1

Using EOF analysis over a large area for assessing the climate impact of small 1

scale afforestation in a semi-arid region 2

3

By: 4

G. Yosef1, P. Alpert2, C. Price2, E. Rotenberg3 and D. Yakir3 5

6 1 Department of Applied Mathematics, Israel Institute for Biological Research, 7

Ness-Ziona, Israel 8 2 Department of Geosciences, Tel Aviv University, 9

Tel Aviv, Israel 10 3 Department of Earth and Planetary Sciences, Weizmann Institute of Science, 11

Rehovot, Israel 12

13

14

Submitted to: Journal of Applied Meteorology and Climatology 15

16

17

(June 2017) 18

19

------------------------------------------------------------------------------------------------------ 20

Corresponding Author: Dr. Gil Yosef, Dept. of Applied Mathematics, Israel 21

Institute for Biological Research, Ness-Ziona, Israel. 22

e-mail: gily@iibr.gov.il 23

24

Prof. Pinhas Alpert and Prof. Colin Price, Dept. Geoscience, Tel Aviv University, 25

Tel Aviv, Israel, 69978. 26

e-mail: pinhas@cyclone.tau.ac.il, Tel: 972-3-6405722, Fax: 972-3-6405205 27

e-mail: colin@post.tau.ac.il, Tel: 972-3-6406029, Fax: 972-3-6409282 28

29

Prof. Dan Yakir and Dr. Eyal Rotenberg, Dept. of Earth and Planetary Sciences, 30

Weizmann Institute of Science, Israel, Rehovot, 76100. 31

e-mail: dan.yakir@weizmann.ac.il, Tel: 972-8-9342549 32

e-mail: eyal.rotenberg@weizmann.ac.il, Tel: 972-8-9344227 33

34

2

ABSTRACT 35

We suggest an approach to analyze the effects of small-scale afforestation on the 36

surrounding climate, of a large heterogenic area. While simple statistics have 37

difficulty identifying the effect, we use a well-known eigenvector technique in order 38

to overcome several specific challenges, due to a limited research region, complex 39

topography, and multiple atmospheric circulation patterns. We applied this approach 40

to investigate the influence of the isolated Yatir forest, at the north edge of Israel’s 41

Negev desert. It was found that this forest does influence the daily climate, primarily 42

seen in the main pattern of the Empirical Orthogonal Function (EOF) of temperature 43

and humidity. The EOF explains 93% and 80%, respectively, of the total variance in 44

the data. Although the Yatir forest is small, it is significant in regulating the climate in 45

the near-by surroundings, as it is located in a sharp transition area toward an arid 46

climate. A clear pattern of the effect of the dense vegetation in the research area (for 47

the surface air temperature) was observed in the third pattern EOF3 that explain 1-3%. 48

The results are presented as maps of correlation and regression between the 49

normalized PC time series of each pattern as well as other time series of the raw data 50

and spatially interpolated data stations. Analysis of short-term campaign 51

measurements around the Yatir forest supports the EOF results, and shows the forest’s 52

influence to the south, mainly during nighttime when the forest becomes cooler than 53

its surroundings. Overall, results suggest that in areas of transition to semi-arid 54

climates, forests regulate the surrounding surface air temperature and humidity fields. 55

Wind analysis based on complex EOF (CEOF) technique reveals the pattern of the 56

daily cycle of surface wind over the region. 57

3

1. Introduction 58

There is much interest in the impact of land-use change (LUC) on local climate 59

(Costa and Pires, 2010; Fall, et al., 2010a, 2010b; Brovkin, et al., 2013 and Deng, et al., 60

2013). Numerical simulations using complex models of the atmosphere have shown that the 61

presence or absence of vegetation could influence the local climate of the region in question, 62

e.g., Charney, et al., (1975); Shukla and Mintz (1982). Deforestation and overgrazing, as 63

studied in the Amazon forest and the Sahel Zone, decrease precipitation and daytime 64

evapotranspiration, and increase average daytime temperature, all due to decreased humidity 65

near the ground and changes in the energy balance, e.g., see Otterman (1974), Charney et al., 66

(1975), Idso (1980), Shukla et al., (1990), Shukla and Xie (1993) and Xue and Shukla 67

(1996). On the other hand, afforestation processes, studied in arid areas like the Sahel, 68

modify the energy balance and accelerate changes of circulation, leading to increased 69

precipitation (Ortiz et al., 2000; Patricola and Cook, 2007; Davin, et al., 2007; Bonan, 2008; 70

Cowling, et al., 2009). Studies in central/south Israel on the influence of LUC on climate 71

have found a decrease of the daytime temperature amplitude (Alpert and Mandel, 1986), 72

enhancement of convective rainfall due to increased air humidity, and the creation of 73

thermals that enhance convection (Otterman, et al., 1990, Ben-Gai, et al., 1993). 74

Previous studies on the climate impact of LUC over semi-arid areas suggest a tight 75

coupling between the atmosphere and land surface (Koster, et al., 2004). This coupling is 76

characterized by exceedingly complex nonlinear interactions with extensive degrees of 77

freedom (modes). Consequently, a challenging task in climate science is to find ways to 78

reduce the dimensions of the system and reveal the most important patterns behind the 79

variations (Thompson and Wallace, 1998). Numerical climate models (Grotch and 80

MacCracken, 1991; IPCC, 2007; Maraun et al., 2010) are widely used for analyzing small 81

areas. However, this technique is ineffective for revealing high resolution or fine-scale spatial 82

resolution of effects (Lluch-Cota, et al., 2003). In addition, the heterogenic distribution of 83

4

meteorological stations in such areas makes the analysis more complex. In order to reveal the 84

patterns over small research areas, eigentechniques can be applied to capture most of the 85

variance in the data in a small number of modes (Venegas, et al., 1997; Venegas, 2001). Each 86

of the modes is represented by its spatial pattern and time series, which are derived from the 87

eigenvalues and eigenvectors of the covariance matrix. Using the efficiency of the 88

eigentechniques, we investigated the impacts of LUC in Israel, caused by changes in land use 89

in the central-north Negev (Otterman, et al., 1990; Ben-Gai, et al., 1993; Sharon and Angert, 90

1998). This study focused on the Yatir afforestation project (Rotenberg and Yakir, 2010). 91

Based on LIDAR and ceilometers measurements at Yatir, Eder et al., (2015), showed that 92

there is a secondary air circulation created over the forest that likely changes surface air 93

temperature and relative humidity values over the forest compared with the surroundings thus 94

providing a mechanistic explanation of the connection between the two. 95

Based on studies mentioned above, we hypothesized that a relatively small-scale 96

afforestation area in a semi-arid region can influence the climate of the surrounding area and 97

the Yatir forest was chosen as a case study. We evaluate the forest's contribution to meso- 98

meteorological cycles, and the scope of the impact of the forest on the local climate. The 99

research was performed on the basis of data collected in the summer months, when there is a 100

semi-permanent synoptic circulation pattern that can be isolated, thus enabling us to study the 101

pattern generated by the forest itself. We used simple statistics on the stations measurements 102

in the area, and showed the difficulty in identifying clearly the influence of the dense 103

vegetation, in particular the small scale Yatir forest on large research area (Fig. 2). Therefore 104

we used the statistical analysis of eigentechniques in the following procedure; first, using 105

several interpolations on a synthetic database and interpolation schemes; second, directly on 106

the raw data stations, producing interpolated correlation and regression maps (see discussion 107

in the method section). 108

Eigentechniques were applied to temperature, humidity and wind observations on a 109

5

small scale with high temporal (1hr) and spatial (~2.5km) resolution. Section 2 outlines the 110

methodology that describes the summer climatology, dataset collection, generating of 111

synthetic maps by several interpolation methods, and briefly introduces the applied methods 112

of eigentechniques. Spatial and temporal patterns of temperature and humidity are presented 113

in section 3. Section 4 summarizes the conclusions. 114

115

2. Methodology 116

During the summer months, Israel is influenced by two global synoptic circulation 117

patterns: the Subtropical High that causes regional subsidence over the Eastern 118

Mediterranean with the associated marine inversion (Alpert, et al., 1990), and the Persian 119

trough that influences the region through dominance of the northwest winds (Yair and Ziv 120

1994). Subsidence of the subtropical air creates a stable atmosphere, leading to the creation 121

of thermal circulations, such as the nighttime land breeze and daytime sea breeze along the 122

Mediterranean coast, investigated by Doron and Neumann (1977); Segal, et al., (1985); 123

Lieman and Alpert (1993); Alpert and Rabinovich (2003). An additional circulation develops 124

between the coast and the inland mountains as a result of the topographic gradient (Fig. 1a): 125

Anabatic winds are created during the day and enhance the sea breeze. At night a katabatic 126

westward wind strengthens the land breeze. Another dominating circulation is the Red Sea 127

trough that generally creates eastern winds whose strength is modified by the land and sea 128

breezes. 129

The research area is a region considerably larger than the Yatir forest and was chosen 130

to emphasis the ability of the technique to identify the effects of a small forest on its 131

surroundings. This region includes various geographical features, such as mountain slopes, 132

the coastal plain area, varying land-use, etc. (Fig. 1b), and several patches of forest (indicated 133

by the ellipses in Fig. 1a). A larger forest is in the north, along the hills towards Jerusalem, 134

and the smaller Yatir forest in the south, at the edge of a rapid transition zone from semi-arid 135

6

to arid desert. The black box in Fig. 1a shows the research area between longitude 34.6!𝐸 136

and 35.22!𝐸 and between latitude 31.25!𝑁 and 31.85!𝑁. The research area is ~3920 km2. 137

We used meteorological station data provided by the Israel Meteorology Service 138

(IMS), data from the Yatir forest and three sites of campaign measurements around the Yatir 139

forest provided by the Department of Earth and Planetary Sciences at the Weizmann 140

Institute. The Yatir forest data collection is automatic, with half hour resolution, and 141

converted to hourly resolution to fit the IMS database. The measured parameters are surface 142

air temperature, relative humidity, and wind, at surface level for the months of August – 143

September, 2002-2005 in all the stations, with additional two stations in the north located in 144

and out of the forest (number 16 and 17) for the August –September, 2013-2016 (Table 1). 145

To support EOF analysis results, campaign measurements during summertime around Yatir 146

forest were used. Three sites were used, located northeast (c1), west (c2) and southwest of 147

the forest (c3) (Fig. 1b). Two consecutive weeks of hourly resolution at each site were 148

analyzed: in (c1) August-2015, (c2) Jun-2012, and (c3) August-2013. 149

The period August –September was chosen as in this period it is easier to observe and 150

isolate the regional circulation (the subtropical ridge creates a marine inversion, and the 151

Persian trough- creates western winds) and the local circulation (sea and land breeze, 152

anabatic and katabatic winds). This enabled us to isolate the daily anomalies due to processes 153

related to land coverage. 154

Limited meteorological stations and their heterogeneous scattering in complex terrain 155

present difficulties for spatial analysis, and for defining the spatial extent of climate data. 156

Therefore, spatial interpolation is utilized for building gridded maps. In order to provide 157

reliable meteorological fields (surface air temperature, relative humidity, and wind), we 158

compare two different interpolation approaches. The first technique was the combination of 159

Linear with the Nearest Neighbor interpolation methods (LNN) where the linear interpolation 160

uses two grid point values to define a value of a grid point in between, while the Nearest 161

7

Neighbor Interpolation considers only one grid point, closest to the interpolated point used, 162

due to a lack of data at the domain's edges. The second technique was the Kriging 163

interpolation (Kr), an advanced geo-statistical procedure based on models that include 164

autocorrelation, which weighs the contribution of the measured point according to statistical 165

relationships; for more details see Oliver and Webster (1990) and Cressie (2015). 166

Comparisons between LNN (Fig. S1) and Kr (Fig. 3a-c and 5a-c) show, in general, the same 167

spatial patterns for the surface air temperature and relative humidity variables, however they 168

showed different directions of influence on Yatir forest (see discussion in the results 169

section). We based the analysis on Kr interpolation procedure for 3 km grid spacing. 170

Simple statistical techniques (e.g. the spatial mean, anomaly and standard deviation) 171

were applied on the stations measurements. They show that for surface air temperature, the 172

mountain area is characterized with lower mean (Fig. 2a) and anomaly values (Fig. 2b), 173

while higher values are observed on the coastline and in the west slopes of the mountain. The 174

analysis of the relative humidity (Fig. 2d,e) shows that the coastline was observed with the 175

highest values, while Yatir forest and eastern slopes of mountain had the lowest values (for 176

further discussion physical process see the results section). The standard deviation (Fig. 2c) 177

show the effect of the forest in moderating the daily cycle of surface air temperature but over 178

a larger area, which is not obvious in the observation but confirmed with the EOF (Fig. 3). 179

The relative humidity (Fig. 2f) does not show any effect. These results emphasize the need 180

for a more complex statistical technique to identify a clear influence of the density 181

vegetation, in particular of the small scale Yatir forest. In the case of eigentechniques it gives 182

an explanation of the phenomenon, its significance and a unique time series for the pattern. 183

Overview of EOF and CEOF –The first technique used was the Empirical Orthogonal 184

Function (EOF). The technique is widely applied for the analysis of the spatial and temporal 185

variability of large multidimensional datasets and is commonly used in meteorological 186

studies (Lorenz, 1956). It can be used to identify and separate the dominant and the 187

8

underlying processes and essential parameters that in our case control the surface air 188

temperature and relative humidity patterns. Its main goal is to decompose the parameters of 189

the local climate region, and present a few patterns for each variable. As a result, we hope to 190

show the influence of the forest on the local climate. The following procedure of EOF was 191

performed separately for the surface air temperature and the relative humidity. The popular 192

configuration of the raw data matrix, in atmospheric research, is referred to as S-mode 193

analysis and represents the pattern's location (EOF) and the time series of the patterns called 194

Principal Component (PC). The matrix has the structure of 𝑀×𝑁, where 𝑀 is the space 195

dimensions and 𝑁 is the shared sampling dimension. This method was adopted by Halldor 196

and Venegas (1997) and von Storch and Zwiers (1999). EOF analysis is conducted on 197

datasets where the mean is removed from the measurements; the covariance matrix F can 198

then be constructed (Eq. 1) as 199

(1) 𝐹 = !!𝑅!𝑅 200

were 𝑅 is the raw data matrix, 𝑅! the transpose matrix and N is the number of samples. Then 201

reconstructing the raw data 𝑅 (Eq. 2) with a new time series PC and the eigenvectors (EOF); 202

(2) 𝑅 𝑡, 𝑥 = 𝑃𝐶 𝑡 𝐸𝑂𝐹(𝑥)!""#$%& . 203

The decomposed patterns are considered "empirical models" (EOF’s), which are orthogonal 204

to each other in space and in time (Mestas-Nuñez and Enfield, 1999). 205

The fraction of each k-th mode 𝐸𝑂𝐹! accounts for a variance (Eq. 3) is 206

(3) 𝜎! = 𝐿!/ 𝐿!!!!! ∗ 100 207

where L is the eigenvalue value of the 𝑘!! eigenvector 𝐸𝑂𝐹!. 208

The second technique was CEOF, an extension of EOF from a scalar field to vector 209

field, which was used for the wind analysis. Kundu and Allen (1976) introduced this 210

technique with the intention of studying ocean currents (Dominguez and Kumar 2005). The 211

9

CEOFs procedure is the same as EOF, except from the construction of the covariance matrix 212

𝐹!"#$%&' (Eq. 4) as 213

(4) 𝐹!"#$%&' =!!𝑅𝑅∗ 214

Where 𝑅∗ denotes the complex conjugate transpose, and the 𝐹!"#$%&' matrix is Hermitian, 215

which guarantees that the eigenvalues are real. 216

Output display - The eigentechnique results were displayed by PC's, which represent 217

normalized time series oscillations of the pattern. To find the location of the pattern in the 218

research region, EOF's are presented as correlation maps, resulting from the correlation 219

between the normalized PC time series of each pattern and all other time series of the raw 220

data (stations or interpolated gridded data). The study area is characterized by high 221

correlation; therefore the spectrum of the correlation index is limited in order to identify the 222

small biases. In addition the EOF’s are presented as a regression map to investigate the 223

spatial distribution of amplitude of daily cycle. 224

We used a sequence of methods first Kriging interpolation and second 225

eigentechniques (Kr-Eig). To verity that there was no “circular reasoning” of the statistical 226

relationship, the eigentechniques were also applied directly on the stations raw data and then 227

applying Kriging interpolation method (Eig-Kr) (Ha, H et al., 2014; Nazzal, Y et al., 2015). 228

The MODIS (MCD12Q1) (LA DAAC, 2003) land cover was projected to the maps, 229

showing only the relatively dense vegetation class in the research area (woody and closed 230

shrubland). 231

Analysis procedure - In order to show the forest influence, two sets of EOF analyses 232

were done; the first serves as control (CON), and includes all the meteorological stations 233

except the Yatir forest and the second analysis (YF) includes the Yatir forest station. Then 234

the difference between YF and CON analyses was calculated, to show the direct influence of 235

the Yatir forest on the spatial climate and also the fraction of the total variance. 236

10

To support the results of the EOFs analysis, the daily average differences map 237

between Yatir forest and the near surrounding meteorological stations was calculated for the 238

surface air temperature and relative humidity (in Fig. 3c and Fig. 5c the dashed line represent 239

the analyzed area). Three of the IMS meteorological stations (Arad, Beer-Sheva and Shany) 240

together with three campaigns measurements (c1,c2, and c3) were compared to the Yatir 241

forest station, and used to calculate the differences over hourly an time step for two 242

consecutive weeks. Although the measurements were taken in different years, they represent 243

the same summertime season behavior of this region. Also no synoptic trend was observed 244

between those years. We assume homogeneity over the Yatir forest. Therefore, to calculate 245

the map differences, six points with zero values were used to cover the location of the Yatir 246

forest area, according to MODIS’s index of relatively dense vegetation (Fig. 7). 247

248

3. Results 249

Comparisons between LNN and Kr interpolation procedures - The main biases 250

between the methods are located around the edges of the research area. We expect that the 251

LNN is less accurate along the edges as compared to the Kr method, since it uses the Nearest 252

Neighborhood (NN) interpolation, which considers only one grid point as a reference. 253

Another main bias was found in size and direction of the forest influence, measured by the 254

differences between YF and the CON analysis (Fig. S1c,f, Fig.3c and Fig. 5c), which shows, 255

a larger area at the Kr interpolation and inconsistent direction of influence with the LNN. The 256

LNN interpolation show the impact of forest area, which extends northwest of the forest, 257

while Kr interpolation is consistent with observation (Fig. 7) and shows influence south of 258

the forest. 259

Temperature analysis - For the procedure Kr-Eig (Eig-Kr), the main pattern of the 260

surface air temperature field EOF1 at CON and YF analyses account for 94% (86%) and 261

93% (86%) of the variance respectively (Fig. 3 and Fig. 6). The pattern time series PC1 (Fig. 262

11

3g) of CON and YF, are intertwined with each other, where the local maxima are obtained at 263

around noon and the local minima around midnight. This cycle is results of the energy 264

balance due to daily solar radiation, with highest intensity every 24hr. The higher mode 265

EOF3 account for the 1% (3%) of the variance for the procedure Kr-Eig (Eig-Kr). Although 266

it has a low value, it shows clear pattern of the forests influence on the region (Fig. 4). 267

Humidity analysis - For the procedure Kr-Eig The main pattern of the relative 268

humidity field EOF1 at CON and YF analyses accounts for 81% and 80% of the variance, 269

respectively (Fig. 5a-b). As in the surface air temperature, the pattern time series PC1 (Fig. 270

5g) of CON and YF are intertwined with each other. The daytime cycle reaches variance 271

local maximum peak after midnight, until sunrise, when temperatures are lowest, and a local 272

minimum humidity point around noon. 273

Underlying features in the main pattern – For the procedure Kr-Eig, the surface air 274

temperature and relative humidity fields show in the correlation maps of EOF1 (Fig. 3a-b, 275

and Fig. 5a-b) for both of the CON and YF analysis that most of the research area has a high 276

correlation with the PC1, following the 24hr cycle of the daily solar radiation. Within the 277

area of high positive correlation, underlying features at four different locations are revealed: 278

coastline, mountain slopes, mountain, and the forests areas. Where the mountain slope has 279

the heights values compare to the other areas. The regression maps show the same features 280

and the same hierarchy of values as correlation maps, except for the relative humidity 281

variable that has a maximum over part of the mountain area compared to the correlation 282

maps that have maximum values in the mountain slopes. 283

The results shows that the mountain slopes with the height correlation and linear 284

regression values have the highest daily amplitude range of surface air temperature (see 285

stations comparison in Fig. S2). The daily amplitude range at the coastline area is affected by 286

the higher humidity that acts as greenhouse effect and moderates the minimum temperature 287

at night and increases the relative humidity. Around the midday the sea breeze dominates 288

12

with a northwest wind direction, decreasing the maximum temperature and keeping the 289

highest relative humidity. In the mountain area, clear nights and lower humidity (as a 290

function of distance from the sea) lead to greater upward thermal radiation, which in turn 291

leads to lower minimum surface air temperature and the lowest relative humidity. During the 292

day the maximum surface air temperature is lowest as compared to other areas, as a result of 293

the cooling sea breeze (Fig. 8a,b) that flows at a higher altitude, so less affected by the 294

warming surface. In the south, Yatir forest is located in the Negev Desert influenced by an 295

arid regional climate, as in Beer Sheva (point 6) and Arad (point 7). However, one should 296

note that in reality, Yatir forest (point 8) is distinct from its surrounding region with a low 297

correlation and linear regression in relation to its near surroundings. 298

The third pattern of the surface air temperature (Fig. 4) represents a time series of the 299

dense vegetation areas. EOF3 shows mainly positive correlation and linear regression in the 300

Yatir forest and its surroundings. The factor causing this pattern is vegetation, by moderating 301

between the day and night, and therefore regulating the arid climate from its immediate 302

surroundings. Due to its size and its regulating capacity, the clear pattern of the forests 303

appears only in the third pattern, which represents small deviations from the average. The 304

pattern reaches a maximum variance around noon, and minimum variance at night hours, 305

which means that the forest creates temperature regulation during the hot hours of the day. 306

The differences between YF and CON analyses both in surface air temperature and 307

relative humidity show the impact of forest area, which extends southwest of the forest (Fig. 308

3c,f and 5c,f). The adjacent station Shany (point 9) at the north of the forest determines the 309

north boundary of the forest’s influence. The same shape of the forest’s influence structure 310

was observed at the daily average differences map, where the forest creates an enclave 311

surrounded with small differences of surface air temperature (Fig. 7b) and relative humidity 312

(Fig. 7d) that stretches south of the forest. There are differences between the two analyses, in 313

the forest's influence extending to the southwest in the EOF analysis as compared to the 314

13

southeast from analyses of difference maps. This can be related to the results of the new 315

campaign sites measurements west of the forest. 316

In general around the enclave to the south, a higher temperature and lower relative 317

humidity is observed. In regarding to relative humidity, the enclave is smaller and close to 318

the forest size area, meaning the forest’s influence is more prominent in surface air 319

temperature than the relative humidity. As the distance from the forest toward the southeast 320

increases, the surface air temperature and relative humidity increase and decrease, 321

respectively. This relates to the transition of climate from semi-arid to arid conditions. The 322

shapes of the daily average differences maps are mainly affected by the night hours when 323

surface air temperature around the forest is lower than the surroundings and relative humidity 324

is higher (Fig. 7a,b). The rest of the day is influenced also from the direction and intensity of 325

the wind that acts on the surface air temperature and relative humidity, which limits the 326

forest's influence. 327

Intuitively one would think that the forest’s surface air temperature at night should be 328

higher, as the net radiation is higher over the forest than the bare soil, leading to more 329

observed energy. In contrast, Rotenberg and Yakir (2010) presented a study on the radiation 330

budget of a forest in a semi-arid area. They show that in spite of increased radiation 331

absorption due to the forest low albedo, and small latent heat flux due to lack of water, such 332

forests develop large sensible heat fluxes, which in turn results in decreasing surface 333

temperature and reduced thermal radiation emission. 334

For the procedure Eig-Kr the same underlying features (mentioned above) are 335

observed, where the mountain slopes have the highest values in the correlation and regression 336

maps, compared to the coastline and forest areas (Fig. 6). Compared to the Kr-Eig procedure, 337

Yatir forest is much more prominent in the difference between YF and CON. 338

Wind analysis – The two main patterns of the wind field account for almost 85% of 339

the variance. The first pattern CEOF1 accounts for 73% of the variance and shows the pattern 340

14

of daily average wind vectors, dominated by the intense western sea breeze. The sea breeze is 341

well developed around noon. Toward evening, it a clearly shows high dominant U wind 342

component (east-west), while the V wind component (north-south) is hardly noticeable (Fig. 343

8c). The second pattern CEOF2 accounts for 11% of the variance (Fig. 8d), shows wind 344

directions that are observed after mid-night until the early morning. Two distinct areas are 345

shown in CEOF2: the mountain area with western component, and the remaining area 346

including the Yatir forest with weak southeast components, which in the observations 347

appears as easterly winds. The analysis of the results is accompanied by a comparison to 348

those in Segal's article (1985), which evaluated the wind flows during summer in southern 349

Israel. The results show no influence of the Yatir forest on the creation of horizontal local 350

wind patterns on the scale of the research area. Although we concentrate on the 351

climatological summer season and not the daily cycle, it is important to note that the 352

direction and strength of the wind at different times of the day can change the area influenced 353

by the Yatir forest as was discussed in the previous section. 354

355

4. Conclusions 356

This study attempts to find the effect of small scale dense vegetation such as Yatir 357

forest on the surrounding local climate, in large area, using a small number of meteorological 358

stations. It was shown that merely using simple statistics makes it hard to identify, from a 359

large heterogenic area, a clear impact of the small-scale dense vegetation. Therefore there is 360

need for a more complex statistical procedure, such as the eigentechniques, that can reveal 361

the forest time series pattern and use it to estimate the spatial influence. Regarding the 362

primary EOF patterns of surface air temperature and relative humidity, it is seen that they are 363

all influenced by the diurnal cycle, so that the approximated maximum or minimum values of 364

the temperature patterns are obtained around 12:00 and 24:00 local time. We found that in 365

cases of different interpolation procedures, LNN and Kr, on long observation periods, the 366

15

Yatir forest creates considerable differences in the local climate and its surroundings in both 367

temperature and humidity amplitude regulation. For the procedure Kr-Eig, the main pattern 368

EOF1 (YF analysis) of the surface air temperature and relative humidity fields accounts for 369

93% and 80% of the variance, respectively. The main patterns in both fields show spatially 370

high correlation with PC1 with underlying features at four different locations. Except for the 371

mountain slopes that have the highest correlation, the coastline, mountain, and Yatir forest 372

areas have lower correlations, thus a lower daily amplitude range. The same results were 373

obtained in the analysis of the regression map, which indicates a strong linear relationship of 374

daily amplitude maximum range in the mountain area compared to coastline and north forest, 375

and Yatir in the south. 376

To verify that there was no “circular reasoning”, the procedures Kr-Eig vs. Eig-Kr 377

were examined and showed that there was no significant difference between the results, 378

where the same underlying features of the first pattern were highlighted in both correlation 379

and regression maps. Yatir forest is prominent in the difference between YF and CON using 380

Eig-Kr. 381

The climate at the region is characterized with sea breeze during the day that 382

moderates the maximum surface air temperature at the coastline, while in the mountain area 383

it has an important role in reducing the maximum temperature as it is less affected by the 384

warm surface. At night, surface air temperature at the coastline is moderated by the humidity 385

that acts as a greenhouse effect. In the Yatir forest, the development of large sensible heat 386

fluxes decreases surface air temperature and reduces thermal radiation emission. The location 387

of the forest is in a transition area from semi-arid to arid conditions, and differentiates itself 388

by its source of moisture. Those effects of lower surface temperature and higher relative 389

humidity become stronger at night when there is no dominant wind, such as the sea breeze, to 390

limit the forest influence. 391

16

The analysis of the differences between YF and CON in the two procedures Kr-Eig 392

and Eig-Kr, yields the area of the Yatir forest footprint (i.e., the forest’s influence), both in 393

surface air temperature and relative humidity. It shows the impact of the forest area that 394

extends southwest of the forest. A similar shape was observed for the daily average 395

differences map based on campaign measurements around the forest, where the forest creates 396

an enclave surrounded with small differences of surface air temperature and relative humidity 397

The main pattern of wind field CEOF1 and CEOF2 accounts for almost 85% of the 398

variance, and shows the development of the sea breeze and land breeze, respectively. The 399

Yatir forest does not show influence on the creation of horizontal local wind patterns on the 400

scale of the research area. 401

The general conclusion is that the relatively dense vegetation is the basis for a local 402

climate change; for example 5 km away, in the southern region of the forest center, the day is 403

cooler by 0.5-1 deg. The expected dry region afforestation may affect the surroundings. 404

Hence, if the Yatir forest expands, so will its influence, assuming that the climate influence is 405

cumulative. For a higher sensitivity study and in order to quantify the region of influence and 406

extent of the changes, there is a need for a larger number of observations in a more 407

homogeneous regime, with further numerical simulations. 408

409

410

Acknowledgment. We wish to thank the Israel Meteorological Service for providing the data 411

of the meteorological stations. The study was partially funded by the German Helmholtz 412

Association within the DESERVE project. We also acknowledge Dr Yehudah Alexander for 413

fruitful discussions regarding this work. 414

415

416

417

17

REFERENCES 418

Alpert, P., and Mandel, M, 1986: Wind variability-An indicator for a mesoclimatic 419

change in Israel. J. Climate Appl Meteor, 25(11), 1568-1576. 420

Alpert, P., Abramsky, R., and Neeman, B. U., 1990: The prevailing summer synoptic 421

system in Israel―Subtropical high, not Persian Trough. Israel journal of 422

earth-sciences, 39(2/4), 93-102. 423

Alpert, P. and Rabinovich-Hadar, M., 2003: Pre-and Post-Sea-Breeze Frontal Lines-A 424

Meso-γ-Scale Analysis over South Israel. Journal of the atmospheric 425

sciences, 60(24), 2994-3008. 426

Ben-Gai, T., Bitan, A., Manes, A., and Alpert, P., 1993: Long-term change in 427

October rainfall patterns in southern Israel. Theoretical and Applied 428

Climatology, 46(4), 209-217. 429

Bonan, G.B., 2008: Forests and climate change: forcings, feedbacks, and the climate 430

benefits of forests. science, 320(5882),1444-1449. 431

Brovkin, V., Boysen, L., Raddatz, T., Gayler, V., Loew, A. and Claussen, M., 2013: 432

Evaluation of vegetation cover and land-surface albedo in MPI-ESM CMIP5 433

simulations. Journal of Advances in Modeling Earth Systems, 5(1), 48-57. 434

Charney, J.G. 1975: Dynamics of deserts and drought in the Sahel. Quart. J. Roy. 435

Meteor. Soc, 101, 193-202. 436

Chamey, J., Stone, P.H. and Quirk, W.J., 1975: Drought in the Sahara: a 437

biogeophysical feedback mechanism. Science, 187, 434-435. 438

Cook, K.H., 1999: Generation of the African easterly jet and its role in determining 439

West African precipitation. Journal of climate, 12(5), 1165-1184. 440

Costa, M. H. and Pires, G. F., 2010: Effects of Amazon and Central Brazil 441

deforestation scenarios on the duration of the dry season in the arc of 442

18

deforestation. International Journal of Climatology, 30(13), 1070-1979. 443

Cowling, S.A., Jones, C.D. and Cox, P.M., 2009: Greening the terrestrial biosphere: 444

simulated feedbacks on atmospheric heat and energy circulation. Climate 445

dynamics, 32(2-3), 287-299. 446

Cressie, N. 2015: Statistics for spatial data. Wiley & Sons. 447

Davin, E.L., de Noblet-Ducoudré, N. and Friedlingstein, P., 2007: Impact of land 448

cover change on surface climate: Relevance of the radiative forcing 449

concept. Geophysical Research Letters, 34(13). 450

Deng, X., Zhao, C. and Yan, H., 2013: Systematic modeling of impacts of land use 451

and land cover changes on regional climate: a review. Advances in 452

Meteorology, 2013. 453

Dominguez, F. and Kumar, P., 2005: Dominant modes of moisture flux anomalies 454

over North America, Journal of Hydrometeorology, 6, 194-206. 455

Doron, E., J. Neumann., 1977: Land and mountain breezes with special attention 456

to Israel's mediterranean coastal plain, Israel Meteor. Res. Papers, 1, 109-122. 457

Eder, F., De Roo, F., Rotenberg, E., Yakir, D., Schmid, H.P. and Mauder, M., 2015. 458

Secondary circulations at a solitary forest surrounded by semi-arid shrubland and 459

their impact on eddy-covariance measurements. Agricultural and Forest 460

Meteorology, 211, 115-127. 461

Fall, S., Niyogi, D., Gluhovsky, A., Pielke, R.A., Kalnay, E. and Rochon, G., 2010a: 462

Impacts of land use land cover on temperature trends over the continental 463

United States: assessment using the North American Regional 464

Reanalysis. International Journal of Climatology, 30(13), 1980-1993. 465

Fall, S., Diffenbaugh, N.S., Niyogi, D., Pielke, R.A. and Rochon, G., 2010b: 466

Temperature and equivalent temperature over the United States (1979– 467

19

2005). International Journal of Climatology, 30(13), 2045-2054. 468

Grotch, S.L. and MacCracken, M.C., 1991. The use of general circulation models to 469

predict regional climatic change. Journal of Climate, 4(3), 286-303. 470

Halldor, B. and Venegas, S.A., 1997. A manual for EOF and SVD analyses of climate 471

data. McGill University, CCGCR Report, (97-1), p.52. 472

Ha, H., Olson, J.R., Bian, L. and Rogerson, P.A., 2014: Analysis of heavy metal sources in 473

soil using kriging interpolation on principal components. Environmental science & 474

technology, 48(9), pp.4999-5007. 475

Idso, S. B. 1980: Surface energy balance and genesis of deserts, Arch. Meteor. 476

Geophys. Bioklim, A30, 253-260. 477

IPCC: Climate Change 2007: The Physical Science Basis, edited by: Solomon, S., 478

Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tignor, M., and 479

Miller, H. L., Cambridge University Press, Cambridge, UK, and New York, USA, 480

2007. 481

Koster, R.D., Dirmeyer, P.A., Guo, Z., Bonan, G., Chan, E., Cox, P., Gordon, C.T., 482

Kanae, S., Kowalczyk, E., Lawrence, D. and Liu, P., 2004: Regions of strong 483

coupling between soil moisture and precipitation. Science, 305(5687), 1138-1140. 484

Kundu, P. K. and Allen, J. S., 1976: Some three-dimensional characteristics of low- 485

frequency current fluctuations near the Oregoncoast. J. Phys. Oceanogr, 6, 181–199. 486

Land Processes Distributed Active Archive Center (LP DAAC), 2003, - Land Cover 487

Type Yearly L3 Global 0.05Deg CMG MCD12C1, USGS Earth Resources 488

Observation and Science (EROS) Center, Sioux Falls, South Dakota 489

(https://lpdaac.usgs.gov), accessed January 1, 2012. 490

Lieman, R. and Alpert, P., 1993: Investigation of the planetary boundary layer height 491

variations over complex terrain. Boundary-Layer Meteorology, 62(1-4), 492

129-142. 493

20

Lluch-Cota, D.B., Wooster, W.S., Hare, S.R., Lluch-Belda, D. and Parés-Sierra, A., 494

2003: Principal modes and related frequencies of sea surface temperature 495

variability in the Pacific coast of North America. Journal of 496

Oceanography, 59(4), 477-488. 497

Lorenz, E. N: Empirical Orthogonal Functions and Statistical Weather Prediction, 498

Departemnt of Meteorology, MIT, 49 pp, 1956. 499

Maraun, D., Wetterhall, F., Ireson, A.M., Chandler, R.E., Kendon, E.J., Widmann, 500

M., Brienen, S., Rust, H.W., Sauter, T., Themeßl, M. and Venema, V.K.C., 501

2010: Precipitation downscaling under climate change: Recent developments 502

to bridge the gap between dynamical models and the end user. Reviews of 503

Geophysics, 48(3). 504

Mestas-Nuñez, A.M. and Enfield, D.B., 1999: Rotated global modes of non-ENSO 505

sea surface temperature variability. Journal of Climate, 12(9), 2734-2746. 506

Nazzal, Y., Zaidi, F.K., Ahmed, I., Ghrefat, H., Naeem, M., Al-Arifi, N.S., Al-Shaltoni, 507

S.A. and Al-Kahtany, K.M., 2015: The combination of principal component 508

analysis and geostatistics as a technique in assessment of groUndWater 509

hydrochemistry in arid environment. Current Science, 108(6). 510

Oliver, M.A. and Webster, R., 1990: Kriging: a method of interpolation for 511

geographical information systems. International Journal of Geographical 512

Information System, 4(3), 313-332. 513

Ortiz, J., Guilderson, T., Adkins, J., Sarnthein, M., Baker, L. and Yarusinsky, M., 514

2000: Abrupt onset and termination of the African Humid Period:: rapid 515

climate responses to gradual insolation forcing. Quaternary science 516

reviews, 19(1), 347-361. 517

Otterman, J., 1974: Baring high-albedo soils by overgrazing- A hypothesized 518

desertification mechanism. Science, 186(4163), 531-533. 519

21

Otterman, J., Manes, A., Rubin, S., Alpert, P. and Starr, D.C., 1990: An increase of 520

early rains in southern Israel following land-use change?.Boundary-Layer 521

Meteorology, 53(4), 333-351. 522

Patricola, C.M. and Cook, K.H., 2007: Dynamics of the West African monsoon under 523

mid-Holocene precessional forcing: regional climate model simulations. Journal of 524

Climate, 20(4), 694-716. 525

Preisendorfer, R. W. 1988: Principal Component Analysis in Meteorology and 526

Oceanography. Development in Atmospheric Science, 17. 527

Rotenberg, E. and Yakir, D., 2010: Contribution of semi-arid forests to the climate 528

system. Science, 327(5964), 451-454. 529

Segal, M., Mahrer, Y., Pielke, R.A., and Kessler, R.C. 1985: Model evaluation of the 530

summer daytime induced flows over southern Israel. Israel Journal Earth Science, 531

34, 39-46. 532

Sharon. D, and A. Angert. 1998: Long-Term changes in Northern Negev rains from a 533

regional perspective. Met in Israel 5, 38-50 (In Hebrew) 534

Shukla, J. and Mintz, Y., 1982: Influence of land-surface evapotranspiration on the 535

earth's climate. Science, 215(4539), 1498-1501. 536

Shukla, J., Nobre, C. and Sellers, P., 1990: Amazon deforestation and climate 537

change. Science(Washington), 247(4948), 1322-1325. 538

Shukla, J. and Xie, Y., 1993: A numerical experiment to study the 539

influence of change in the land surface properties on Sahel climate. 540

Part 1: Desertification. J. Climate, 6, 2232-2245. 541

Thompson, D.W. and Wallace, J.M., 1998: The Arctic Oscillation signature in the 542

wintertime geopotential height and temperature fields. Geophysical research 543

letters, 25(9), 1297-1300. 544

Venegas, S.A., Mysak, L.A. and Straub, D.N., 1997. Atmosphere-ocean coupled 545

22

variability in the South Atlantic. Journal of Climate, 10(11), 2904-2920. 546

Venegas, S.A., 2001: Statistical methods for signal detection in climate. Danish 547

Center for Earth System Science Report, 2, 96. 548

Von Storch, H. and Zwiers, F.W., 1999: Statistical analysis in climate research. 549

Cambridge university press. 550

Williams, C.R., 1997: Principal component analysis of wind profiler 551

observations. Journal of Atmospheric and Oceanic Technology, 14(3), 386-395. 552

Willmott, C.J. and Matsuura, K., 1995: Smart interpolation of annually averaged air t 553

emperature in the United States. Journal of Applied Meteorology, 34(12), 2577-2586. 554

Xue, Y. and Shukla, J., 1996: The influence of land surface properties on Sahel 555

climate. Part II. Afforestation. Journal of Climate, 9(12), 3260-3275. 556

Yair, Y. and Ziv, B., 1994: An introduction to Meteorology. The Open University, 557

Tel-Aviv, Israel. 558

559

560

561

562

563

564

565

566

567

568

569

570

571

23

Figures captions: 572

Fig. 1 a. The box represents the research area in southern Israel, the black ellipses indicate 573

two main patches of forest in the area. b. Topography height in the research area base on 574

digital elevation model (DEM) database. Black dots indicate meteorological stations. 575

Contour and color scale represent the topography height. Units in [m] 576

Fig. 2 Spatial interpolation using Kriging (Kr) method of the meteorological stations (black 577

dots); surface air temperature (a) Mean, (b) spatial anomaly and (c) spatial standard deviation 578

and Relative humidity (d-f). The grid maps were prepared with the Kr procedure, for 3 km 579

grid spacing. Topography (contours) and vegetation (gray). Contours and colorbar units in 580

[m] and [0C] respectively. The same for relative humidity (d-f). 581

Fig. 3 The first EOF analysis of surface air temperature for the procedure starts with Kriging 582

spatial interpolation to the stations database and then applying Eigentechniques (Kr-Eig). The 583

value at each grid point is the correlation (Corr) and regression (Reg) coefficient between the 584

time series of the principal component and the surface air temperature at that point (with the 585

mean removed, see method section). The interpolation method, coefficient and the fractions 586

of variance explained by the respective modes are indicated in the parenthesis at the title of 587

each map. The correlation maps (a) spatial pattern of the control without the station of Yatir 588

forest measurements (CON), (b) Yatir forest (YF), (c) bias between (YF) control (CON). The 589

same for the regression maps (d-f), with scale multiplied by 100. (g) The first normalized 590

time series (PC) (the first 5 days are presented) of YF and CON. In the EOF maps, black dots 591

indicate the meteorological stations, topography in contours and vegetation in gray. Contours 592

units in [m], and colorbar represents correlation index. 593

Fig. 4 The same as Fig. 3, for EOF3. 594

Fig. 5 The same as Fig. 3, for relative humidity analysis. 595

24

Fig. 6 The same as Fig. 3 for surface air temperature with the procedure starting with 596

applying Eigentechniques to the stations database and then using Kriging for spatial 597

interpolation (Eig-Kr). 598

Fig. 7 Daily differences between forest and surroundings: “out” minus “in” of the Yatir 599

forest (YF), (a-b) surface air temperature (c-d) and relative humidity. In the map (b and d): 600

blue dots set with zero values represent the forest location (gray) based on satellite products 601

of Modis. Black dots indicate the IMS meteorological stations and circled dots the campaigns 602

measurements (c1, c2, and c3). The grid maps are preformed with the Kr procedure, for 3 km 603

grid spacing. Contours (topography) units in [m], and colorbar represents the difference. 604

Fig. 8 Wind analysis: Hourly average value of wind direction and speed (a) Yatir and (b) 605

Jerusalem. The main CEOFs analysis of wind: (a) CEOF1 and (b) CEOF2. The value at each 606

grid point is the complex eigenvectors. In the maps black dots indicate the meteorological 607

stations, topography in contours. Contours units in [m]. 608

609

610

611

612

613

614

615

616

617

618

619

620

621

25

Table 1: Location of meteorological stations by Lat(N)/Lon(E) and altitude above mean sea 622

level. 623

624

625 626

Num Station Lat Lon Altitude (m)1 Revadim 31.77 34.82 902 Ashdod 31.83 34.67 103 Azrikam 31.75 34.7 304 Negba 31.65 34.67 905 Lachish 31.6 34.78 1256 Beer-Sheva 31.25 34.8 2807 Arad 31.25 35.18 5688 Yatir 31.33 35.03 6509 Shany 31.35 35.05 69510 Jerusalem 31.77 35.22 81511 Rosh-Tzurim 31.67 35.12 95512 Netiv H L H 31.68 34.97 28513 Bet-Gimal 31.72 34.98 36014 Yavne 31.82 34.72 5015 Besor 31.27 34.38 10616 Tzova 31.78 35.12 73017 Mevo Horon 31.85 35.02 195c1 Northeast to YF 31.37 35.02 575c2 East to YF 31.33 34.99 522c3 Southeast to YF 31.32 34.98 460

26

627

Fig. 1 a. The box represents the research area in southern Israel, the black ellipses indicate 628

two main patches of forest in the area. b. Topography height in the research area base on 629

digital elevation model (DEM) database. Black dots indicate meteorological stations. 630

Contour and color scale represent the topography height. Units in [m] 631

632

633

634

635

636

637

638

27

639

640

641 642

Fig. 2 Spatial interpolation using Kriging (Kr) method of the meteorological stations (black 643

dots); surface air temperature (a) Mean, (b) spatial anomaly and (c) spatial standard deviation 644

and Relative humidity (d-f). The grid maps were prepared with the Kr procedure, for 3 km 645

grid spacing. Topography (contours) and vegetation (gray). Contours and colorbar units in 646

[m] and [0C] respectively. The same for relative humidity (d-f). 647

648

649

650

a.

b.

c.

d.

e.

f.

28

651

652

653 654

Fig. 3 The first EOF analysis of surface air temperature for the procedure starts with Kriging 655

spatial interpolation to the stations database and then applying Eigentechniques (Kr-Eig). The 656

value at each grid point is the correlation (Corr) and regression (Reg) coefficient between the 657

time series of the principal component and the surface air temperature at that point (with the 658

mean removed, see method section). The interpolation method, coefficient and the fractions 659

of variance explained by the respective modes are indicated in the parenthesis at the title of 660

each map. The correlation maps (a) spatial pattern of the control without the station of Yatir 661

forest measurements (CON), (b) Yatir forest (YF), (c) bias between (YF) control (CON). The 662

same for the regression maps (d-f), with scale multiplied by 100. (g) The first normalized 663

time series (PC) (the first 5 days are presented) of YF and CON. In the EOF maps, black dots 664

indicate the meteorological stations, topography in contours and vegetation in gray. Contours 665

units in [m], and colorbar represents correlation index. 666

667

668 669 670 671 672

a. b. c.

d. e. f.

g.

29

673

674

675 676

Fig. 4 The same as Fig. 3, for EOF3 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703

a. b. c.

d. e. f.

g.

30

704

705

706 707

Fig. 5 The same as Fig. 3, for relative humidity analysis. 708

709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733

a. b. c.

d. e. f.

g.

31

734

735

736 737

Fig. 6 The same as Fig. 3 for surface air temperature with the procedure starting with 738

applying Eigentechniques to the stations database and then using Kriging for spatial 739

interpolation (Eig-Kr). 740

741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762

a. b. c.

d. e. f.

g.

32

763

764

Fig. 7 Daily differences between forest and surroundings: “out” minus “in” of the Yatir 765

forest (YF), (a-b) surface air temperature (c-d) and relative humidity. In the map (b and d): 766

blue dots set with zero values represent the forest location (gray) based on satellite products 767

of Modis. Black dots indicate the IMS meteorological stations and circled dots the campaigns 768

measurements (c1, c2, and c3). The grid maps are preformed with the Kr procedure, for 3 km 769

grid spacing. Contours (topography) units in [m], and colorbar represents the difference. 770

771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789

a.

c.

b.

d.

33

790

791

Fig. 8 Wind analysis: Hourly average value of wind direction and speed (a) Yatir and (b) 792

Jerusalem. The main CEOFs analysis of wind: (a) CEOF1 and (b) CEOF2. The value at each 793

grid point is the complex eigenvectors. In the maps black dots indicate the meteorological 794

stations, topography in contours. Contours units in [m]. 795

796

c.

d.

a.

b.