.Vol. 35, pp. 137–157.

[4] Hartl H. (2000), “Development of a continuum mechanics based tool for 3d FEA of RC Structures and application to problems of soil-structure interaction” Ph.D. thesis, Faculty of Civil Engineering, Graz Univ. of Technology.

[5] Markou G. (2011), “Detailed Three-Dimensional Nonlinear Hybrid Simulation for the Analysis of Full-Scale Reinforced Concrete Structures”, Ph.D. Thesis, Institute of Structural Analysis and Seismic Research, National Technical University of Athens.

[6] Cervenka V. (2010), “Large Deflections”, Cervenka Consulting.

[7] Hristovski V.T. and Noguchi H. (2002), “Finite Element Modeling of RC Members Subjected to Shear”, Third DIANA World Conference, Tokyo, Japan 9-11 October 2002.

[8] Markou G. and Papadrakakis M. (2012), “An efficient generation method of embedded reinforcement in hexahedral elements for reinforced concrete simulations”, Advances in Engineering Software ADES, Vol. 45(1), Pp 175-187.


41

[9] Barzegar F. and Maddipudi S. (1994), “Generating reinforcement in FE modeling of concrete structures”, Journal of Structural Engineering Vol. 120, pp.1656 –1662.

[10] Elwi A.E. and Hrudey T.M. (1989), “Finite element model for curved embedded reinforcement”, Journal of Engineering Mechanics, Vol. 115, pp.740 –754.

[11] ASCE Task Committee on Concrete and Masonry Structures. (1982). "Finite element analysis of reinforced concrete." ASCE.

[12] Abdel-Halim M.A.H. and Abu-Lebdeh T.M. (1989), “Analytical study of concrete confinement in tied columns”, J. Struct. Engrg., ASCE, Vol. 115(11), pp. 2810-2828.

[13] Gonzalez Vidosa F., Kotsovos M.D. and Pavlovic M.N. (1990), "Three-dimensional finite element analysis of structural concrete”, Proc., Second Int. Conf. on Computer Aided Anal. and Des. of Concrete Struct., N. Bicanic, and H. Mang, eds., Vol. II, Pineridge Press, Swansea, Wales, pp. 1029-1040.

[14] Markou G. (2012), “Computational limitations of an embedded reinforcement mesh generation method for large-scale RC simulations”, IJCM, Submitted for Publication.

[15] Siemens P.L.M. Software (2009), “World-class finite element analysis (FEA) solution for the Windows desktop”, Siemens Product Lifecycle Management Software Inc.

43


AN ENHANCED MODEL FOR NUMERICAL INVESTIGATION OF MASS TRANSPORT IN AN

OPERATIONAL PEM FUEL CELL

Rihab Jaralla1, Jun Cao1, Mama Chacha2, Tawfiq J. Jaber2*

1Department of Mechanical and Industrial Engineering, Ryerson University 350 Victoria St., Toronto, ON, M5B 2K3

2Department of Mechanical and Industrial Engineering, ALHOSN University,P.O. Box 38772, Abu Dhabi, U.A.E

ABSTRACT: A mathematical model for the proton exchange membrane fuel cell (PEMFC) is developed in this study. The model features assessing the thermodiffusion effect on the PEM fuel cell performance, which has been conventionally neglected in existing fuel cell modeling studies. Also, instead of treating catalyst layers simply as interfaces, the model assigns a finite thickness for catalyst layers, allowing for a more realistic description of electrochemical reaction kinetics arising in the operational PEM fuel cell. The complete model consisting of the equations of continuity, momentum, energy, species concentrations, and electric potentials in different regions of a PEM fuel cell are numerically solved using the finite element method. Various flow and transport phenomena in an operational PEM fuel cell are simulated using the newly developed model. Through comparison of numerical simulation results using the model developed in this study and the conventional thermodiffusion-free model, no significant impact of thermodiffusion is demonstrated on the performance of a PEM fuel cell during its steady operation.

KEYWORDS: Proton exchange membrane (PEM) fuel cell, thermodiffusion, numerical simulation

NOMENCLATURE

catalyst surface area per unit

volume ( )2 3/m m

AMEA Full active MEA area (m2)

c concentration (mole fraction, or mass fraction) (-)

C electric charge (C) C Concentration (mol/m3)

pC specific heat capacity at constant

pressure (J/kg/K) D diffusion coefficient (m2/s) TD thermal diffusion coefficient

(m3/s/K) e membrane constant (mol/m3) E potentiel (V)

equivalent molecular weight of the membrane (kg/mol)

f swelling coefficient of membrane (-)

F Faraday’s constant, (96487 C/mol) i current density (A/m2) I current (A)

exchange current density (A/m2)

k thermal conductivity (W/m/K)

electro-kinetic permeability (m2)

hk hydraulic permeability (m2)

pk permeability of electrode/membrane (m2)

M molecular weight of mixture (kg/mol)

MEA membrane electrode assembly (-)

iM molecular weight of component i (kg/mol)

n molar number of electrons transferred (-)

a

mEW

0refj

kφ

* Corresponding Author: email: [email protected]

RIHAB JARALLA, JUN CAO, MAMA CHACHA, TAWFIQ J. JABER

44

1. INTRODUCTION

Fuel cells are devices that convert chemical energy of the reactants directly into DC electricity and heat with high efficiency. This research focused on the PEMFC which is the most common type of “regenerative fuel cell”. The high power density and rapid adjustment to power demands make proton exchange membrane fuel cells (PEMFCs) one of the best candidates for a clean alternative energy source in the twenty-first century, especially for transportation applications.

The polymer electrolyte membrane fuel cells (PEMFCs), sometimes also referred to as proton exchange membrane fuel cells, consists of a negatively charged electrode (anode), a positively charged electrode (cathode), and an electrolyte membrane. An electrochemical oxidation reaction at the anode produces electrons that flow through the bipolar plate/cell interconnect to the external circuit, while the ions pass through the electrolyte to the positive electrode (cathode). The electrons return from the external circuit, while the ions pass through the electrolyte to the positive electrode. The electrons return from the external circuit to participate in the electrochemical reduction reaction at the cathode. The reactions at the electrodes are:

Anode: ( ) ( )-

2 aqH 2H 2eg+→ +

(1)

electro-osmotic drag coefficient (-)

N molar flux (mol/m2/s) P pressure (Pa)

R universal gas constant (8.3145 J/mol/K)

RH Relative Humidity (-) S mass source/sink (kg/m3/s) T temperature (K)

cell potential (V) velocity vector (m/s) power (W)

x molar fraction (-) y y-coordinate (m) z z-coordinate (m) Z charge number (-) Greek Symbols

transfer coefficient (-) porosity (-)

volume fraction of water (-)

over-potential (V) membrane water content (-) viscosity (kg/m/s) density (kg/m3)

ω mass fraction (-) φ electric potential (V)

θ volume fraction of membrane in the catalyst layer (-)

χ cell efficiency (-) σ electrical conductivity (S/m) Subscripts a Anode act Activation c Cathode e Equilibrium state f Fixed charge g Gas phase i Species i ij Gas pair i , j in a mixture mem Membrane l Liquid phase ohm Ohmic m Membrane phase rev Reversible s Solid phase w Water 0 Reference conditions Superscripts eff Effective value of parameter ref Reference value sat Saturation state T Thermal 0 Standard state

dn

VVW

αεmwε

ηλμρ

ENHANCED MODEL FOR NUMERICAL INVESTIGATION OF MASS TRANSPORT IN AN OPERATIONAL PEM FUEL CELL

45

Cathode: 1 + -

2 2 (g) (aq) 2 ( )/ O +2 H +2e H O l→ (2)

Overall: ( ) ( ) ( )1

2 22 g 2/ H O H O g l+ → (3)

where “ ”refers to the gaseous state, “ ” stands for a substance in the aqueous phase that is dissolved in water [1], and “ ” denotes the liquid state.

Since the early nineties, many papers on single PEM fuel cell models have been published to investigate different aspects of the heat and mass transport processes in the fuel cell. Springer et al. [2] and Bernardi et al. [3] were the first to publish complete fuel cell models, which are isothermal, one-dimensional, and steady state, but one-dimensional models are unable to simulate the species and phase distribution along the channel within the gas diffusion layer (GDL). In late nineties, the models were more advanced; involving multi-dimensionality (2D or 3D), multiphase flow, and entire fuel cell structure.

There exist an enormous number of fuel cell analytical/numerical studies; however, no research has been done to investigate the effect of thermodiffusion on fuel cell performance other than the work by Jaralla et al. [4]. In recent years, thermodiffusion has become a subject of extensive scientific research both theoretically and experimentally. If the temperature of a liquid mass varies with spatial position, there will be a transport of energy from the hotter regions to the colder ones, [5]. A temperature gradient applied to a liquid mixture not only causes a heat flux but also gives rise to a diffusion current of constituent components. The resulting separation of the components causes a concentration gradient parallel or antiparallel with respect to the temperature gradient. This cross-effect between temperature and concentration is known as thermodiffusion or Ludwig-Soret-effect, since this effect was discovered by Ludwig [6] and systematically investigated by Soret [7] for liquid mixtures. Thermodiffusion plays a crucial role in many important processes. Research on thermal-solutal convection in porous media has gained more attention; such attention has been focused on areas including underground diffusion of nuclear waste, oil reservoir analysis, tar sand extraction. The significant role played by thermodiffusion in many applications motivated the present work.

A two-dimensional, steady-state, and general-purpose PEMFC model was developed by Jaralla et al [4], with emphasis placed on effects of thermodiffusion that have been neglected in previous fuel cell modeling studies. A noticeable impact of thermodiffusion is demonstrated on the constituent species of an operating PEM fuel cell. The oxygen consumption in the presence of thermodiffusion is found to be 3.4% less than in the case of nil thermodiffusion; that warns PEMFC researchers against the overestimation of oxidant supply using conventional computer models that ignore the thermodiffusion effect. Further investigation of thermodiffusion effects is done in this work, as an extension to Jaralla et al. [4].

The investigation of the thermodiffusion effect in the current model is based on the determination of DT (thermal diffusion coefficient). For multicomponent gas mixtures, the thermal diffusion coefficient DT is more commonly used as a measure of thermodiffusion, and at moderate pressures, the Maxwell-Stefan equations in the form developed by Curtiss and Bird [8] will be used to model the transport of reactants in the electrodes. By considering DT in the Maxwell-Stefan equations, the thermodiffusion effects on the molar fraction of each species can be investigated [4]. In this work, DT is determined for each species: hydrogen, water vapor and carbon dioxide at the anode side and, oxygen, water vapor and nitrogen at cathode side. Important electrochemical and physical phenomena during operation of PEM fuel cell were successfully simulated in this study.


46

2. MATHEMATICAL MODEL OF A PEM FUEL CELL

2.1 Computational domain

As illustrated in Fig. 1, a two-dimensional cell model established in the y-z plane is studied. There are five primary transport phenomena during fuel cell operation, namely, the heat transfer in the solids and in the gases, the flow of reactant gases, the convection and diffusion of different species, as well as the transport processes for the proton and the liquid water.

For computation convenience, these five flow and transport phenomena are modeled using seven computational sub-domains (from the top to the bottom as schematically shown in Figure 1: on the anode side, the collector plate, gas diffusion layer (GDL) and catalyst layer (CL); the ionomeric membrane; the catalyst layer, gas diffusion layer, and collector plate on the cathode side. In the y-z plane, components involved in the two-dimensional model include the two current plates, two GDLs, and two CLs at the anode and cathode sides, along with a membrane in the middle. However, the anodic and cathodic gas channels may be extracted from the computational domain since the cross flow within the channels is insignificant.

Fig. 1: Schematic diagram of computation model of a 2-D PEM fuel cell for (y-z) plane.

2.2 Assumptions

The following assumptions will be invoked to make the model more tractable and the computation faster:1. The transport processes are steady-state. 2. The thermodiffusion is taken into account. 3. The flow in the gas-distribution channels is laminar and all of the gaseous mixtures are assumed to

be ideal gases and incompressible. Though heat generation due to the electrochemical reaction is considered, the fluid properties are assumed to be independent of temperature.

4. The gas diffusers, the catalyst layers, and the membrane are all considered as isotropic and homogeneous porous media.

5. No water phase change is taken into account within each component except that all water vapor at each interface between the catalyst layer and the membrane is entirely transformed to liquid water.


47

2.3 Governing equations

2.3.1 Mass equation

The continuity equation describing the conservation of mass is used for the entire fuel cell

(4)

with ρ denoting the density, V is the velocity of the fluid mixture and ε is the porosity

( )s at CL , at GDL at membranect g mε ε ε ε ε ε= = = .

2.3.2 Momentum equation

To describe the momentum conservation in porous media, the gas mixture flow in the porous electrodes is governed by Darcy’s law in its revised version:

(5)

(6)

where p is the pressure, kp and kh are the permeability of the porous electrode and the hydraulic permeability, respectively; µ is the viscosity of the fluid; and r (2) is a coefficient describing the effect of the porosity of the medium to the viscous force. The coefficient r (2) can be determined using [9]:

( ) ( )( )

2

2

2

12.25 in GDLs

g

g

rε

ε

−=

(7)

( ) ( ) ( )2

2

2

12.25 in CLs and membranect

ct

rε

ε

−=

(8)

where gε is the porosity of the gas diffusers, ctε is the effective porosity of the catalyst layer which can

be calculated by:

ct m mcε ε θ= ⋅ (9)

with mε and mcθ denoting the porosity of the membrane and the volume fraction of the membrane in the catalyst layer, respectively. It is obvious that Eqs. (5) and (6) reduce to Darcy’s law when r(2) =1.

Since gas diffusion layers (GDLs), catalyst layers (CLs) and the membrane domains are porous structure, Darcy’s law can be apply within these entire domains. The pressure boundary values are prescribed at the operating pressures for the interfaces between the gas channels and GDLs for both anode and cathode sides. That is:

For Boundaries #1, #9: p = pa (10)

For Boundaries #1', #9', p = pc

(11) where p

a and p

c are the operating pressures at the anode and cathode sides, respectively.


48

2.3.3 Mass transfer equation

Based on the Maxwell-Stefan equations in the form developed by [9], the transport model for the multi-gases passing through the GDLs and CLs accounts for both diffusion and convection:

(12)

where jx and represent the mole and mass fraction of the species j in the mixture. In Eq. (12), the

effective binary diffusivities effijD for the flow in porous media are obtained using the original binary

diffusivities via Bruggemann correction formula [10]:

1.5effij ijD D ε= (13)

and

1.5

0 00 0

0

( , )ij ij

p TD D T p

p T

⎛ ⎞= ⎜ ⎟

⎝ ⎠ (14)

where 0ijD , 0T , 0p are the reference binary diffusivities, temperature and pressure, respectively.

In addition, Eq. (12) takes into account the thermodiffusion effect due to the presence of temperature gradient. The multicomponent thermodiffusion coefficient, DT , is calculated by [11]:

0.5110.511

7 0.659 1

0.511 0.489

1 1

2.59 10

n

i iT i i ii f in n

i i i ii i

M wM w

D T xM w M w

− =

= =

⎛ ⎞ ⎛ ⎞⋅⎜ ⎟ ⎜ ⎟⋅

⎜ ⎟ ⎜ ⎟= − ⋅ ⋅ − ⋅⎜ ⎟ ⎜ ⎟⋅ ⋅⎜ ⎟ ⎜ ⎟⎝ ⎠ ⎝ ⎠

∑

∑ ∑

(15)

where Mi indicates the mole mass of the species i in the mixture, and T

f is the fluid temperature, x

i the

molar fraction of the species i, and wi is the term of mass fraction of species i.

As indicated in Eq. (12), the right-hand side corresponding to the GDLs vanishes because there is no reaction on the sites; but a sink term at the anodic CL must be prescribed:

2 22a

H H

jS M

F= −

(16)

20H OS =

(17)

due to the oxidization of hydrogen; meanwhile a sink term for oxygen and a source term for water should be considered at the cathodic CL representing the oxygen reduction reaction:

2 24c

O O

jS M

F=

(18)

2 22c

H O H O

jS M

F= −

(19)


49

In the above source/sink terms, M is the molecular weight of the species, ja and j

c are the anodic

and cathodic exchange current densities, respectively, which can be modeled by the Butler-Volmer equations [12]:

( ) 2

2

1/ 2

0,

exp expa e

Href a ca a aa

H ref

c F Fj aj

c RT RT

α αη η

⎛ ⎞ ⎡ ⎤⎛ ⎞ ⎛ ⎞= − −⎜ ⎟ ⎢ ⎥⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎝ ⎠ ⎝ ⎠⎣ ⎦⎝ ⎠

(20)

( ) 2

2

0,

exp expc c

Oref a cc c cc

O ref

c F Fj aj

c RT RT

α αη η

⎛ ⎞ ⎡ ⎤⎛ ⎞ ⎛ ⎞= − −⎜ ⎟ ⎢ ⎥⎜ ⎟ ⎜ ⎟⎜ ⎟ ⎝ ⎠ ⎝ ⎠⎣ ⎦⎝ ⎠

(21)

The surface over-potential for an electrochemical reaction, η is considered the driving force for an electrochemical reaction, and can be described as:

0

at

at a s p

c s p

anode

E cathode

η φ φ

η φ φ

= −⎧⎪⎨

= − −⎪⎩ (22)

where sφ is the solid-phase potential, pφ is membrane-phase potential, and 0E is the thermodynamic open circuit potential for overall reaction, which is expressed by the Nernst equation [13] as a function of the reactant and product concentrations at the interface:

( ) ( )2 2

3 20 1.23 0.9 10 298 2.3 log

4 H O

RTE T p p

F−= − × − +

(23)

The boundary conditions at the anodic and cathodic interfaces between the gas channels and GDLs, i.e., boundaries #1,#1', #9, and #9' (as shown in Fig. 1), for each of the mass fractions of H

2, O

2 is prescribed

using a constant inlet value. The mass fractions of water vapor at the anode and cathode sides can be determined as follow respectively:

2

2 ,

satH o

H O aa

p MRH

p Mω =

(24)

2

2 ,

satH o

H O cc

p MRH

p Mω =

(25)

where RH is the relative humidity, pa and pc are the operating pressures at the anode and cathode sides; psat is the saturated water partial pressure calculated using the following empirical equation [2]:

3725

10 104454.1101837.902953.01794.2log TTTpsat −− ×+×−+−= (26)

2.3.4 Potential equations

To describe the fuel cell potential distribution, two kinds of potentials are modeled. One is the membrane-


50

phase potential, which is obtained by solving the equation of proton transport in the catalyst layer and membrane:

(27)

where pφ is the ionic potential, and pσ is the ionic conductivity.

The proton current density at the interface between the catalyst layers and the GDL is set to zero. Dirichlet boundary conditions are used to solve the protonic potential equations at the interface between the membrane and catalyst layers at anode and cathode sides. Elsewhere homogeneous Neumann boundary conditions are applied.

The second kind of potential in the model is the solid-phase potential which results from the solution to the equation of electron transport in the catalyst layers, gas diffusers and collector plates:

(28)

where eφ is the electric potential, and eσ is electrical conductivity.

The solid-phase potential sφ is the cell voltage; the value of solid-phase potential along the cathode collector plate edge, i.e., Boundary #5', is prescribed while the value of the cell voltage along the anode current plate edge, i.e., Boundary #5, is assumed to be zero.

2.3.5 Water transport equation

The water transport in the membrane is driven by the electro osmotic drag, diffusion, and hydraulic permeation, which is induced respectively by the moving protons, the water concentration difference, and the pressure difference between the two sides of the membrane. A good water management procedure aims at balancing the three water fluxes such that neither flooding of the electrodes nor drying out of the membrane occurs. The following equation describes the distribution of water concentration, wc , within the membrane:

(29)

It is obtained through combination of the diffusion represented by the diffusion coefficient, Dw, the pressure drop, , and the electro osmotic drag that is related to the current density, . More details about Eq. (29) and its derivation can be found in Cao et al. [14]. In this model equilibrium is assumed between the gas phase and the membrane phase of water in Nafion membrane. Since the water can be transported through the catalyst layers to the membrane, Dirichlet boundary conditions should be applied at the interface between the membrane and the catalyst layers at the anode and cathode sides. And the water content at these interfaces can be calculated using [2]:

2 30.043 17.8 39.85 36 0 1a a a for aλ = + − + < ≤ in catalyst layers (30)


51

where a is the activity of water vapor defined as:

2H O

sat

x pa

p=

(31)

As the water mole fraction exceeds saturation, a linear relation is assumed between the water content and water activity [2]:

14 1.4( 1) 1 3a for aλ = + − < ≤ in the membrane (32)

the Neumann boundary condition is applied at the left and right sides of the membrane:

(33)

where denotes the unit vector normal to the boundaries.

2.3.6 Energy equation

In an operational PEM fuel cell, the overall mechanism of water transport is further complicated by the globally exothermic electrochemical reaction. The heat generated increases water evaporation rates at the cathode and, in situations where the membrane is allowed to dehydrate, can combine with ohmic heating in the membrane leading to a deterioration of the membrane-electrode bond. The temperature distribution can be obtained by solving the following energy equation in the GDLs, CLs and membrane of a PEM fuel cell:

(34)

At GDLs, the last term, 2eeGDL

i

σ , represents the heat source term produced as a result of the ohmic heating

of electron current, ie , as there is an electronic resistances through the gas diffusers. In the membrane

an additional Joule heating source, 2p

pm

i

σ, arising from protonic resistances through the membrane has

been added to the energy equation, pmσ appears in the Joule heating source term representing the ionic

conductivity in the membrane. At the catalyst layers, on the right hand side of the above equation, the second and third terms describe the ohmic heating of both proton current ip and electron current ie within catalyst layers; and the last term represents the heat generation or absorption because of electrochemical reaction at the catalyst.


52

3. NUMERICAL PROCEDURES

The PEM fuel cell modeling equations describing fluid flow, multi-species transport, heat transfer, and electric potentials are strongly nonlinear and coupled with each other. To numerically solve this large set of nonlinear equations, a finite element computational fluid dynamics package, COMSOL 3.4 is selected. In finite element analyses, it is often necessary to resolve the geometry in more details for more accuracy, thus, numerical tests were performed for the base case geometry to ensure that the solutions were independent of further refinement for the mesh. In the current simulation, the total number of elements is 13,264, and a stationary non-linear solver is used together with Direct (UMFPACK) linear system solver in this simulation.

The relative tolerance for the error criteria was 1.0 x 10-6.

4. RESULTS AND DISCUSSION

4.1 Model Validation

To validate the fuel cell model presented in this paper, a comparison using fuel cell performance curves is made between the simulation results and the available experimental data [15] corresponding to the same operating conditions grouped in Table 1. Table 2 lists the base case operational parameters and electrode properties for the current model.

The polarization curve is the most important characteristic of a fuel cell. It may be used for diagnostic purposes. As shown in Fig. 2, the sample points used to plot polarization curves are picked in such a way that the cell voltage can range from 0.3V to 0.9V with 0.1V as increment. The polarization curve looks steeper and exhibits slight nonlinearity in the activation region.

Table 1: Physical dimension of the PEM fuel cell, and operating parameters under a base case computation, [15]

Description Value Unit

Gas diffuser width (z-direction) 3 x 10-3 m

Gas diffuser height (y-direction) 2.54x10-4 m

Collector width 1.3x10-3 m

Catalyst layer thickness 2.87x10-5 m

Membrane thickness (y-direction) 2.3x10-4 m

T : fuel cell (ambient) temperature 333 K

aζ : stoichiometric ratio at anode 1.3 -

cζ : stoichiometric ratio at cathode 3 -

pa : fuel inlet pressure at anode 1 atm

pc : air inlet pressure at cathode 3 atm

RH : relative humidity of inlet gas mixture 100 %


53

Then, the two sets of results reaches good agreement at intermediate current densities

, showing the cell voltage drops clearly in a linear trend as the current density increases within this ohmic loss region. However, a remarkable discrepancy is found for

high current densities ( ). This is due to the lack of a successful mathematical model to accurately quantify the effect of mass transport losses. As of today, all existing fuel cell models underestimate the mass-transport limitation. As shown in Fig. 2, the limiting current density captured from the simulation results is close to 0.8 A/cm2, which appears generally in good agreement with the findings through experiments.

Table 2: Electrode and electrochemical properties of PEM fuel cell

Description Value Unit Ref.

�g : gas diffuser porosity 0.4 - [16]

�m : membrane porosity 0.28 - [16]

�mc : volume fraction membrane in catalyst layer 0.5 - [16]

kp : permeability to air in the gas diffuser 1.76x10-11 m2 [16]

kh : hydraulic permeability of the membrane 1.58x10-18 m2 [16]

k� : electrokinetic permeability of membrane 1.13x10-19 m2 [16]

kair : air thermal conductivity 30.x10-2 W / m / K [17]

kgr : thermal conduc. of matrix of gas diffuser 150.6 W / m / K [16]

km,dry : thermal conductivity of dry membrane 100 W / m / K [16]

cp,air : air specific heat at constant pressure 1008 J / kg / K [15]

cf : fixed charged site concentration in memb. 1.2x103 mol / m3 [15]

zf : charge of sulfonate site in memb. -1 - [16]

drymρ : membrane solid dry mass density 1980 kg / m3 [16]

EWm : equivalent membrane weight 1.1 kg / mol

f : membrane swelling coefficient 0.0126 - [13]

0,refaaj : reference exchange current density times specific area at the anode 1.0x109 A / m3 [15]

0,refaaj : reference exchange current density times specific area at the anode 2.5x103 A / m3 [15]

2 ,H refc : reference molar concentration 40.88 mol / m3 [15]

2 ,O refc : reference molar concentration 40.88 mol / m3 [15]

caα : anodic transfer coefficient at cathode 0 - [16]

ccα : cathodic transfer coefficient at cathode 1.2 - [16]

acα : cathodic transfer coefficient at anode 1/2 - [16]aaα : anodic transfer coefficient at anode 1/2 - [16]


54

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.2 0.4 0.6 0.8 1 1.2

Cel

l Vo

ltag

e (V

)

Current density (A/cm2 )

Cal

Exp.

Fig. 2: Comparison of modeling results with experimental data by Ju and Wang [15].

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

Cel

l vo

ltag

e (V

)

Current density (A/cm2)

with DT

without DT

Fig. 3: Comparison of modeling results with and without thermodiffusion on the PEM fuel cell performance


55

4.2 Effect of thermodiffusion on polarization curve

A Set of systematic comparisons are made between the results with the thermodiffusion effects turned off against the thermodiffusion term switched on. By using the two types of simulation results, Figure 3 shows the comparison of modeling results with and without thermodiffusion and their effects on the PEM fuel cell performance. All comparisons show that the overall performance of a fuel cell during its steady-state operation does not significantly change no matter whether the thermodiffusion term is switched on or off; because of the relatively small magnitude of the thermodiffusion coefficients of the species at anode and cathode sides, as shown in Figs. 4 and 5.

4.3 Results of the base case

Transport phenomena play an important role in many engineering applications and scientific research. In this numerical simulation, detailed information of transport phenomena in the fuel cell that are usually hard to examine by in situ measurements can be provide by the comprehensive current model. In the following sections, distributions of different physical variables, including the velocity, distributions of mole fractions of reactant gases, cell temperature, local current density, membrane water content, potential loss, and the activation over-potential distribution at the cathode side will be examined for better understanding the fuel cell working mechanism; taking the thermodiffusion into account. The numerical simulation employs fuel cell operational conditions as described in Table 1.

4.3.1 Distributions of reactants and products

Figures 6-9 illustrate the important effects of the transport of reactant and product gases on the performance of PEM fuel cell. Figure 6 depicts the hydrogen mole fraction distribution inside the anode-side porous electrode for three different nominal current densities: on the top a higher one (0.9179 A/cm2), on the middle (0.6959 A/cm2), and the lower one (0.2862A/cm2) on the bottom.

In all cases, the hydrogen mole fraction decreases gradually from the inlet toward the anode catalyst layer due to the consumption of the hydrogen through the hydrogen oxidation reaction (HOR) that takes place at the anode side catalyst. Also the hydrogen mole fractions decrease in higher rate in GDLs from the side toward the center where the current collector is positioned due to lack of convection and higher diffusivity on the sites; as, the hydrogen has high diffusivities in both carbon-oxide (CO2) and in water vapor. Thus, no significant reduction of hydrogen mole fraction is found, as shown on Fig. 6, even at the high current density of i = 0.-179 A/cm2.


56

Thermal diffusion coefficient of hydrogen

Fig. 4: The Thermal diffusion coefficients of species (H2, H2O, CO2) at anode side


57

Fig. 5: The Thermal diffusion coefficients of species (O2, H2O, N2) at cathode side


58

Fig. 6: Hydrogen mole fraction at anode-side catalyst layer and gas diffusion layer for three loading conditions: 0.9179 A/cm2 (top), 0.6959 A/cm2 (middle), and 0.2862A/cm2 (bottom)

Figure 7 shows the oxygen mole fraction distribution inside the cathode-side porous electrode for three different current densities. In general, the oxygen concentrations decrease gradually from inlet at cathode side toward the cathode catalyst layer due to the consumption of oxygen through the oxidation-reduction reaction (ORR) in the cathode catalyst layer. In addition the oxygen depletion is more noticeable than the hydrogen because of the relatively low diffusivity of the oxygen compared to that of hydrogen and the low concentration of oxygen in ambient air that leads to slower oxygen diffusion in the cathode, causing dramatic oxygen mass fraction gradients. From Figs. 6-8 one can observe that as the loading


59

current density increases, more hydrogen and oxygen are depleted and higher water vapor concentration is produced, as a result, the mass transport limitation for the oxygen within GDL and catalyst layer becomes significant.

Fig. 7: Oxygen mole fraction at anode-side catalyst layer and gas diffusion layer for three loading conditions: 0.9179 A/cm2 (top), 0.6959 A/cm2 (middle), and 0.2862A/cm2 (bottom)

4.3.2 Temperature distribution

Maintaining the desired temperature in PEM fuel cell is required to remove the heat generated by the electrochemical reaction in order to prevent drying of the membrane. Figure 9 illustrates the temperature distribution in the y-z plane for PEM fuel cells for high, intermediate and low load conditions. The presence of current plates as a medium of cooling in y-z plane causes significant temperature gradient along the z-direction. The peaks of temperature are observed within the cathode gas diffusion layer at the corner of current plate for the high and intermediate current density. While, for the low current density the peak of the temperature can be observed at the corner of the cathode gas channels; where, the concentration of the oxygen is higher at the inlet of the cathode. It can be obviously seen that the temperature distribution is symmetric about its vertical central line due to the symmetry of flow field. Also, the temperature at the cathode side is slightly higher than at the anode side; as, air at the cathode


60

side is closer to the major heat source. It is also noticed from Fig. 9 that the nominal current density has an influence on the increasing of temperature during fuel cell operation, as when the current density increases, the temperature maximum gets increased as a result of more heat generated from the chemical reaction.

Fig. 9: Temperature distribution across fuel cell in z-y plane for the three different current densities: 0.9179 (top), 0.6959 (middle) and 0.2862 (bottom)


61

4.3.3 Water content in the membrane

The water transport in the membrane is mainly caused by the electro-osmotic drag related to the protonic current in the electrolyte that leads to the transport of water from anode to cathode; and transport through diffusion related to the water-content gradients in the membrane.

These transport mechanisms usually effect the distribution of the water in the fuel cell. Figure 10 shows the profiles for water content in the membrane for the base case conditions at three different average current densities. The influence of the electro-osmotic drag and back diffusion are obvious from these results on Fig. 10. Across the membrane, the water content at the anode side is lower than that at the cathode side, as water is dragged toward the cathode via electro-osmotic transport and due to insufficient back-diffusion to the anode side from the cathode side, where water is produced.

Fig. 10: Water content distribution in membrane z-y plane for the three different current densities: 0.9179 (top), 0.6959 (middle) and 0.2862 (bottom)


62

Figure 10 demonstrates that the water content at the anode side decreases from 14.01 to 14.002 as the current density increases from 0.2862 to 0.9179 , while at the cathode side rises from 14.273 to 14.427. As when current density increases, a higher electro-osmotic drag drives more water from the anode side to the cathode side that leading to lower water content at the anode side while a higher water content at the cathode side; also, more water is produced at the cathodic catalyst layer in response to a higher current density.

Fig. 11: Membrane-phase potential distribution in MEA z-y plane for the three different current densities: 0.9179 (top), 0.6959 (middle) and 0.2862 (bottom)


63

4.3.5 Membrane-phase potential loss

The Protons generated within the anodic catalyst layer and pass through the membrane to consume inside the cathodic catalyst layer cause what is called membrane-phase potential loss as a result to the resistance to proton transport across the membrane electrode assembly (MEA), which is a combination of a membrane sandwiched between two catalyzed porous electrodes. Figure 11 shows the membrane-phase potential loss distribution for three different loading conditions. For all the three current densities, the electric potential on the upper boundary of the membrane is assumed to be a constant value-zero at the anode side for convenience of computational simulation.

The membrane-phase potential loss at the lower current density is small owing to the small amount of protons flowing through the membrane. The membrane-phase potential loss becomes significant as the current density increases. The maximum potential loss can be observed at the inlet of the anode side, due to the highest rate of electrochemical reaction that results from the fastest consumed of the protons.

5. CONCLUSION

A two-dimensional, steady-state, and general-purpose PEMFC model was developed in this work with emphasis placed on effects of thermodiffusion that have been neglected in previous fuel cell modeling studies. The simulation based on this new model can predict the overall performance of PEMFC, which reach good agreement with available experimentally-obtained data. Simulation results also provided valuable information about the detailed distribution of the reactant gases inside the PEM fuel cell. Due to the relatively small magnitude of the thermodiffusion coefficient in the fuel cell application, the overall performance of a steady-state PEMFC exhibits no significant change after the thermodiffusion term is incorporated into the model.

The comprehensive fuel cell model developed in this study can be used to examine details of mass and heat transport encountered in an operational PEMFC, which are usually unobservable by in-situ measurements due to the tiny dimension of the cell. The simulation results also provide insights that may assist practitioners in optimizing PEMFC design and reducing the manufacturing cost.

ACKNOWLEDGMENTS

This research work was supported by the Discovery Grant of the Natural Sciences and Engineering Research Council of Canada (NSERC).

REFERENCES

[1] Fine L. W., Beall H., Stuehr J., (2000). Chemistry for Scientists and Engineers. Preliminary Edition, Saunders Golden Sunburst Series.

[2] Springer T. E., Zawordzinski T. A., Gottesfeld S., (1991). “Polymer electrolyte fuel cell model.” J. Electrochemical Society, Vol. 138 (8), pp 2334-2342.

[3] Bernardi D.M., Verbrugge M.W., (1991). “Mathematical model of a gas diffusion electrode bonded to a polymer electrolyte.” AIChE Journal, Vol. 37 (8), pp 1151-1162.


64

[4] Jaralla R., Cao J., (2010). "Numerical investigation of the thermodiffusion effects on PEM fuel cell performance." Modern Physics Letters B, Vol. 24 (13), pp 1329-1332.

[5] Costeseque P., Pollak T., Platten J.K., Marcoux M., (2004). “Simultaneous evaluation of Soret and Fick coefficients in a free and a packed vertical gradient Soret cell.” 6th International Meeting on Thermodiffusion, IMT6, Varenna, Italy.

[6] Ludwig C., Sitzber A., (1856). Wiss, Vien Math-Naturw., Vol. 20, pp 539.

[7] Soret C., (1880). “Influence de la température sur la distribution des sels dans leurs solutions.” Acad. Sci., Paris, C. R.,Vol. 91, pp 289-291.

[8] Curtiss C.F., Bird R.B., (2001). “Multicomponent diffusion.” Ind. Eng. Chem. Res.Vol. 40, pp 1791

[9] Gagan G., (1989). Flow and Transport in Porous Formation. Springer-Verlag, Berling.

[10] Bernadi D. M., Verbrugge M.W., (1992). “A mathematical model of the solid polymer electrolyte fuel cell.” Journal of Electrochemical. Society, Vol. 139 (9), pp 2477-2491.

[11] Kuo K.K.Y., (1986). Principles of Combustion: John Wiley, and Sons, New York.

[12] Furry W. H., Jones R.C., Onsager L., (1939). “On the theory of isotope separation by thermal diffusion.” Physical Review, Vol. 55, pp1083-1095.

[13] Djilali N., Lu D., (2002). “Influence of heat transfer on gas and water transport in fuel cells.” International Journal of Thermal Sciences 41 (1): 29-40.

[14] Cao J., Djilali N., (2005). “Numerical modeling of PEM fuel cells under partially hydrated membrane conditions.”, ASME J. Energy Resources Technology, Vol. 127 (1), pp 26-36.

[15] Ju H., Wang C.Y., (2004). “Experimental validation of a PEM fuel cell model by current distribution data.” Journal. Electrochemical Society, Vol. 151 (11), pp 1954-1960.

[16] Gurau V., Liu H., Kakac S., (1998). “Two-dimensional model for proton exchange membrane fuel cells.” AIChE J. Vol. 44 (11), pp 2410-2422.

[17] Zhou T., Liu H., (2001). “A general three-dimensional model for proton exchange membrane fuel cells.” Int. J. Transport Phenom.Vol. 3 pp 177-198.

65


UAE TRAFFIC ACCIDENTS TRENDS AND INSIGHTS OVER THE PAST DECADE

Sharaf A. Alkheder1*, Reem Sabouni1, Hany El Naggar2

1 Department of Civil Engineering, Faculty of Engineering & Applied Sciences, ALHOSN University, Abu Dhabi, UAE, P.O. BOX: 38772

2 Department of Civil Engineering, University of New Brunswick, H-124, Head Hall, 17 Dineen Drive, P.O. Box 4400, Fredericton, N.B., Canada E3B 5A3

ABSTRACT Traffic accidents are a critical safety issue in the Gulf Cooperation Council (GCC), especially in the United Arab Emirates (UAE). The economic prosperity over the past few decades associated with excessive growth in population, major highway infrastructure development projects, and high levels of car ownership have led to dramatic increase in traffic accidents. To address this issue, this article conducts a comprehensive study to investigate traffic accidents rates and causes in UAE to propose recommendations for reducing the traffic accidents hazard. Traffic accident data for the UAE for the period from 2000 to 2009 were collected from the Ministry of Interior (MOI). The collected data included detailed information mainly from the traffic accident reports stored in the database of the General Administration of Traffic Coordination department. Using the obtained data, a comprehensive critical analysis was conducted. A pattern of conditions and causes of traffic accidents were established with different accidents trends being extracted. The analysis revealed aggregated trends and correlations between accidents, injuries, fatalities, vehicle population and type, driver’s age, gender, and nationality.

KEYWORDS: Traffic Accidents, UAE, Critical Analysis, Pattern of Conditions & Causes.

1. INTRODUCTION

Road traffic accidents is a major public problem facing most developed and developing countries in the world resulting yearly in huge social, economical, and environmental losses. It has a higher severity level in developing countries due to the poor transportation infrastructure system and low enforcement levels of traffic laws. Much of the existing literature has emphasized on the strong correlation between aggressive & faulty driving and traffic accidents occurrence (see Bener et al. [2]). Based on a study carried out by Treat et al. [13], human factor contribute solely to around 57% of total accidents and to more than 90% as a contributing factor while only 2.4% and 4.7% were the sole contribution of vehicle and environment factors, respectively. Most accidents are caused by human error such as over-speeding, not giving the right of way, sudden change of lanes etc.

The Manchester Driver Behavior Questionnaire (DBQ), originally developed by Reason et al. [10], is one of the tools that has been used extensively to study the relationship between human driving errors/violations and involvement in accidents. DBQ has been used by many researchers in different countries as in the work of Dobson et al. [4], Kontogiannis et al. [7], and Lajunen et al. [8]. Bener [3] used DBQ in a comparison study between Qatar and UAE.

Statistical approaches have been widely used by researchers in the literature to extract important


SHARAF A. ALKHEDER, REEM SABOUNI, HANY EL NAGGAR

66

patterns and relationships embedded in traffic accident data. Lord and Mannering [9] provide an extensive review of the statistical methods being in used for decades to analyze traffic accidents data. Example of these methods include developing statistical models to highlight highway locations with traffic safety problems (Hauer [6]; Stokes and Mutabazi [11]; Tarko et al. [12]). Abdel-Aty and Radwan [1] used negative binomial regression to model traffic accidents occurrence. Virtisen [14] used different approaches to analyze hot spots locations with high accidents frequency. Other models tried to relate traffic flow characteristics, such as in the work of Zhou and Sisiopiku [16] where they use v/c ratio, with accident rates.

This article conducts a comprehensive study to investigate traffic accidents rates and causes in UAE to propose recommendations for reducing the traffic accidents hazard. Traffic accident data for the UAE for the period from 2000 to 2009 were collected from the Ministry of Interior (MOI). The collected data included detailed information mainly from the traffic accident reports stored in the database of the General Administration of Traffic Coordination department. Using the obtained data, a comprehensive critical analysis was conducted to establish a pattern of conditions and causes of traffic accidents with different accidents trends being extracted. The analysis revealed aggregated trends and correlations between accidents, injuries, fatalities, vehicle population and type, driver’s age, gender, and nationality.

2. TRAFFIC ACCIDENTS IN THE UNITED ARAB EMIRATES

Traffic accidents in the UAE is a serious problem; the country suffers a lot due to accidents in loosing mostly young people at the same time paying cost of treatment, insurance and lifelong liability of injured persons. The objective of this section is to investigate the causative factors of traffic accidents in the UAE. A number of issues such as causes of accidents and characteristics of road users were considered in the study.

The traffic accidents data for the last ten years (2000-2009) used in this report were collected from the Ministry of Interior, MOI (The General Administration of Traffic Coordination). The collected data included the traffic accident report data for the UAE as whole, as well as, the same set of data for each of the seven emirates.

The data were analyzed to investigate the predominant causes of accidents and the change of the rate of accidents in the last decade. Figure 1 presents the total number of traffic accidents in each emirate during the period from 2000 to 2009. In addition, the variation in the total number of traffic accidents in the UAE during the same period is shown in Figure 2. As it can be noticed from Figure 2, the total number of traffic accidents in the UAE dropped from about 10600 to 7900 over the last ten years (a reduction of about 25.5%).

To better illustrate the reduction trend in the traffic accidents, Figure 3 presents numbers of traffic accident per 100 thousand of population. Consequently, according to the per 100 thousand representation the reduction in the number of accidents is about 70% (2009 in comparison with 2000). Figure 4 shows the share of each emirate in the average number of accidents during the period from 2000 to 2009.


67

Fig. 1: Total number of traffic accidents in each emirate from 2000 to 2009

10579

87318018

86528269 8254 8443

72897874 7904

0

2000

4000

6000

8000

10000

12000

2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Year

Nu

mb

er o

f A

ccid

ents

Fig. 2: The variation in the total number of traffic accidents in the UAE

Fig. 3: Number of traffic accidents in the UAE per 100 thousand of population.


68

Abu Dhabi 38%

Dubai 19%

Sharjah 20%

Ajman 4%

Umm al-Quwain

4%

Ras al-Khaimah

11%

Fujairah4%

Fig. 4: Share of each emirate in the total traffic accidents

According to the traffic accident report format used by the UAE Traffic Department, the causes of accidents are divided into nine different types: disrespect of other road users, changing lanes frequently, not yielding, not leaving appropriate distance (t), crossing red traffic light, speeding, negligent driving, tire bursting, and other (which includes all other causes). Figure 5 shows a pie chart illustrating the share of the different factors in the total number of traffic accidents over the whole study period (2000-2009). Among many causes of accidents the main cause of accidents is the disrespect of other road users, according to Figure 5 it account for almost 22%. The “other” category of the accident causes was found to account for nearly 29% of the causes as revealed from the data collected from the traffic accident reports. This very high percentage indicates that the accident report format needs some alternations to better categorize the traffic accident causes and make the judgment easier to traffic officers in the accident site. According to these data more than 90% of the accidents maybe associated with the driver faults. In addition, the most predominant cause of fatalities that accompany the accidents is speeding cause as concluded from the related data analysis (accounts for almost 47%). Furthermore, accidents data investigation shows that in terms of fines the speeding cause surpassed all other causes by 87% share of the total number of fines in the 2006-2009 period.

Fig. 5: Traffic accident causes in the whole UAE.

3. DISCUSSION AND CRITICAL ANALYSIS OF THE TRAFFIC ACCIDENTS DATA

As of March 2008 the Unified Federal Traffic Law stiffened the penalties for driving offences and also implemented a black point system for drivers in an attempt to improve safety on the roads. The


69

new laws work on a three-tier penalty system with fines for some offences, add black points and cars being impounded for more serious misdemeanors. Repeat offenders could ultimately face suspension of their license. Before the application of Unified Federal Traffic Law each of the seven emirates had its own traffic law, thus, basis for comparison for causes of traffic accidents before the year 2008 is not consistent. Consequently, it was decided that for the critical analysis of the traffic accidents data it more appropriate to consider only the period following the application of the Unified Federal Traffic Law (i.e., from 2008 to present).

3.1 Types and Frequency of Traffic Accidents in the UAE

According to the traffic accident report format used by the UAE Traffic Department, the accidents are divided into four categories: collision, roll-over, run over, and others (which includes all other types of accidents). Table 1 presents the number of traffic accidents per accident category for each of the seven emirates. It can be noticed for Table 1 that collision accidents type dominates by almost 60 to 63% followed by run over accidents type which accounts for about 20% to 23%, whereas, roll-over share was about 12% to 13%. Table 2 presents a comparison between the monthly numbers of accident for the whole UAE. It can be noticed that the distribution of the accidents is almost constant year round with a decrease during the month of July. The decrease in the number of accidents in July is justifiable as the population decreases during this month due to the summer vacation.

3.2 Traffic Accidents in the UAE with Respect to Driver’s Gender and Age Group

It seems that age has a direct bearing on how you drive and the likelihood of one being involved in an accident. Detail data made available to this study show that the MOI has detail age-wise break-up of those causing or involved in road mishaps. Drivers in the age group 31 to 45 are among the most dangerous, followed by drivers in the age group 46 to 60 (Figure 6a). As per Figure 6a, drivers above 30 years of age caused 81% of the traffic accidents during the three years period. In addition, Males was found to be the predominant gender involved in traffic accidents as shown in Figure 6b.

3.3 Traffic Accidents with Respect to Vehicle Type and Lighting conditions

The light vehicles are the major type of vehicles involved in traffic accidents as shown in Figure 7. The light vehicles dominated by 81%, followed by heavy trucks that were involved in 8% of the accidents. More than 92% of traffic accidents happened in good lighting conditions (either in daylight or in an acceptable lighting level).

Table 1: Number of traffic accidents per accident category for each of the seven emirates

Emirate2008 2009

collision roll-over run over others Total collision roll-over run over others TotalAbu Dhabi 1631 535 661 136 2963 1773 549 649 120 3091

Dubai 1683 98 423 15 2219 1433 91 275 28 1827Sharjah 375 70 321 19 785 473 107 313 21 914Ajman 217 27 182 4 430 193 23 187 14 417

Umm al-Quwain 62 48 22 7 139 101 71 12 1 185

Ras al-Khaimah 479 104 87 60 730 649 121 114 47 931

Fujairah 381 114 108 5 608 353 112 70 4 539Total 4828 996 1804 246 7874 4975 1074 1620 235 7904


70

Table 2: Comparison between the monthly numbers of accidents (in thousands) for the whole UAE over the last three years

TotalPercentageYears

Month20082009

20089612748Jan

18779661692Feb

18038579710March

17408637625April

18158619673May

16968584662June

16317599579July

16768602600Aug

18829748622Sep

18378712624Oct

19479748670Nov

20299773699Dec

2194110078747904Total

Fig. 6a: Traffic accidents in the UAE with Respect to age groups

(after 2008 period)


71

Fig. 6b: Traffic accidents (percentage) in the UAE with Respect to Gender (after 2008 period)

Fig. 7: Traffic accidents (after 2008 period) in the UAE with Respect to vehicle type

4. UAE ROAD TRAFFIC ACCIDENTS FATALITIES AND INJURIES

Reducing the number of fatalities and injuries are the most important goals of any traffic study. Important aspects related to traffic accident fatalities and injuries are presented in this section. This part gives an overview on the fatalities and injuries due to traffic accidents for the last ten years in the UAE. First the


72

data available on the fatalities are discussed and correlations with key factors are made. Then the injuries data is discussed followed by studying some correlations between fatalities and injuries. This part is concluded by a brief discussion on the cost of traffic accidents.

4.1 Traffic Accident Fatalities

The relation between the number of traffic accidents fatalities for the period from 2000 to 2009 and different key factors are discussed in the following sections.

4.1.1 Emirate-Based Traffic Accident Fatalities

The overall numbers of fatalities in the UAE for the last ten years are shown in Figure 8a. This figure shows that the years 2007 and 2008 witnessed a noticed increase in the number of fatalities compared to the previous years. Whereas, the year 2009 showed a decrease in the number of fatalities. For most of this period Abu Dhabi had the highest rate of fatalities followed by Dubai and then Sharjah. The rest of the Emirates had comparable numbers of fatalities. The figure shows also that Dubai had a major reduction in the last two years (2008 and 2009) which may be attributed to the implementation of the new traffic laws. Figure 8b showed that the fatality per 100 thousands had the highest rate of 24 at 2003, the rates have been decreasing since then, and it reached a rate of 12 in 2009. This shows that there is a reduction of 50% in the number of fatalities per 100 thousands of population from the peak value at 2003. The ratio of the number of road traffic violations to the number of fatalities due to road accidents is reviewed for each Emirate. Ajman has the largest ratio among all emirates. Comparing these ratios to the average ratios for all emirates over the ten years period showed that up to the year 2004 only Ajman had ratios higher than the average. In 2009 Ajman had the highest ratio followed by Dubai, then. Umm Al Quwain.

(a)


73

(b)

Fig. 8: Fatalities in the UAE (2000 – 2009): a) Overall numbers of, b) Number of fatalities per 100 thousands of population

4.1.2 Traffic Accident Fatalities According to Age-group, Gender and Citizenship

The ages of traffic accident fatalities can be divided into five groups as shown in Figure 9. The distribution of the total number of fatalities for the last 10 years among these five age groups showed that about 39% of the fatalities lied in the age group (from 18 to less than 30 years old) and about 33% in the age group (from 30 to less than 45 years old). Only about 28% of the fatalities lied in the age groups younger or older than these two age groups. These results show that the people ageing from 18 to less than 45 years are more prone to being a traffic accident fatality. The same conclusion can be withdrawn from the distribution of the fatalities in different age groups for each year in the last ten years.

Fig. 9: Distribution of fatalities percentages according to the age-group for the 2000 - 2009 period


74

In terms of gender, the percentage traffic accident fatalities of males and females shows that males form from 85% to 93% of the fatalities, whereas females form from 7% to 15%. This may be attributed to the fact that the number of male licensed drivers is higher than that for females. A review of the accidents data based on the citizenship category reveals that expats form from 67% to 79.5% of the fatalities, whereas UAE citizens form from 20.5% to 32%. These percentages indicate that expats’ fatalities from traffic accidents are more than twice that of UAE citizens. The distribution of the total fatalities among different nationalities in the UAE for the last ten years is shown in Figure 10. This figure shows that Asians’ had the highest percentage of fatalities. This result can be better interpreted through correlation with the demographic classification of UAE population. Literature showed that the demographic distribution of UAE is about 19% UAE citizens; 23% Arab, 50% Asian and 8% for other expatriates of the total population, respectively.

Fig. 10: Total fatalities distribution among different nationalities in UAE (2000-2009)

4.2 Traffic Accidents Injuries

The relations between number of injuries from traffic accidents and different key factors are discussed in the following sections.

4.2.1 Emirate-Based Traffic Accidents Injuries

The overall numbers of injuries due to traffic accidents in the UAE for the last ten years are shown in Figure 11. This figure shows that the overall injuries rates are slightly increasing over the last years. For this period Abu Dhabi had the highest rate of injuries followed by Dubai and then Sharjah and Ras Al Khaimah. The rest of the Emirates had comparable numbers of injuries. The figure shows also that Dubai had a slight reduction in the last year (2009). The injuries per 100 thousands had a decreasing trend among the last ten years with the rate of 147 in year 2009. This shows that there is a reduction of 58% in the number of injuries per 100 thousands among the last ten years. Studying the average number of injures per accident in each emirate for the last ten years shows that it ranged from 0.75 to 1.66. Among all emirates Dubai had the largest followed by Abu Dhabi and Ajman. On the other hand, Umm Al Quwain had the smallest number of injuries per accident.


75

Fig. 11: Overall numbers of traffic accidents’ injuries in UAE for the last ten years

4.2.2 Traffic Accidents Injuries According to Citizenship

The distribution of the total injuries among different nationalities in the UAE for the period from 2006 to 2009 is shown in Figure 12. The percentage distribution of injuries among nationalities is close to that of fatalities (discussed earlier). This figure shows that Asians had the highest percentage of injuries and the citizens of the Gulf countries had the lowest percentage.

5. TRAFFIC ACCIDENT COST ESTIMATION

Based on the information provided by the Ministry of Interior, the cost of each traffic accident fatality in million dirhams is estimated to be 6.49 and the cost of injuries is about 0.9 millions (based on the distribution shown in Table 3). The trend of the costs due to fatalities and injuries from traffic accidents followed the same trend as that for the accidents as expected (Figure 13). The average cost per year of fatalities for the last ten years was ADE 5672 millions and the average cost per year of injuries was ADE 9873 millions. The cost of fatalities and injuries for each emirate shows that it was the highest for Abu Dhabi then Dubai.

Fig. 12: Distribution of the total injuries among different nationalities in UAE


76

Table 3: Cost estimation of a traffic accident injury

Estimated CostCost ($)Injury PercentageInjury type

356400019800001.80Very Severe

44730006300007.10Severe

758845034200022.19Moderate

868085112600068.90Simple

24306301

243063.01Total ($)

0.2431Cost (Million $)

0.8945Cost (Million AED)

(a)

(b)Fig. 13: Fatalities (a) and Injuries (b) costs in million AED for the period 2000 – 2009


77

6. TRAFFIC ACCIDENTS IN UAE AND HUMAN/DIVER FACTOR

Driver has a very important role in traffic accidents and most of the time he holds the major responsibility as he has the highest control level on the road. In this section we will focus on identifying the relationships that drivers› licensing, drivers› citizenships and vehicle types have with traffic accidents.

Reviewing the statistics related to driver licensing over the past ten years from 2000 to 2009 (Figure 14) reveals that the total number of driver›s licenses issued over these years has increased significantly reaching a total number of 496138 in 2009 compared to 262045 in 2000. The majority of these licenses were issued for the first time (new licenses especially after 2003) while the rest were renewals. The large number of newly issued licenses can be related to traffic accidents in that most of new license holders are usually lack the driving experience which increase the risk of getting involved in an accident. From these statistics we calculate the average number of licenses per accident, per fatality and per injury (Figure 15). As can be seen from the figure, the average number of licenses per accident, per fatality and per injury is increasing over time with the rate for fatality being the highest. This match the increase in the total numbers of issued licenses. However in order to understand the real trend we need to compare these rates to the total number of issued licenses. So we divided these license averages per accident, per fatality and per injury by the total number of licenses (Figure 16). Results show that roughly for accident and injury categories, the average number of licensees associated with each accident/injury is almost constant over years. While for fatality, it›s clear that the average number of licensees associated with each fatality is decreasing over years; which indicates that there is an increase in the accident severity (fatality) over time with more cars being involved in these accidents upon their occurrence.

Fig. 14: Total, new and renewed license numbers for a ten year period


78

Fig. 15: Average numbers of licenses per accident, per fatality and per injury (in thousands)

Finally, looking at the statistics that’s related to the average number of fines per license (Figure 17) shows that there is a significant increase in the number of fines issued per license (13 in 2009 and 5 in 2000 on average). This can be related to the large increase in the number of driving licenses issued over the past ten years (especially new ones) from one side and the higher level of law enforcement from the other side.

Fig. 16: License averages per accident, per fatality and per injury as percentage of the total # of licenses


79

On the other side, looking at accidents› totals and percentages distributions according to the citizenship of the driver for all vehicle types reveals that few citizenships have a major contribution to traffic accidents. Results indicate that for all types of vehicles, the highest percentage of accidents was for emirates nationals drivers followed by Pakistanis then come Indians drivers. The fact that many Pakistanis and Indians work as drivers may explain their high percentage of involvement in traffic accidents. A proposal might be to develop a traffic safety trainings and brochures in different languages to increase the traffic safety awareness for different citizenships.

Fig. 17: Average numbers of fines per license (total in thousands)

Based on the obtained results, a general trend can be seen also for all citizenships where there is an increase in the number of traffic accidents related to light vehicle type. The rate of increase for the major citizenships is slower between the years 2008 and 2009 after the implementation of the united traffic enforcement law compared to the rates of the previous years.

7. COMPARISON OF ROAD TRAFFIC ACCIDENT STATISTICS WITH OTHER COUNTRIES

To compare the road accident fatalities statistics for the UAE with those for the Gulf countries, Arabic countries and worldwide, the latest statistics provided by the World Health Organization (WHO) was used. The latest WHO report (WHO [15]) provided statistics for road accidents facilities for the year 2007 for most counties and that for 2006 or 2005 when not available for latter years.

The UAE shows the second highest number of accident fatalities and injuries among the Gulf countries after Saudi Arabia. If you sort the Gulf countries once based on the number of road accidents fatalities and another time based on the population you will almost get the same order in both cases. The number of accidents per 100 thousand of population is used to eliminate the effect of the difference in population. Oman had the highest number of fatalities per 100 thousand of population followed by


80

Saudi Arabia and then UAE and Qatar. Bahrain had the lowest number of fatalities per 100 thousand. On the other hand, Bahrain had the largest number of road injuries followed by Oman then Kuwait then UAE. The Saudi Arabia had fewer injuries per 100 thousand of population than UAE. The percentage of each gender (males and females) from the total number of road accident fatalities shows that in all Gulf countries the percentage of female fatalities is way smaller than that for male fatalities. This may be attributed to the fact that more males have licenses than females in these countries.

A review of the data showing which Gulf countries has seatbelt laws, the level of enforcement of these laws and the seatbelt wearing rate had been done. All the Gulf countries had seatbelt laws for front seat passengers and only Saudi Arabia had laws for all passengers. The level of enforcement of these laws was equal to or less than 61% for all Gulf countries except Oman that had a 95% level of enforcement. There was no correlation found between the rate of wearing the seatbelt for the front passenger and the number of accident fatalities or injuries per 100 thousand of population.

The road accident fatalities and injuries for six selected Arabic countries (Libya, Egypt, Jordan, Lebanon, Yemen and Iraq) were reviewed. Egypt had the highest number of fatalities (12,295) and injuries (154,000) among the selected Arabic countries, whereas the UAE had the fifth highest number of fatalities and the fourth highest number of injuries. Based on the number of fatalities per 100 thousand of population the UAE (24) had the second largest number among the selected Arabic countries after Libya (35). It also had the second largest number of injuries per 100 thousand of population after Jordan (303). Among the selected Arabic countries the percentage of female fatalities is way smaller than that for male fatalities. This may be attributed to the fact that more males have licenses than females in these countries.

Furthermore, the number of fatalities and accidents in the UAE is compared to some selected worldwide countries of interest: USA, Canada, England, Switzerland, Sweden, Japan and South Africa. The UAE shows the second highest number of accident fatalities per 100 thousand of population among the selected worldwide countries after South Africa. On the other hand, it has the lowest number of road accident injuries per 100 thousand of population. The percentage of each gender (males and females) from the total number of road accident fatalities is reviewed. In the selected countries the percentage of female fatalities is way smaller than that for male fatalities. The difference between the percentage of male and female accident fatalities is smaller in these selected worldwide countries than those for the Gulf countries or the selected Arabic countries.

8. SUMMARY

The UAE has since its establishment in 1971 comes a long way in a fairly short time. Prosperity such as that experienced today was unheard of in the 1970s but with the discovery of oil in the early 1960s the level of income (and expenditure) began to rise and motor vehicle ownership began to spread amongst the population. Accordingly, an increase in vehicle numbers accompanied by rapidly expanding road construction took place. The result has been a large increase in the numbers of road traffic accidents with causalities and fatalities creating serious public problems. Consequently, this article came to investigate traffic accidents rate and causes and to find recommendations to reduce the traffic accidents hazard. Data on traffic accidents in the UAE were collected from the General Administration of Traffic Coordination department at the Ministry of Interior (MOI). The collected data covered the period from 2000 to 2009. The collected data included the traffic accident report data for the UAE as whole, as well as, the same set of data for each of the seven emirates. The data included accident date, time, location, collision type, number of injuries, and number of fatalities.


81

Using the obtained data, a comprehensive critical analysis was conducted to establish a pattern of conditions and causes of traffic accidents where different accidents trends were deduced. The analysis revealed aggregated trends and correlations between accidents, injuries, fatalities, vehicle population and type, driver’s age, gender, and nationality. Furthermore, the study includes a comparison between the causes and rates of traffic accidents in the UAE with different GCC countries, selected Arabic countries and selected worldwide countries.

Through investigating the causative factors of traffic accidents in the UAE, a number of issues such as causes of accidents and characteristics of road users were considered in the study. Based on the performed critical analysis, the total number of traffic accidents in the UAE per 100 thousand of population dropped by about 70% over the last ten years (2009 in comparison with 2000). Abu Dhabi witnessed 38% of the traffic accidents; the highest number among the seven emirates, followed by Sharjah (20%), then Dubai (19%). Among many causes of accidents, the main cause of accidents is the disrespect of other road users, it accounted for almost 22%. The “other” category of the accident causes was found to account for nearly 29% of the causes as revealed from the data collected from the traffic accident reports. This very high percentage indicates that the accident report format needs some alternations to better categorize the traffic accident causes and make the judgment easier to traffic officers in the accident site. More than 90% of the accidents maybe associated with the driver faults/behavior. The most predominant cause of fatalities that accompany the accidents is the speeding cause (accounts for about 47%).

As per critical analysis of the data following the application of the Unified Federal Traffic Law (2008), for the period from 2008 to 2010, collision accidents was found to dominate by almost 60 to 63% followed by run over accidents type which accounts for about 20% to 23%, whereas, roll-over share was about 12% to 13%. The monthly distribution of the traffic accidents is almost constant year round with a decrease during the month of July. The decrease in the number of accidents in July is justifiable as the population decreases during this month due to the summer vacation.

Drivers in the age group 31 to 45 are among the most dangerous, followed by drivers in the age group 46 to 60. Drivers above 30 years of age caused 81% of the traffic accidents during the three years period. Males were found to be the predominant gender involved in traffic accidents. The light vehicles were involved in 81% of traffic accidents, followed by heavy trucks that were involved in 8% of the accidents.

The article also provided an overview of the fatalities and injuries due to traffic accidents for the last ten years in the UAE. In addition, the economic impact of traffic accidents was discussed. The overall numbers of fatalities in the UAE for the last ten years witnessed a noticeable increase in year 2007 and 2008 compared to the previous years. Whereas, the year 2009 showed a decrease in the number of fatalities. The number of fatalities per 100 thousand of population dropped from 21 in 2000 to 12 in 2009 which reflects a major improvement in reducing the severity of traffic accidents. For most of this ten years period Abu Dhabi had the highest rate of fatalities followed by Dubai and then Sharjah. The distribution of the total number of fatalities for the last 10 years among different age groups showed that the people ageing from 18 to less than 45 years are more prone to being a traffic accident fatality. 39% of the fatalities lied in the age group (from 18 to less than 30 years old) and about 33% in the age group (from 30 to less than 45 years old). The same conclusion can be extended to the traffic accident injuries for the period from 2007 to 2009.

The overall numbers of injuries due to traffic accidents in the UAE showed that the overall injuries numbers are slightly increasing over the last years. On the other hand, the number of injuries per


82

100 thousand of population showed a 58% reduction in the number of injuries per 100 thousands among the last ten years. For this ten years period Abu Dhabi had the highest numbers of injuries followed by Dubai and then Sharjah and Ras Al Khaimah. The distribution of the total injuries among different nationalities in the UAE for the period from 2006 to 2009 showed that Asians’ had the highest percentage of injuries and the citizens of the Gulf countries had the lowest percentage.

Based on the information provided by the Ministry of Interior, the average cost per year of fatalities for the last 10 yours was found to be AED 5672 millions and the average cost per year of injuries ADE 9873 millions. Abu Dhabi had the highest cost of fatalities and injuries followed by Dubai.

Driver has a very important role in traffic accidents and most of the time he/she holds the major responsibility (more than 90%) as he/she has the highest control level on the road.

There has been an excessive increase in the number of new driver›s licenses issued over the past 10 years where they almost doubled in number in 2009 as compared to that in 2000 (496138 in 2009 while it was 262045 in 2000). Most of these licenses are newly issued licenses (in 2009, 325523 new licenses and 170616 renewed licenses). The lack of experience that usually associates new drivers can play a major role in traffic accident occurrence. For fatality associated accidents, results indicate that the average number of licenses associated with each fatality is decreasing over years which indicate that there is an increase in the accident severity (fatality) over time with more cars being involved in these accidents upon their occurrence.

A significant increase in the number of fines issued per license can be seen clearly which rises on average from 5 fines/license in 2000 to 13 fines/license in 2009. The major increase happened after year 2008 after the implementation of the unified law which reflects that a stronger traffic enforcement law took place which played a major role in traffic accidents reduction. Regarding the citizenship of the drivers who are involved in traffic accidents, results indicated that for all types of vehicles emirates nationals drivers represent the citizenship with the highest involvement rate in traffic accidents followed by Pakistanis then comes the Indians drivers.

9. RECOMMENDATIONS

Based on the statistical analysis performed in this article and our field work, it was clearly noticed that the traffic accident report needs to be modified to include more details about the traffic accident such as traffic accident exact location using a GPS system to record the traffic accidents location coordinates by traffic police at the time of accident. Accident locations can be then identified on digital GIS maps to highlight the High Accidents Locations (HALs) for further remedy of such sites. Accident reasons need to be revised and to be made easier to measure. For example, disrespect of other road users is hard to measure by the traffic policeman and other reasons that can be related to the traffic accident such as not yielding, not slowing down at pedestrian crossing areas…etc can be used. It is better to include the geometric details of the highway at the accident location in the accident report such as the highway width, median geometry...etc to better understand the relationship between road geometry and traffic accidents. This can also include information about traffic control devices used at the intersection/highway to measure the commitment level of drivers. Details related to passengers such as whether they were wearing the seatbelt or not can be added to the report. Sections about injuries, fatalities can be also added to the traffic accident form.


83

A GIS system needs to be developed. This system should present an interactive database for all traffic accidents distributed according to their spatial locations on the street network. The database needs to include also all other attributes related to these traffic accidents. This can help in understanding the spatial variability of these accidents over the road network and in identifying the HALs sections of the network. The GIS system can allow us also to relate these accidents with different parameters such as driver characteristics, geometric design..etc.

As speed is considered a crucial factor in fatality related accidents, a review of the speed limit over highway sections that are exposed to high accidents rates is necessary. According to prevalent traffic, road, and accidents condition, speed limit might need to be revised to enhance traffic safety levels. It’s also recommended to check other geometric highway details such as highway stopping sight distance, intersections’ sight distances …etc to check their compatibility with the current road conditions. Traffic lights performance need to be checked to ensure a better traffic circulation system especially in terms of signals coordination, dilemma zones eliminations, etc. It’s important to perform traffic volumes counts for highway sections with high accidents rates to check if the road is working at or near capacity and how can this be related to traffic accidents records. Vehicle registration (especially for insurance) can be related to driver history of fines and traffic accidents. Higher insurance can be assigned to drivers with large fines and accidents record. Development of special programs for training, assessment and monitoring for the highest involved citizenships in traffic accidents is needed. These can be done in different languages to ensure their effectiveness.

Finally, several traffic safety issues are still in need of further analysis and discussion. For example, the causes reported by the police are only specific to each case. These causes are not the same as overall (system-wide) causes for traffic accidents. In other words, while these data might show that 90% accidents are caused by driver error, how do we know that diver error is not caused by, say, “improper licensing”, or “overworked taxi drivers”, or “faulty highway design” or any other reason. Further, we do not know what the police means by “disrespect of other road users”, we also do not know why many of the other causes (e.g. “tail-gating”) do not fall within the “disrespect” category. There is a need to clarify what causes are contained within the “disrespect” category.

Further, the critical analysis in the paper still needs important checks to be implemented in the future. For example, the analysis finds that drivers in the age group 31-45 the most “dangerous”. Maybe this age group has the highest number of drivers, and so it is normal that they also have higher representation in the accidents. This data must be compared with data on drivers/ license holders to obtain a clearer picture about whether or not a group is overrepresented in the accidents. Citizenship in section 4 needs to be analyzed further. The categories used for analysis are “UAE”, “GCC Arab”, “Non-GCC Arab”, “Asian” and “Others”. These categories do not a measure citizenship. A “critical” analysis ought to raise a question about this categorization.

The paper has to also study many of the potential causes of accidents – including but not limited to – highway design, licensing procedures, management of taxis, policing issues etc. With the available data it is difficult to make any real recommendations – and as it turns out the recommendations made is simply a request for more data.

10. REFERENCES

[1] Abdel-Aty M.A. and Radwan A.E., (2000). Modelling traffic accident occurrence and involvement. Accident Analysis and Prevention, Vol. 32 (5):pp 633-642.


84

[2] Bener A., Crundall D., Haigney D., Benisiali A., Al-Falasi A., (2004.) Driver Behavior, Stress, Error and Violations on the Road: A cross-cultural comparison study, 3rd International Conference on Traffic &Transport Psychology, 5-9 September 2004, Nottingham, UK.

[3] Bener A., Ozkan T., Lajunen T., (2008). The Driver Behaviour Questionnaire in Arab Gulf countries: Qatar and United Arab Emirates, Accident Analysis & Prevention, Vol. 40 (4): pp 1411-1417. [4] Dobson A., Brown W., Ball J., Powers J., McFadden M., (1999). Women drivers’ behaviour, socio-demographic characteristics and accidents. Accident Analysis & Prevention, Vol. 31: pp 525-535

[5] Elsenaar P. and Abouraad S., 2005. Road Safety Best Practices: Examples and Recommendations, Global Road Saftey Partnership , retrieved on December 15, 2012 from http://www.grsproadsafety.org/themes/default/pdfs/Road%20Safety%20Best%20Practices.pdf

[6] Hauer, E., (1996.) Detection of Safety Deterioration in a Series of Accident Counts, Transportation Research Record No. 1542 Safety and Human Performance/Statistical Methods and Accident Analysis for Highway and Traffic Safety, Transportation Research Board/National Research Council, National Academy Press, Washington, D.C., 1996, p. 38-43.

[7] Kontogiannis T., Kossiavelou Z., Marmaras N., (2002.) Self- reports of aberrant behaviour on the roads: errors and violations in a sample of Greek drivers. Accident Analysis & Prevention, Vol. 34: pp 381-399.

[8] Lajunen T., Parker D., Summala H., 2004. The Manchester Driver Behaviour Questionnaire: across-cultural study. Accident Analysis & Prevention, Vol. 36: pp 231-238.

[9] Lord D. and Mannering F., (2010) The statistical analysis of crash-frequency data: A review and assessment of methodological alternatives. Transp. Res. A: Policy Pract., Vol. 44(5) :pp 291-305.

[10] Reason, J., Manstead, A., Stradling, S., Baxter, J., & Campbell, K., (1990). Errors and violations on the roads: A real distinction?. Ergonomics,Vol. 33 ,pp 1315−1332.

[11] Stokes, R. W. and Mutabazi M. I., (1996). Rate-Quality Control Method of Identifying Hazardous Road Locations. Transportation Research Record No. 1542 Safety and Human Performance/Statistical Methods and Accident Analysis for Highway and Traffic Safety, Transportation Research Board/National Research Council, National Academy Press, Washington, D.C., , p. 44-48.

[12] Tarko, A. P., Sinha K. C., and Farooq O., (1996). Methodology for Identifying Highway Safety Problem Areas”, Transportation Research Record No. 1542 Safety and Human Performance/Statistical Methods and Accident Analysis for Highway and Traffic Safety, Transportation Research Board/National Research Council, National Academy Press, Washington, D.C., , p. 49-53.

[13] Treat, J. R., et.al (1977) Tri-level study of the causes of traffic accidents. Report No. DOT-HS-034-3-535-77 (TAC).

[14] Vistisen, D., (2002) Models and methods for hot spot safety work, PhD thesis, Technical University of Denmark.


85

[15] World Health Organization, (2009) Global status report on road safety time for action. retrieved on December 15, 2012 from: http://www.who.int/violence_injury_prevention/road_safety_status/2009/en/index.html.

[16] Zhou, M. and Sisiopiku, V. P., (1997) Relationship Between Volume-to-Capacity Ratios and Accident Rates, Transportation Research Record No. 1581, Safety and Human Performance/Traffic Records, Accident Prediction and Analysis, and Statistical Methods, Transportation Research Board/National Research Council, National Academy Press, Washington, D.C., , p. 47-52.

87


A GIS-BASED PLANNING STUDY FOR A GEO-ENVIRONMENTAL DISTRICT SELECTION IN NORTH

JORDAN

Sharaf A. Alkheder1*, Ahmad A. Al-Malabeh2, Mohammad N. Sada3

1 Department of Civil Engineering, Faculty of Engineering & Applied Sciences, ALHOSN University, P.O. BOX: 38772 Abu Dhabi, UAE

2 Department of Earth and Environmental Sciences, Faculty of Natural Resources and Environment, Hashemite University, P.O. Box 150459, Zarka , 13115, Jordan

3 Hashemite University, P.O. Box 150459, Zarka , 13115, Jordan

ABSTRACT: Jordan is witnessing an active process of land development in major cities and surrounding rural areas associated with the prosperous and open investment environment. This is applied also to north Jordan area that is involved in such excessive development process related to the fast growth in population and demand on infrastructure services. One target of land development planning in north Jordan is to reserve important environmental areas/districts for parks establishment within the master plan of the area to provide open spaces for recreational activities for citizens and at the same time to preserve the geo-environmental diversity existing at such areas. This paper focuses on using multi-criteria approach and GIS to select a geo-environmental district in north Jordan within the richest bio-geo-diverse system (Wadi Zeglab area) to be marked on the area’s master plan for future land development. Multiple criteria for district selection including accessibility to major roads, slope, and location from natural & urban areas were used. All selection parameters were mapped and digitized in GIS to perform the selection among all available layers of data. Results indicated the existence of a number of candidates that can be selected as protected geo-environmental districts in north Jordan that were highlighted on the final GIS maps. These suggested areas need to be included in future land development plans of the region.

KEYWORDS: Multi-Criteria Selection, GIS, Geo-Environmental District, Land Development, Planning.

1. INTRODUCTION

With the fast pace of land development activities taking place mainly in major urban areas and extending to various rural areas, it becomes so vital to highlight districts of environmental importance for protection. This usually done as a part of land development planning to provide proper enforcement level for preserving such important districts in any future development plans. However, it’s important first to identify the locations of such environmental areas with high importance according to predefined set of selection criteria.

Multi-criteria analysis as integrated with GIS techniques provide unlimited capabilities for land management and planning purposes. Much literature can be found on using such integrated techniques for criteria-based site selection. A major area of application is land suitability mapping for different purposes (Hopkins [10]). Collins et al. [7] and Malczewski [18] presented a review of the application of multi-criteria analysis and GIS in land-use suitability analysis. Joerin and Musy [12] developed a GIS-based multi-criteria model (MAGISTER) that was used, with a set of 8 criteria, to generate a


SHARAF A. ALKHEDER, AHMAD A. AL-MALABEH, MOHAMMAD N. SADA

88

land suitability map for housing purposes taking into account the decision-makers’ perception of land management principles. Land-use suitability analysis was also performed using GIS and outranking multicriteria analysis to select most suitable areas for housing activities in a selected district in Switzerland (Joerin et al. [13]).

Malczewski [19] preformed an in-depth recent review of the rich literature available regarding GIS-based multi-criteria decision analysis. The review regarding the strong connectivity between GIS and multi-criteria analysis techniques as can be seen in the work of Laaribi et al. [16], Malczewski [17], Thill [31], Chakhar and Martel [6] showed clearly the different types of applications of such integrated techniques in the literature. Malczewski [19] classified the application arenas of such integrated techniques into a number of major categories including: environmental modeling as in the work of Noss et al. [26] and Seppelt and Voinov [30], transportation studies as shown in the work of Jha et al. [11] and Bowerman et al. [5], water and waste management (Martin et al. [20]; Leaõ et al. [15]) and many other application areas.

Land management planning is a necessary tool of interest for engineers and urban planners in Jordan to specify the different land uses and reduce the degradation of certain important environmental districts in the country especially in the northern area (Wadi Zeglab district). Many efforts can be seen in this direction starting by the 1965 soil survey done by a team from Durham University (Fisher [9]). The survey provided primary information regarding the watershed management system at Wadi Zeglab district and produced detailed soil, land-use and land-suitability maps for the area (Fisher [9]). Later in 1995, a hydro-geological study was performed to assess the water resources, in terms of rainfalls, runoff, evaporation and infiltration, in Wadi Zeqlab catchment area besides investigating the sedimentary rocks formations that precipitated in the marine environment (Al-Zu`bi [4]). Nuafleh [27] studied the different types of mass wasting features in Wadi Zeglab drainage basin such as soil and rock creep related landslides besides the human factors that contribute to such mass wasting. Such landslides were further investigated by Field and Banning [8] focusing on studying their Geomorphologic and sedimentological characteristics. In a recent study (Ziadat et al. [33]), high resolution suitability maps were produced for the region, based on a detailed soil map, through the use of remote sensing and GIS technologies. Radaideh [28] performed a research through which he evaluated the woodlands and ranges in Wadi Zeglab district besides identifying the effect of natural and human factors on such resources.

The importance of Wadi Zeglab district reflected in its important geographic location in the northern region of Jordan and its unique bio- geo- diversity system raises the issue of designating a protected area within the district to be marked on the area’s future development plans as a part of a sustainable land development planning scenario. This will protect the area from different kinds of environmental threats that already taking place at some parts of the Wadi such as turning certain places into liquid/solid landfills besides the unplanned excavation of the region to use its valuable materials in different construction industries (Al-Malabeh et al. [2]).

2. RESEARCH SIGNIFICANCE

This work identifies the GIS application of multi-criteria decision making. The study area is in northern Jordan and represents an important region of the country for agricultural and eco-tourism activities. The fact that there is a lot of literature on the GIS application of multi-criteria modeling, the advancement of knowledge in this research is on the application of the method in Jordan. There is a lot of published literature about the use of GIS, Multi-criteria or suitability analysis. However, none of the published work focus on an area similar to the one used in our case study. The study area is witnessing a clear practice

A GIS-BASED PLANNING STUDY FOR A GEO-ENVIRONMENTAL DISTRICT SELECTION IN NORTH JORDAN

89

of improper planning of land use as can be seen in uncontrolled urban sprawl, mixed land uses, absence of regulatory laws regarding land management and the increasing threats to the natural environment through man-made projects. Implementing this methodology to this area for proper and sustainable management of the available natural rich bio-diversity environment, through Geopark protection area selection, is unique to the area and will highlight these areas for protection from external human threats on the area’s master plan. Building such extensive GIS database will also provide a benchmark for temporal and spatial monitoring of the study area over time to assess the positive and negative changes in the landscape and to design and implement suitable mitigation plans to respond to these changes.

3. STUDY AREA FEATURES

North Jordan holds marvelous and unique bio-geo-diverse environments scattered all over the area (Al-Malabeh and Al-Shreideh [3]). Among these is Wadi Zeglab district (an approximate total area of 107 km2) that is located at about 80 km to the north of the capital city of Jordan (Amman) and 25 km northwest of Irbid (a major city in north Jordan) as can be seen in Figure 1. Wadi Zeglab is part of the north rift basin (Figure 2 (Ministry of Water & Irrigation [22]) & Figure 3) with a lowland zone at the west side (lands with elevations less than 400m above MSL) (Mohammad [23]). The Wadi contains distinguished natural land cover of forests and agricultural crops, waterfalls (three major ones with an approximate total water discharge rate of 8 MCM) and dams, and unique geological formations of various sedimentological, structural and geomorphologic features (Al-Malabeh and Al-Shreideh [3]). Wadi Zeglab geomorphology is a result of the headward erosion of the major Wadis into the Arabian plate following the subsidence and faulting of Dead Sea-Jordan Valley area (Mohammad [23]).

Regarding the Wadi climate, it is considered a part of the Mediterranean climate and usually categorized as semi-humid climate with all precipitations take place in winter (Fisher [9]). Its climate is located between upland climate in the east (moderate temperatures in summer, and cold winter) and Ghor climate in the west (high temperatures in summer and warm winter). Metrological data for the 1986 to 2003 period taken from the two closest station to the study area, one located at upland area (Ras Munief station) and the other at Ghor area (Der Alla station), regarding monthly variation (starting January (denoted on the graph as 1) to December (denoted on the graph as 12)) of the average temperatures (Figures 4.1 & 4.2) and relative humidity (Figures 4.3 & 4.4) are shown (Meteorological Department [21] and Jordan Valley Authority [14]).

4. RESEARCH OBJECTIVES AND METHODOLOGY

Due to the scarcity of research directed to study such important region, this work comes to fill the gap in this regard. The main research objectives are to use GIS to create a complete digital database of all natural (geological, hydrological and biodiversity spatial data) and man-made features in the study area and use it along with Multi-criteria analysis to select a geo-environmental district in Wadi Zeglab area as a protected area to be marked on the future land development masterplans. This selection is based on a set of expert-based predefined criteria implemented on the geo-referenced GIS multi-layered system including data on topography, road network, water resources, soil, and land use. Such selection of the geo-environmental protected area will ensure a suitable environment to conserve the diverse and valuable ecosystem resources available in Wadi Zeglab. This action will contribute significantly in protecting these natural resources especially vegetation from different environmental threats such as blazing, trees overcutting, over grazing, pollution, and human consumption of agricultural lands for construction activities such as roads and housing.


90

Fig. 1. A location map of Wadi Zeglab


91

Fig. 2. Wadi Zeglab location according to main water basins


92

Fig. 3. Google Earth images showing the Wadis


93

Fig. 4.1. Monthly average temperatures in Ras Munief - Ajlun station for the period 1986 to 2003

Fig. 4.2. Monthly average temperatures in Der Alla station for the period 1986 to 2003

Fig. 4.3. The monthly average humidity in Ras Munief- Ajlun station for the period 1986 to 2003

Fig. 4.4. The monthly average humidity in Der Alla station for the period 1986 to 2003


94

4.1 Multi-Criteria Decision Making and GIS

According to the International Society on Multiple Criteria Decision Making [36], Multi-Criteria Decision Making is defined as “the study of methods and procedures by which concerns about multiple conflicting criteria can be formally incorporated into the management planning process”. Drobne and Lisec [37] provides in depth review of the theory behind spatial Multi-Criteria Decision Making. They show comparison between Classic and GIS-based spatial decision-making procedures. GIS-based spatial decision-making procedure (Figure 5.1) is the one that was followed in this work.

Fig. 5.1. GIS-based multi-criteria spatial decision-making procedure (Drobne and Lisec [37])


95

4.2 Data Collection and Processing

The main raw spatial data used in this research were gathered from different sources as shown in Table 1. Further data were collected from hydro-geo-spatial data archives, Aerial photographs, through field investigation, and through GIS digitization of topographical and geological maps, roads, drainage networks, biodiversity, and other types of maps. Most of the collected spatial data layers were in Palestine Grids and Jordan Transverse Mercator (JTM) coordinate systems, so they were registered to Universal Transverse Mercator (UTM) to be used as the standard reference system for this work.

Table 1. Raw spatial data

Layer Name Layer Preview Details Source

Geologic Map

Resolution = 1:50,000Format: Raster

Natural Resource Authority (NRA)

(Jordan)

Topographic Map

Resolution =1:50,000Format: Raster

Royal Jordanian Geographic Centre

Catchment Area

Format: VectorMinistry of Water and

Irrigation, Jordan

Landuse

Format: Vector Mohawesh [24]

As our study area is covered by four topographical maps with a scale of 1:50000, mosaic process was used to get them as one layer as shown in Figure 5.2 (RJGC, 2008). A contour map, with a 50 m contour


96

interval, was generated for the study area through digitizing the topographic map in ArcGIS as shown in Figure 6. Huge variation in elevation can be easily noticed that ranges from as low as 100 below MSL for certain areas to reach an elevation as high as 1050 above MSL.

Fig. 5.2. Topographic maps mosaic for the study area


97

Fig. 6. A contour map of the study area resulted from digitizing the topographic map

Using 3D analyst, the contour map was converted to TIN surface to better understand the topography and hydrology of the study area as shown in Figure 7. Spatial analyst was used then to convert the TIN to a slope layer as can be seen in Figure 8.

Through digitization process in ArcGIS, simplified and detailed versions of geologic maps were produced as shown in Figure 9 (reproduction after NRA, 2006). These figures show that the study area is mainly composed of three geological groups: Ajlun Group, Belqa group and Jordan Valley Group. These groups are presented in details in Table 2 (different references such as Mohammad [23] and Abed [1]). These groups refer to the Mesozoic (upper cretaceous) Mohammad [23]. Also, it can be clearly seen that Wadi Al-Sir Formation is the dominant formation in the study area.


98

Fig. 7: TIN model for the study area

Fig. 8. Slope layer of Wadi Zeglab in degrees


99

Fig. 9.1. Simplified geological map of the study area

Fig. 9.2. Detailed geological map of the study area


100

Table 2. Geological groups and formations in Wadi Zeglab

Group Formation AgePleistocene and

Holocene SedimentSoil over bedrock, Calcrete Pleistocene- Holocene

Jordan Valley GroupIrkheim Lower Pleistocene

Waqqas Conglomerate Lower Pleistocene

Belqa Group

Shalala Chalk B5 Eocene

Umm Rijiam Chert B4 Paleocene-Eocene

Muwaqqar chalk Marl B3 Maestrichtian

Al-Hisa Fhosphorit B2b Campanian-Maestrichtian

Amman Silicifed B2a Campanian

Wadi Umm Ghudran B1 Santonian

Ajlun GroupWadi Al-Sir A7 Turonian

Wadi Sheib A5-A6 Turonian

Based on the Geological maps, a structural geology map was digitized as shown in Figure 10. Three main faults sets can be identified in Wadi Zeglab area: the NNW and NNE, N-S and E-W faults that are mostly normal faults (Nuafleh [27]).

Regarding the hydrology of Wadi Zeglab, two GIS layers were produced presenting the rainfalls stations (Figure 11 (Jordan Valley Authority [14])) and the stream system in the Wadi (Figure 12). Figure 12 was produced through digitization of the topographic map. Part of the hydrology system in the area is Wadi Zeglab dam shown in Figure 12 that has a maximum height of about 47.5 m and original reservoir volume of about 4.3 MCM (Mohammad [23]). Figure 13 shows some photos for the dam.

Fig. 10. Structural geology map of the study area


101

Fig. 11. Wadi Zeglab rainfalls station

Fig. 12. Stream system of Wadi Zeglab


102

A

Dam Body

B

Up Stream

Fig. 13 (A&B Wadi Zeglab Dam Due the important role of soil characteristics in selecting the best candidate site for geo-environmental district area, a GIS layer for Wadi Zeglab soil classification system is needed. This layer was obtained from the database of the Jordanian National Soil Map and Land use Project [25] as can be seen in Figure 14. The soil classification system divides the soil into map units where each unit is given a number associated with the properties of the soil it represents. The map units’ scale ranges from map unit No. 1 to map unit No. 82 with map unit No.0 being bare soil, and map unit No.999 being urban area. Wadi Zeglab area includes the following map units’ numbers with a brief description of their soil characteristics (National soil map [25]):


103

Fig. 14. Soil map for the study area

• Map unit No. 1: Extremely deep soil having mature profiles with about 120cm of brown silty clay to clay overlaid by extremely indurate red brown clay with parent rock being cherty limestone. Calcium carbonate concretions are noticeable in lower horizons. Such soil exists in areas with 300m to 670m altitudes and 0 to 16% slope (National soil map [25]).

• Map unit No. 7: Thin soil categorization associated with outcrops of more cherty limestone that exists at areas with 400m - 1150m elevations and 0-25% slope. The texture of this soil class ranges from clay loams through silt loams to silty clays that are dominantly brown in color. This soil type has a 44-62% clay content range, 27-43% silt content, PH range of 7.1 to 7.8, and cation exchange capacities between 39 to 55 m.e.q/100g. The parent rock of this category is cherty limestone (National soil map [25]).

• Map unit No. 10: This soil category, moderate to shallow depth, is available in areas with 450m to 1000m elevation and slope range of 5 to 40%. The parent rock of this soil class is limestone (National soil map [25]).

• Map unit No. 17: Moderate deep to shallow soil, of alluvial and colluvial origin that usually overly slope deposit. Soil ranges from loosely massive loams at surface to heavily compacted clays lower down


104

with limestone being the parent rock. Such soil can be found in areas with 400m-1000m elevations and 9-60% slope (National soil map [25]).

• Map unit No. 23: Soil in this class is characterized by heavy texture with clay content of about 55% to 73%, 21% to 64% Silt content, and low uniform sand content. Parent rock for this soil is limestone and its PH ranges between 7.2 and 7.8, and the cation exchange capacities ranges from 39 to 57 m.e.q/100g. This soil exists in areas with elevations of 250m to 600m and 600m to 1100m and with slope range of 0% to 40 % (National soil map [25]).

• Map unit No. 25: This type of soil exists at 669m to 690m elevations and slope between 0 and 16 percent. The soil has a moderate to shallow depth with a fine particle size and limestone parent rock (National soil map [25]).

Another GIS data layer was created to show the 22 administrative divisions within the study area associated with the main villages existing within each division as can be seen in Figure 15 (after Mohawesh [24]).

Fig. 15. Wadi Zeglab administrative divisions


105

The land use shapefile (see Table 1 and Figure 16 (Mohawesh [24])) is one of the most important layers that were obtained. The study area consists mainly from forest, farms, range and urban as can be seen in the land use map.

Fig. 16. Land use of the study area

Another important layer that was produced is the roads network as can be seen in Figure 17. This layer, digitized from the topographic map, presents the road networks that cover the study area that can be classified mainly into main roads (divided/undivided) and secondary roads.


106

Fig. 17. Local road networks in the study area

4.3 Geospatial Analysis and GIS-based Multi-Criteria Site Selection

Based on the complete geospatial database collected in the previous section, the main objective of this section is identified in selecting the best area for a geo-environmental protected district or what’s known as ‘Geopark’ (UNESCO [32]). Geopark is usually classified into three types according to the covered area namely, mega, macro or micro Geopark. Wadi Zeglab as a whole with a total area of 107 Km2 can be considered as a mega GeoPark, but it is not possible to assign it as mega GeoPark because it includes certain land uses (e.g., urban areas) that need to be excluded from such classification. Therefore, some areas suitable for establishing a macro GeoPark need to be selected within Wadi Zeglab area that represent the most valuable resources in the Wadi (e.g., geological, natural, etc.) For these reasons the suggested Geopark location is best to be selected in forest and range land uses within Wadi Zeglab area because these areas include a rich biodiversity system (see Figure 18 (after Mohawesh [24])). So, since these forest and range areas are considered suitable locations for Macro-Geopark, this will be one of the selection criteria for the best candidate site.


107

Fig. 18. Suitable land uses for Geopark (from land use layer)

The rest of the selection criteria are identified based on our experience in the area and through consultation with some experts in the field and can be summarized as following (UNESCO [34] and Frey et al [35]):

A. Geopark shouldn’t be located at steep slopes areas, slopes between 0 to 25 degrees are recommended.

B. Geopark shouldn’t be so close to roads to avoid pollution resulted from vehicles, a distance of 50m to 100m from the highway is recommended.

C. Geopark should be little bit faraway from main streams to avoid flooding, being away of a distance of at least 30m is recommended.

D. Geopark should be located at forest and range land uses and away from farms and urban areas.

Collected data over which such criteria are implemented included: main and secondary roads layer, main Wadi streams and minor streams layer, contour map layer and land use layer.

The major step in the multi-criteria Geopark site selection is represented in identifying all unsuitable areas for Geopark (e.g., urban areas). After highlighting such areas in separate layers, overlaying for all of these unsuitable areas was done. Once unsuitable areas for Geopark in Wadi Zeglab are identified, the remaining areas represent potential locations for Geopark establishment. The following steps show clearly the detailed process of selecting the best candidate site for Geopark using mainly the layers of roads, elevation, streams, land use and soil:


108

• Based on the contour map, the digital elevation model (DEM) surface was developed as shown in Figure 19 along with the slope layer (see Figure 8) that was reclassified into nine intervals with classes six to nine being considered as unsuitable areas for Geopark as can be seen in Figure 20. Class six has slope range between 25˚ to 30˚, number seven between 30˚ to 36˚, number eight between 36˚ to 45˚ and number nine ranges between 45˚ and 66˚. All of these areas were excluded from being candidates for Geopark area as they don’t match the slope criterion.

Fig. 19. Digital Elevation Model for the study area.


109

Fig. 20. High slope region in the study area (> 25˚)

• The land use layer is grouped into 4 land uses: Forest, Range, Farms (tree crop, olive trees, orchard and irrigated area), and Urban area. The early mentioned land use criterion was weighted based on the fact that the Geopark should be away from urban area and farm area. Therefore, farms (Figure 21 (modified after Mohawesh [24])) and urban (Figure 22 (modified after Mohawesh [24])) layers are unsuitable for Geopark establishment and were excluded.


110

Fig. 21. Farms in the study area (from landuse layer)

Fig. 22. Urban area in the study area


111

• For roads layer, the criterion was defined as that Geopark should not be close enough to roads. Geopark needs to be away from main roads by about 100 m and from secondary roads by about 50 m. Therefore, buffer for the roads was developed to exclude the areas within these specified distances form roads as can be seen in Figure 23.

• Regarding the streams criterion, Geopark should be away from main streams by about 30 m and from minor streams by 5 m. Therefore a buffer for the Wadi was generated to exclude these areas from Geopark selection as can be seen from Figure 24.

Fig. 23. Buffer for the main and secondary roads

• All layers including the excluded areas based on the early defined criteria (high slope, farm layer, urban layer, road buffer layer, and Wadi buffer layer) were overlaid to identify all areas within Wadi Zeglab than are unsuitable for establishing the Geopark as can be seen in Figure 25. The available gaps among Figure 25 are suitable areas for Geopark but still need to be investigated further.


112

Fig. 24. Buffer for main and minor Wadi

From Figure 25, we conclude that there are seven potential areas that can be used as a Geopark highlighted in a separate shape file as can be seen in Figure 26. These seven possible areas for Geopark are located at lower, middle and upper part of the catchment area and stand in forest and range areas. All of these candidate sites satisfy the earlier defined criteria for Geopark selection but we still needs to select the most appropriate site. Table 3 summarizes the attributes associated with each one of the seven potential sites for Geopark establishment. These site attributes include: elevation range, location, area, type of climate zone, and accessibility to road network.


113

Fig. 25. All unsuitable areas for Geopark.

Fig. 26. Potential Geopark locations in Wadi Zeglab that satisfy the defined criteria.


114

Table 3. Attributes of the seven potential Geopark sites

Geopark Elevation (m) Accessibility to road network

Climate zone

Area (Km2)

Location

1 820m to 900 m Linked with roads highlands 2.06 Forest

2 350m to 420m Linked with roads highlands 0.44 Range

3 850m to 920m Linked with roads Ghor Zone 0.4 Forest and Range

4 800m to 880m Linked with roads highlands 1.9 Forest

5 250m to 310m Linked with roads highlands 1.42 Forest

6 490m to 650m Linked with roads highlands 1.48 Forest and Range

7 240m to 400m Linked with roads Ghor 0.58 Range

Based on the data presented in Table 3, Geoparks number one, four, five, and six are most likely the best candidates for the Geopark site mainly because they have adequate large area, they are located in highlands and Ghor zones’ climate with elevation reaches about 900 m above MSL, and they are linked with roads grids in addition to their location at forest and range area. According to these reasons, Geoparks number one (best option), four, five, and six are the most suitable areas for Macro Geopark that still needs further investigation to identify the best candidate among these four.

5. CONCLUDING REMARKS

The work presented in this paper emphasizes on the important role of GIS and multi-criteria analysis in land management using Wadi Zeglab in northern Jordan region as a case study. Wadi Zeglab area holds a unique and valuable geo-bio-diversity system that needs to be protected and highlighted on any future development plan for the area. For this reason, this paper aimed to select the best suitable locations for establishing a geo-environmental protected area ‘Geopark’ in Wadi Zeglab. Spatial data was collected from different sources and the rest of the data were produced using GIS analysis capabilities. Multi-criteria analysis was run on the GIS spatial layers using the predefined selection criteria to identify the potential candidates to establish the Geopark. Four potential locations satisfied the selection criteria and hence proposed as suitable sites for a Geopark. Future work is still needed to merge such sites with the masterplans and future development scenarios for the area and how this will affect the area future planning. Designing the necessary environmental standards to ensure the region sustainability is another target that needs to be covered in the future.

6. REFERENCES

[1] Abed, A., Moumani, K., and Ibrahim, K. (2001). Geology of Jordan, Jordanian Geologists Association, Amman.

[2] Al-Malabeh, A., Al-Oqlah, A., Wahib, M., and Bni-Yasine,T. (2007). Tourism Investment for Barqash and Al-Kura area and surroundings archaeology. Municipality of Al-Kura District. Unpublished Report, 11p. Hashemite University, Jordan.


115

[3] Al-Malabeh, A., and Al-Shreideh, A. (2004). Geotourism of Wadi Zeglab Water falls (Al-Kura/Jordan) and their conservation. 3rd Environmental Conference of the Jordanian Geologists Association.

[4] Al-Zu`bi, K. (1995). Surface Water Resources in Wadi Zeglab Catchment Area, M.Sc. Thesis, University of Jordan, Amman. (in Arabic).

[5] Bowerman, R., Hall, B., and Calamai, P. (1995). A multi-objective optimization approach to urban school bus routing: formulation and solution method. Transportation Research Part A: Policy and Practice,Vol. 29, pp. 107–123.

[6] Chakhar, S., and Martel, J-M. (2003). Enhancing geographical information systems capabilities with multi-criteria evaluation functions. Journal of Geographic Information and Decision Analysis,Vol. 7, pp. 47–71.

[7] Collins, M.G., Steiner, F.R., Rushman, M.J. (2001). Land-use suitability analysis in the United States: historical development and promising technological achievements. Environmental Management, Vol. 28 (5), pp. 611–621. [8] Field, J., and Banning, E. (1998). Hillslope Processes and Archaeology in Wadi Zeglab, Jordan. Geoarchaeology, An International Journal, Vol. 13 (6), pp. 595–616.

[9] Fisher, W. B. (1966). Soil Survey of Wadi Zeglab in Jordan, UN pub. report, University of Durham, UK.

[10] Hopkins, L. (1977). Methods for generating land suitability maps: a comparative evaluation. Journal of the American Planning Association, Vol. 43(4), pp. 386 – 400. [11] Jha, M. K., McCall, C. and Schonfeld, P. (2001). Using GIS, genetic algorithms, and visualization in highway development. Computer-Aided Civil and Infrastructure Engineering, Vol. 16, pp. 399–414.

[12] Joerin, F., Musy, A. (2000). Land management with GIS and multicriteria analysis. International Transactions in Operational Research,Vol. 7 (1), pp. 67–78.

[13] Joerin, F., The´riault, M., Musy, A. (2001). Using GIS and outranking multicriteria analysis for land-use suitability assessment. International Journal of Geographical Information Science, Vol. 15(2), pp. 153-174.

[14] Jordan Valley Authority (2008). The locate of rainfall station in Wadi Zeglab catchment.

[15] Leaõ, S., Bishop, I., and Evans, D. (2004). Spatial–temporal model for demand and allocation of waste landfills in growing urban regions. Computers, Environment and Urban Systems, Vol. 28, pp. 353–385.

[16] Laaribi, A., Chevallier, J.J. and Martel, J. M. (1996). A spatial decision aid: a multicriterion evaluation approach. Computers, Environment and Urban Systems,Vol. 20 (6), pp. 351–366.

[17] Malczewski, J. (1999). GIS and Multicriteria Decision Analysis New York: Wiley.


116

[18] Malczewski, J. (2004). GIS-based land-use suitability analysis: a critical overview. Progress in Planning, Vol. 62 (1), pp. 3–65.

[19] Malczewski, J. (2006). GIS-based multicriteria decision analysis: a survey of the literature. International Journal of Geographical Information Science, Vol. 20 (7), 249-268.

[20] Martin, N. J., St Onge, B. and Waaub, J. P. (1999). An integrated decision aid system for the development of Saint Charles River alluvial plain, Quebec, Canada. International Journal of Environment and Pollution,Vol. 12 (2/3), pp. 264–279.

[21] Meteorological Department, Jordan (2008). Climatic data for 1984 to 2003 and 2007. Amman-Jordan.

[22] Ministry of Water and Irrigation, Jordan (2008). Internal Report and Map, Jordan Valley Authority, Directorate of Geographic Information System, Amman-Jordan.

[23] Mohammad, B. (2000). Geological map no.3154-III of Ash Shuna Ash Shamaliyya, Jordan, Scale 1:50000, Geological map no.3154-IV of Dayr Abu Saìd, Jordan, Scale 1:50000, Geological map no.3154-IV of Irbid, Jordan, Scale 1:50000, and The geology of Irbid and Ash Shuna Ash Shamaliyya. Bulletin 46. Geological Mapping Division, The Natural Resources Authority, Jordan.

[24] Mohawesh, Y. (2009). Landuse and villages map. (Personal communication.) [25] National Soil Map and Land use Project, Jordan (1994). The Soil of Jordan. Level 2 Final Report. Soil Survey and Land Research Center. Ministry of Agriculture, Amman, Jordan.

[26] Noss, R. F., Carroll, C., Vance-Borland, K. and Wuerthner, G. (2002), A multicriteria assessment of the irreplaceability and vulnerability of sites in the Greater Yellowstone ecosystem. Conservation Biology,Vol. 16 (4), pp. 895–908.

[27] Nuafleh, K. (1995). Kinds of earth mass wasting in the slopping of Wadi Zeglab valley, M.Sc. Thesis, University of Jordan, Amman (in Arabic).

[28] Radaideh, E. (2006). Woodlands and Range Evaluation in Wadi Zeglab Catchment, M.Sc. Thesis, University of Jordan, Amman (in Arabic).

[29] Royal Jordanian Geographic Center (RJGC) (2008). Topographic Map for Al-Shuna Al-Shamalia, Irbid, Jerash and Der Abu Said. Amman-Jordan.

[30] Seppelt, R., and Voinov, A. (2002). Optimization methodology for land use patterns using spatially explicit landscape models. Ecological Modelling, Vol. 151 (2-3), pp. 125–142.

[31]Thill, J-C. (1999). Multicriteria Decision-making and Analysis: A Geographic Information Sciences Approach New York: Ashgate.

[32] Global UNESCO Network. Retrieved on October 10, 2012 from: http://www.europeangeoparks.org/?page_id=633

[33] Ziadat, F., Taylor, J., and Brewer, T. (2003). Merging Landsat TM imagery with topographic data to


117

aid soil mapping in the Badia region of Jordan. Journal of Arid Environments, Vol. 54 (3), pp. 527-541.

[34] UNESCO. (2008). Guidelines and Criteria for National Geoparks Seeking UNESCO’s Assistance to Join the Global Geoparks Network. Beijing: UNESCO.

[35] Frey, M-L, K. Schafer, G. Buchel, and M. Patzak (2006). Geoparks—A regional, European and global policy. In Geotourism: Sustainability, Impacts and Management, D. Newsome and R. Dowling, (eds). Oxford: Butterworth-Heinemann, pp. 95–117.

[36] International Society on Multiple Criteria Decision Making. Retrieved on October 10, 2012 from: www.mcdmsociety.org/

[37] Drobne, S., and Lisec, A. (2009). Multi-attribute Decision Analysis in GIS: Weighted Linear Combination and Ordered Weighted Averaging. Informatica 33, pp. 459-474.

Documents

Booklet Text.indd