54
European scale AQ mapping European scale AQ mapping (using interpolation and (using interpolation and assimilation) and evaluation assimilation) and evaluation of its uncertainty of its uncertainty Jan Horálek Jan Horálek , Pavel Kurfürst , Pavel Kurfürst Peter de Smet Peter de Smet ETC/ACC ETC/ACC

European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

  • Upload
    debra

  • View
    62

  • Download
    0

Embed Size (px)

DESCRIPTION

European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty Jan Horálek , Pavel Kurfürst Peter de Smet ETC/ACC. - PowerPoint PPT Presentation

Citation preview

Page 1: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

European scale AQ mapping (using European scale AQ mapping (using interpolation and assimilation) and interpolation and assimilation) and

evaluation of its uncertaintyevaluation of its uncertainty

Jan HorálekJan Horálek, Pavel Kurfürst, Pavel KurfürstPeter de SmetPeter de Smet

ETC/ACCETC/ACC

Page 2: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Task „Spatial air quality data“ under ETC/ACC Task „Spatial air quality data“ under ETC/ACC Implementation planImplementation plan

„to provide support in general to any AQ and spatial „to provide support in general to any AQ and spatial related activity“ related activity“ – e.g. providing inputs for CSI, AP Report– e.g. providing inputs for CSI, AP Report (maps) (maps)

Final outputs of the last year:Final outputs of the last year:ETC/ACC Technical Paper 2005/7ETC/ACC Technical Paper 2005/7ETC/ACC Technical Paper 2005/8ETC/ACC Technical Paper 2005/8„Interpolation and assimilation methods for European „Interpolation and assimilation methods for European scale air quality assessment and mapping“, Part I. and II.scale air quality assessment and mapping“, Part I. and II.

this year Task 5.3.1.2. this year Task 5.3.1.2. MNP, CHMI, NILUMNP, CHMI, NILU

Page 3: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Concentration in every place is assessed by measured Concentration in every place is assessed by measured data from surrounding stations, especially using their data from surrounding stations, especially using their linear combination: linear combination:

where where Z(sZ(sii), …, Z(s), …, Z(sii)) are the concentration are the concentration

at the surrounding stations, at the surrounding stations, ii are weights. are weights.

Two classes of interpolation methods: Two classes of interpolation methods: - deterministic (simple, e.g. IDW) - deterministic (simple, e.g. IDW) - geostatistical (utilize spatial structure of the AQ - geostatistical (utilize spatial structure of the AQ field; different types of kriging)field; different types of kriging)

1. Interpolation of air quality data 1. Interpolation of air quality data

n

iii sZsZ

10 )()(

Page 4: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Supplementary (e.g. dispersion model, altitude, Supplementary (e.g. dispersion model, altitude, meteorological parameters, like temperature or wind meteorological parameters, like temperature or wind speed, latitude or longitude) data bring more complex speed, latitude or longitude) data bring more complex information about the whole area.information about the whole area.

Linear regression model of measured AQ data with Linear regression model of measured AQ data with supplementary data + spatial interpolation of residuals supplementary data + spatial interpolation of residuals

wherewhere DD11(s), …, D(s), …, Dmm(s)(s) are supplementary parameters in point are supplementary parameters in point ss

c, c, aa11, …, a, …, amm are parameters of linear regression model are parameters of linear regression model

computed at the basis of data in the places of AQ stations computed at the basis of data in the places of AQ stations

2. Combination o2. Combination off measured AQ data and measured AQ data and different supplementary datadifferent supplementary data

)()(....)(.)( 11 ssDasDacsZ mm

Page 5: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

A.A. Developed mapping Developed mapping methodologymethodology

Page 6: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Mapping methodologyMapping methodology

rural and urban maps are constructed separatelyrural and urban maps are constructed separately(different character of urban and rural air quality) (different character of urban and rural air quality)

final map is created by merging them final map is created by merging them

Page 7: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Rural mappingRural mapping

Linear regression model of measured AQ data and Linear regression model of measured AQ data and different supplementary data + spatial interpolation of different supplementary data + spatial interpolation of residuals by ordinaryresiduals by ordinary kriging kriging

wherewhere DD11(s), …, D(s), …, Dmm(s)(s) are supplementary data in the place are supplementary data in the place ss,,

c,c, aa11, …, a, …, amm are parameters of the regression model,are parameters of the regression model,

computed at the places of AQ measurement.computed at the places of AQ measurement.

)()(....)(.)( 11 ssDasDacsZ mm

Page 8: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Linear regression model – AQ measurement vs. dispersion Linear regression model – AQ measurement vs. dispersion model EMEP, altitude, sunshine duration, 2003model EMEP, altitude, sunshine duration, 2003

measur. vs. lin. regr. model, SOMO35

y = x = 13960 + 0.24*EM + 6.244*alt. + 113.6*s.d.

R2 = 0.59

0

5000

10000

15000

20000

25000

0 5000 10000 15000 20000

linear regression model [µg.m-3.days]

mea

sure

men

ts

[µg

.m-3

.day

s]

measur. vs. lin. r. mod., PM 10 ann. avg

y = x = 1.41*EM - 0.007*alt.+ 0.26*sun.d.

R2 = 0.40

0

10

20

30

40

50

60

70

80

0 10 20 30 40

linear regression model [µg.m-3]m

easu

rem

ents

g.m

-3]

Page 9: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 – rural map (combination of AQ data with EMEP – rural map (combination of AQ data with EMEP

dispersion model, altitude and sunshine duration), 2003dispersion model, altitude and sunshine duration), 2003

Page 10: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 – urban map (rural map + interpolation of urban – urban map (rural map + interpolation of urban

increment „Delta“), 2003increment „Delta“), 2003

Page 11: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Merging of rural and urban map – using population Merging of rural and urban map – using population density mapdensity map

PM10 ann. avg vs. popul. dens. classes

0

5

10

15

20

25

30

35

<50

50-

100

100

-200

200-

500

500-

1000

1000

-200

0

2000

-500

0

>5000

popul. dens. cl. [inhbs.km-2]

PM

10 a

nn

ual

avg

. [µ

g.m

-3]

rural

urb+sub

CLASS [inhbs.km-2]

10

20

30

40

50

60

70

PM

10

an

n.

avg

, 2

00

2

[µg

.m-3

]

Page 12: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - annual average, 2003 annual average, 2003

Combined rural and urban mapCombined rural and urban map

Page 13: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

B. Final European maps for 2003B. Final European maps for 2003

Page 14: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - SOMO35, 2003 SOMO35, 2003

Combined rural and urban mapCombined rural and urban map

Page 15: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - annual average, 2003 annual average, 2003

Combined rural and urban map Combined rural and urban map

Page 16: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - 36. highest daily value, 2003 36. highest daily value, 2003

Combined rural and urban map Combined rural and urban map

Page 17: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - annual average, 2003 annual average, 2003

Concentration map + population densityConcentration map + population density

Page 18: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

C. This year’s activityC. This year’s activity

Page 19: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Actual maps for 2004 Actual maps for 2004 plusplus mapping of mapping of more components resp. parametersmore components resp. parameters

Page 20: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - SOMO35, 2004 SOMO35, 2004

Combined urban and rural mapCombined urban and rural map

Page 21: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - annual average, 2004 annual average, 2004

Combined urban and rural mapCombined urban and rural map

Page 22: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - 36. highest daily value, 2004 36. highest daily value, 2004

Combined urban and rural mapCombined urban and rural map

Page 23: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - 56. highest daily value, 2004 56. highest daily value, 2004

Combined urban and rural mapCombined urban and rural map

Page 24: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

For the purposes of protection of vegetations - rural For the purposes of protection of vegetations - rural background stations only used for mappingbackground stations only used for mapping

In this stage pure interpolation only (no use of supplem. In this stage pure interpolation only (no use of supplem. data in places with no measurements) data in places with no measurements)

82 rural background stations with NO 82 rural background stations with NOxx data in AirBase data in AirBase

For some countries NO For some countries NOxx had to be computed from NO had to be computed from NO

and NOand NO22 data in AirBase (188 stations) data in AirBase (188 stations)

For 23 For 23 stations, in which NO stations, in which NO22 is measured only, N is measured only, NOOxx was was

computed based on linear regression (separately for computed based on linear regression (separately for 4 geographic areas)4 geographic areas)

NONOxx rural mappingrural mapping

Page 25: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

NONOxx rural mapping – relation between NOrural mapping – relation between NOxx and NO and NO22

NOx vs. NO2 - rural background, 2004, North

y = 0.0054x2 + 1.1441x

R2 = 0.9892

0

5

10

15

0 2 4 6 8 10 12

NO2

NO

x

NOx vs. NO2 - rural background, 2004, North-west

y = 0.0278x2 + 0.9208x

R2 = 0.9557

01020304050607080

0 10 20 30 40

NO2

NO

x

NOx vs. NO2 - rural background, 2004, Centre + East

y = 0.0272x2 + 1.0123x

R2 = 0.922

0

10

20

30

40

50

60

0 5 10 15 20 25 30

NO2

NO

x

NOx vs. NO2 - rural background, 2004, South

y = 0.0144x2 + 1.3241x

R2 = 0.9309

01020304050607080

0 10 20 30 40 50

NO2

NO

x

Page 26: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

NONOxx - - rural map, annual average, 2004rural map, annual average, 2004

Page 27: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

SOSO22 - - rural map, annual average,rural map, annual average, 2004 2004

Page 28: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone Ozone -- AOT40 for crops, 2004 AOT40 for crops, 2004

Page 29: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for crops, 2004AOT40 for crops, 2004

„Agricultural Areas at Risk / Damage„Agricultural Areas at Risk / Damage““

Page 30: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for crops, 2004AOT40 for crops, 2004

„Arable Land at Risk / Damage“„Arable Land at Risk / Damage“

Page 31: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

OzonOzonee - - AOT40 for crops, 2004 AOT40 for crops, 2004

„Permanent Crops at Risk / Damage“„Permanent Crops at Risk / Damage“

Page 32: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for crops, 2004AOT40 for crops, 2004

„Pastures at Risk / Damage„Pastures at Risk / Damage““

Page 33: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for crops, 2004 AOT40 for crops, 2004

„Heterogeneous Agricultural Areas at Risk / Damage„Heterogeneous Agricultural Areas at Risk / Damage““

Page 34: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for forests, 2004 AOT40 for forests, 2004

Page 35: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozone -Ozone - AOT40 for forests, 2004 AOT40 for forests, 2004

„Forests at Risk / Damage“„Forests at Risk / Damage“

Page 36: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozon -Ozon - AOT40 for forests, 2004AOT40 for forests, 2004

„Broad-Leaved Forests at Risk / Damage“„Broad-Leaved Forests at Risk / Damage“

Page 37: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozon -Ozon - AOT40 for AOT40 for forests, 2004forests, 2004

„Coniferous Forests at Risk / Damage„Coniferous Forests at Risk / Damage““

Page 38: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Ozon -Ozon - AOT40 AOT40 for forests, 2004for forests, 2004

„Mixed Forests at Risk / Damage„Mixed Forests at Risk / Damage““

Page 39: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Using of actual meteorological Using of actual meteorological instead of instead of long-term long-term climaticclimatic data data

Page 40: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Under IP2005 climatic data were usedUnder IP2005 climatic data were used(averages 1961-1990)(averages 1961-1990)

This year we use actual This year we use actual (2004) (2004) meteorological data meteorological data obtained from ECWMF. obtained from ECWMF.

Improving of results (higher coefficient of determination RImproving of results (higher coefficient of determination R22):):

Using of actual meteorological data instead of Using of actual meteorological data instead of long-term long-term climaticclimatic data data

climatic_61-90 meteo_20041 (altitude, temper., w.speed, sol.rad./sunsh.dur., EMEP) 0.43 0.462 (altitude, w.speed, sol.rad./sunsh.dur., EMEP) 0.42 0.453 (altitude, temperature, wind speed, EMEP) 0.32 0.394 (altitude, w.speed, rel. humidity, EMEP) 0.37 0.415 (altitude, wind speed, sol.rad./sunsh.dur.) 0.27 0.316 (altitude, temperature, wind speed) 0.26 0.31

type of regression modelR2

Page 41: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Major improvement in the usability of supplementary Major improvement in the usability of supplementary parameters – actual wind speed improves parameters – actual wind speed improves the assessment of PMthe assessment of PM1010 (contrary to climatic long (contrary to climatic long

term wind speed)term wind speed) Caused by the differences between actual and climatic Caused by the differences between actual and climatic

wind speed.wind speed.

Using of actual meteorological data instead of Using of actual meteorological data instead of long-term long-term climaticclimatic data data

Page 42: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Comparison of actual meteorological 2004Comparison of actual meteorological 2004and climatic 1961-1990 dataand climatic 1961-1990 data

relative humidity - 2004 vs. 1961-90

y = 0.2587x + 73.919

R2 = 0.8481

80

85

90

95

100

50 60 70 80 90 100

rel. humidity 1961-90 [%]

rel.

hu

mid

ity

2004

[%

]

surf. solar rad. 2004 vs. sunshine dur. 1961-90

y = 201140x + 2599515

R2 = 0,9093

0

5000000

10000000

15000000

20000000

0 20 40 60 80

sunshine duration 1961-90 [%]

surf

. so

lar

r. 2

004

[Ws.

m-2

]

temperature - 2004 vs. 1961-90

y = 0.9227x + 1.4319

R2 = 0.9337

0

5

10

15

20

0 5 10 15 20

total precipitation 1961-90 [mm.year-1]

tota

l p

reci

p.

2004

[m

m]

wind speed - 2004 vs. 1961-90

y = 0.5934x + 1.1393

R2 = 0.3133

012345678

0 2 4 6 8

wind speed 1961-90 [m.s-1]

win

d s

pee

d 2

004

[m

.s-1

]

Page 43: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Analysis of mapping Analysis of mapping error/uncertaintyerror/uncertainty

Page 44: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

CCrossvalidation: interpolation is done without one station,rossvalidation: interpolation is done without one station, repeatedly for all points – stituation in places with no repeatedly for all points – stituation in places with no

measurement is simulated. measurement is simulated. Crossvalidation gives the objective measure of the quality Crossvalidation gives the objective measure of the quality

of interpolation. of interpolation. Several indicators: root-mean-square error (RMSE), Several indicators: root-mean-square error (RMSE),

mean prediction error (MPE), absolute error (MAE) mean prediction error (MPE), absolute error (MAE)

wherewhere Z(sZ(sii) ) is a value of concentration in the is a value of concentration in the ii-th point-th point

ŻŻ(s(sii)) is the estimation in the is the estimation in the ii-th point using other points-th point using other points

MAE should be the smallest and MPE should be the MAE should be the smallest and MPE should be the nearest to zero nearest to zero

N

iii sZsZ

NRMSE

1

2))(ˆ)((1

Crossvalidation analysis of interpolation errorCrossvalidation analysis of interpolation error

N

iii sZsZ

NMPE

1

))(ˆ)((1

N

iii sZsZ

NMAE

1

)(ˆ)(1

Page 45: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Crossvalidation scatterplot: measured values and Crossvalidation scatterplot: measured values and interpolated values interpolated from other stations interpolated values interpolated from other stations are are plottedplotted Linear regression of these values: In ideal case would be Linear regression of these values: In ideal case would be

x=y and Rx=y and R22=1.=1.

Crossvalidation analysis of interpolation errorCrossvalidation analysis of interpolation error

Page 46: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Cross-validation analysis – PMCross-validation analysis – PM1010 rural, annual average, rural, annual average,

interpolation by ord. kriging (left) and cokriging (rightinterpolation by ord. kriging (left) and cokriging (right))

Predicted (in crossvalidation) vs. measured, ordinary kriging

y = 0.3736x + 12.593R2 = 0.3363

0

5

10

15

20

25

30

35

40

0 10 20 30 40 50 60

PM10 measured

PM

10 e

stim

ated

, cro

ssv.

Predicted (in crossvalidation) vs. measured, ordinary cokriging

y = 0.5741x + 8.5817R2 = 0.584

0

5

10

15

20

25

30

35

40

0 10 20 30 40 50 60

PM10 measured

PM

10 e

stim

ated

, cro

ssv.

RMSE 5.92

MPE -0.19

MAE 4.30

R2 0.34

RMSE 4.67

MPE -0.15

MAE 3.31

R2 0.58

Page 47: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Possible only for geostatistic method (krigingPossible only for geostatistic method (kriging etc.) etc.)

Contrary to crossvalidation – this error mapping has some Contrary to crossvalidation – this error mapping has some uncertainty in itself.uncertainty in itself.

Mapping of standard prediction errorMapping of standard prediction error

Page 48: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

AOT40 for crops (rural areas), 2004 AOT40 for crops (rural areas), 2004 ordinary cokriging (using altitudeordinary cokriging (using altitude))

Page 49: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

AOT40 for crops (rural areas), 2004 AOT40 for crops (rural areas), 2004 ordinary cokriging - Prediction Standard Error ordinary cokriging - Prediction Standard Error

Page 50: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PMPM1010 - - annual average, 2003 annual average, 2003

Combined rural and urban map Combined rural and urban map

Page 51: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

PM10 -PM10 - annual average, 2003 – prediction error mapannual average, 2003 – prediction error map

done by done by Marek BrabecMarek Brabec

Page 52: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Further activitiesFurther activities

Page 53: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Further activities Further activities

Improved urban mappingImproved urban mapping

Development of mapping PM2.5 Development of mapping PM2.5

Filling the gaps in maps caused by the lack of population Filling the gaps in maps caused by the lack of population density data (ORNL database). density data (ORNL database).

Population at risk – maps and tablesPopulation at risk – maps and tables

Page 54: European scale AQ mapping (using interpolation and assimilation) and evaluation of its uncertainty

Thank you for attention.Thank you for attention.