Upload
phamkhue
View
216
Download
0
Embed Size (px)
Citation preview
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 1 of 26
Guillem Sòria, José A. Sobrino,Juan C. Jiménez-Muñoz, Mónica Gómez, Juan Cuenca,
Mireia Romaguera and Malena Zaragoza
University of Valencia, Spain
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 2 of 26
IndexIndex
1.1. AATSR LST AATSR LST algorithmsalgorithms proposedproposed
a)a) MethodsMethods ofof atmosphericatmospheric correctioncorrection
b)b) CoefficientsCoefficients calculationcalculation
c)c) SplitSplit--windowwindow andand DualDual--AngleAngle methodmethod
2. Validation over an homogeneous area
3. Study over heterogeneous areas
a) Marrakech Field Experiment 2003
b) Barrax Field Experiment 2004
c) Classification process
4. Conclusions
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 3 of 26
AATSR AATSR CharacteristicsCharacteristics
AATSR: Advanced Along-Track Scanning Radiometer
Provide two views of the surface, nadir and forward (0 and 55 degrees)
and thus improve atmospheric correction
AATSR has 4 NIR/VIS and 3 TIR channels, with a spatial resolution of
1km x 1km at nadir view and 1.5km x 2km at forward view
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 4 of 26
iθτ iθτ iθτ
1. AATSR 1. AATSR algorithmsalgorithms proposedproposed
a) a) MethodsMethods ofof atmosphericatmospheric correctioncorrection
SplitSplit--windowwindow methodmethod::The SW method uses observations at two different wavelengths with the
same observation angle.
2ji εε
ε+
=210 )1()( BBBTTATT jiis εε ∆−−+−−+=
ji εεε −=∆DualDual--angleangle methodmethod::
The DA method uses observations at two different observation angles
within the same wavelength interval
210 )1()( BBBTTATT nfnns θεε ∆−−+−−+= fn εεεθ −=∆
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 5 of 26
iθτ iθτ iθτ
1. AATSR 1. AATSR algorithmsalgorithms proposedproposed
b) b) CoefficientsCoefficients calculationcalculation
The algorithms coefficients were obtained from simulated data to applythem to a large amount of surfaces.
From MODTRAN simulation code:
transmissivity transmissivity θτ i
↑atmiR θ
↓atmiR θupwellingupwelling and downwelling and downwelling
atmospheric atmospheric radianceradiance
waterwater vapor vapor contentcontent W
From laboratory spectral library, a set of emissivity values of:
VegetationVegetation: : GrassGrass, , ConifersConifers, , DecidiousDecidious
BareBare soilssoils, , rocksrocks
WaterWater11µm and 12 µm channels,nadir and forward views.
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 6 of 26
1. AATSR 1. AATSR algorithmsalgorithms proposedproposed
c) c) SplitSplit--windowwindow andand DualDual--AngleAngle algorithmsalgorithms
Algorithm Expression
SW n, Quad: Ts = T2n + 0.61(T2n-T1n) + 0.31(T2n-T1n)2 + 1.92
SW n, Quad, ε: Ts = T2n + 0.76(T2n-T1n) + 0.30(T2n-T1n)2 + 0.10 + 51.2(1-ε)
SW n, Quad, ε, ∆ε: Ts = T2n + 1.03(T2n-T1n) + 0.26(T2n-T1n)2 – 0.11 + 45.23(1-ε) – 79.95∆ε
SW n (W), ε, ∆ε, W: Ts = T2n + (1.01+ 0.53W)(T2n-T1n) + (0.4-0.85W) + (63.4-7.01W)(1-ε) - (111-17.6W)∆ε
SW n, Quad, ε, ∆ε, W: Ts = T2n + 1.35(T2n-T1n) + 0.22(T2n-T1n)2 – (0.82-0.15W) + (62.6-7.2W)(1-ε) - (144-26.3W) ∆ε
SW n, Quad(W), ε, ∆ε, W: Ts = T2n + (1.97+0.2W)(T2n-T1n) - (0.26-0.08W)(T2n-T1n)2 + (0.02-0.67W) + (64.5-7.35W)(1-ε) - (119-20.4W) ∆ε
DA 11 Quad: Ts = T2n + 1.36(T2n-T2f) + 0.18(T2n-T2f)2 + 1.78
DA 11 Quad, ε: Ts = T2n + 1.56(T2n-T2f) + 0.15(T2n-T2f)2 - 0.34 + 51.9(1-ε2n)
DA 11 Quad, ε, ∆ε: Ts = T2n + 1.57(T2n-T2f) + 0.15(T2n-T2f)2 –0.11 + 51.7(1-ε2n) – 25.8∆εθ
DA 11 W, ε, ∆ε, W: Ts = T2n + (1.62+0.3W)(T2n-T2f) + (0.18-0.52W) + (70.1-7.18W)(1-ε2n) - (35.4-3.67W)∆εθ
DA 11 Quad, ε, ∆ε, W: Ts = T2n + 1.92(T2n-T2f) + 0.12(T2n-T2f)2 – (0.39+0.09W) + (71-7.55W)(1-ε2n) - (35.8-3.88W)∆εθ
DA 11 Quad(W), ε, ∆ε, W: Ts = T2n + (2.67-0.07W)(T2n-T2f) - (0.29-0.09W)(T2n-T2f)2 - (0.31+0.28W) + (72.5-7.9W)(1-ε2n) - (35.8-4.1W)∆εθ
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 7 of 26
1. AATSR 1. AATSR algorithmsalgorithms proposedproposed
c) c) SensitivitySensitivity analysisanalysis fromfrom error error theorytheory ofof thethe algorithmsalgorithms
Algorithm σmod (K)
σnoise (K)
σε (K)
σWV (K)
σtotal (K)
SW n, Quad: 1.73 0.07 1.73
SW n, Quad, ε: 1.39 0.07 0.18 1.40
SW n, Quad, ε, ∆ε: 1.05 0.09 0.59 1.20
SW n (W), ε, ∆ε, W: 0.59 0.10 0.83 0.45 1.12
SW n, Quad, ε, ∆ε, W: 0.93 0.11 1.06 0.20 1.43
SW n, Quad(W), ε, ∆ε, W: 0.52 0.15 0.89 0.37 1.10
DA 11 Quad: 1.31 0.11 1.32
DA 11 Quad, ε: 0.72 0.12 0.18 0.75
DA 11 Quad, ε, ∆ε: 0.69 0.13 0.26 0.74
DA 11 W, ε, ∆ε, W: 0.47 0.13 0.35 0.36 0.70
DA 11 Quad, ε, ∆ε, W: 0.57 0.15 0.36 0.17 0.71
DA 11 Quad(W), ε, ∆ε, W: 0.38 0.20 0.37 0.24 0.62
σmod: residual atmospheric error.
σnoise: noise error: NE∆T=0.05 K.
σε: emissivity error: ε(ε)= 0.005.
σWV: water vapor column error:
ε(WV)= 0.5 gcm-2.
{ }2222mod WVnoisetotal σσσσσ ε +++=
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 8 of 26
2. 2. ValidationValidation overover anan homogeneoushomogeneous areaarea
Data Data providedprovided by A. by A. PrataPrata, CSIRO, Australia, CSIRO, Australia
Algorithm σ teoric (K)
σvalidació (K)
Bias (K)
RMSD (K)
SW n, Quad: 1.73 1.51 1.59 2.19 SW n, Quad, ε: 1.40 1.56 1.48 2.15
SW n, Quad, ε, ∆ε: 1.20 1.66 1.28 2.09 SW n (W), ε, ∆ε, W: 1.12 1.93 1.28 2.31
SW n, Quad, ε, ∆ε, W: 1.43 1.81 1.13 2.13 SW n, Quad(W), ε, ∆ε, W: 1.10 2.06 1.17 2.36
DA Quad: 1.32 1.24 0.36 1.29 DA Quad, ε: 0.75 1.29 0.02 1.29
DA Quad, ε, ∆ε: 0.74 1.29 0.02 1.29 DA W, ε, ∆ε, W: 0.70 1.25 -0.21 1.27
DA Quad, ε, ∆ε, W: 0.71 1.39 -0.35 1.43 DA Quad(W), ε, ∆ε, W: 0.62 1.28 -0.39 1.34
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 9 of 26
3. 3. Study over heterogeneous areas
Barrax, Albacete, Spain
5 march 2003
Marrakech, Morocco
17 july 2004
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 10 of 26
3. 3. Study over heterogeneous areas
a) Marrakech a) Marrakech campaigncampaign
Marrakech fieldexperiment took place in an area of thewater-catchment of theTensift river in marchof 2003.
(31º40’ N, 07º35’ W, 600 m elevation)
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 11 of 26
3. 3. Study over heterogeneous areas
a) Marrakech a) Marrakech campaigncampaign
Bare Soil field
Mixed SiteVegetation + Bare Soil
Vegetated field
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 12 of 26
3. 3. Study over heterogeneous areas
b) b) BarraxBarrax campaigncampaign
( 39º03’ N, 02º06’ W, 700 m elevation)Barrax test site is situated within La Mancha, 20 km far away from the capital town Albacete.
The area around Barrax is characterised by a flat morphology and large, uniform landuse units.
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 13 of 26
3. 3. Study over heterogeneous areas
b) b) BarraxBarrax campaigncampaignChris / Proba image acquired during the campaign, near to the AATSR overpass to avoid changes in crop growth.
Alfalfa
Corn
Green grass
Wheat
Bare Soil
Thermal measurements in:
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 14 of 26
c) c) ClassificationClassification processprocess3. 3. Study over heterogeneous areas
The Chris/Proba pixels can be identifiedaccording to their NDVI values.
NDVI Crops
0Nadir view Bare soil
Cut Wheat
0.3
0.5Green grass
Corn
Alfalfa0.9
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 15 of 26
ClassificationClassification processprocess
The Chris/Proba pixels can be identifiedaccording to their NDVI values. NDVI Crops
0Forward view ( 55 degrees)Forward view ( 55 degrees) Bare soil
Cut Wheat
0.3
0.5Green grass
Corn
Alfalfa0.9
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 16 of 26
ClassificationClassification processprocess
Measurement strategy: pixels classified into 3 classes.
Bare Soil
Cut wheat
Alfalfa
NDVI
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 17 of 26
ClassificationClassification processprocess overover a a ChrisChris/Proba /Proba imageimageSupervised - Maximum Likelihood classification
Bare Soil
Cut wheat
Alfalfa
Grid of AATSR pixels
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 18 of 26
ClassificationClassification processprocess overover a a ChrisChris/Proba /Proba imageimageSupervised - Maximum Likelihood classification
Proportion of Crops in every AATSR pixel
From in-situ temperatureand emissivity of eachcrop, it is possible toobtain LST and LSE foreach pixel to validateAATSR data.
EffectiveLST andLSE of eachpixel
Class Crop1 Bare Soil2 CutWheat3 Alfalfa
Class %1 67.5%2 20.6%3 11.9%
Class %1 31.4%2 15.7%3 52.8%
Class %1 39.9%2 33.7%3 26.4%
Class %1 37.3%2 7.4%3 55.3%
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 19 of 26
ValidationValidation ofof LST LST fromfrom algorithmsalgorithms andand insituinsitu datadata
The AATSR algorithms need the following data:radiometric temperature from the AATSR image, value of the water vapor content and emissivity of both spectral bands and view angles.
Data from Barrax campaign Pixel 1 Pixel 2 Pixel 3 Pixel 4
Radiometric Temperature 11µm nadir (K) 303.8 300.8 301.4 300.4
Radiometric Temperature 11µm forward (K) 300.1 301.2 300.1 299.1
Radiometric Temperature 12µm nadir (K) 300.4 297.9 298.5 297.5
W (g/cm2) 2.36 2.36 2.36 2.36
Emissivity nadir (11µm) 0.971 0.980 0.977 0.980
Emissivity nadir (12µm) 0.976 0.985 0.982 0.985
Emissivity forward (11µm) 0.959 0.965 0.963 0.965
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 20 of 26
ValidationValidation ofof LST LST fromfrom algorithmsalgorithms andand insituinsitu datadataDifference between LST from algorithm and LST measured in situ.
Higher errors are observed in the DA algorithms.
Difference LSTalgorithm – LST insitu Pixel 1 Pixel 2 Pixel 3 Pixel 4
1 -1.7 1.9 -0.8 0.9 2 -1.8 1.3 -1.2 0.3 3 -1.3 1.9 -0.7 0.9 4 -1.6 1.7 -0.8 0.8 5 -1.0 2.1 -0.4 1.1
SW algorithms
6 -1.8 1.7 -0.8 0.7 1 0.0 -3.1 -3.3 -1.4 2 1.5 -5.0 -3.2 -2.4 3 1.5 -5.1 -3.4 -2.6 4 1.3 -6.4 -3.6 -2.7 5 1.9 -5.8 -3.4 -2.6
DA algorithms
6 1.0 -6.4 -3.4 -2.6
1 2 34
56
SW
DA-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
LST
diff
eren
ce
Algorithm
Pixel #1
12
34
56
SW
DA-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
LST
diffe
renc
e
Algorithm
Pixel #2
12
34
56
SW
DA-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
LST
diffe
renc
e
Algorithm
Pixel #3
12
34
56
SW
DA-6.00
-4.00
-2.00
0.00
2.00
4.00
6.00
LST
diffe
renc
e
Algorithm
Pixel #4
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 21 of 26
ClassificationClassification processprocess overover a Landsat5/TM a Landsat5/TM imageimage
Supervised - Maximum Likelihood classification
Specific classes selected: Bare Soil Mixed Vegetated
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 22 of 26
ValidationValidation ofof LST LST fromfrom algorithmsalgorithms andand insituinsitu datadata
Marrakech Field Campaign Barrax Field Campaign
Difference LSTalgorithm – LST insitu Bias, K σ, K rmse, K
1 -2.22 1.17 2.51
2 -2.54 1.16 2.79
3 -1.75 1.16 2.10
4 -1.48 1.15 1.88
5 -1.33 1.17 1.77
SW algorithms
6 -1.11 1.16 1.61
1 -1.22 3.58 3.78
2 -1.60 3.80 4.12
3 -1.49 3.82 4.10
4 -1.62 3.79 4.12
5 -1.24 4.26 4.43
DA algorithms
6 -1.79 4.05 4.42
Difference LSTalgorithm – LST insitu Bias, K σ, K rmse, K
1 1.82 3.38 3.84
2 1.38 3.28 3.56
3 1.90 3.30 3.80
4 1.63 3.22 3.61
5 2.16 3.31 3.95
SW algorithms
6 1.54 3.17 3.53
1 0.67 5.22 5.27
2 0.37 5.69 5.70
3 0.26 5.72 5.72
4 -0.26 5.73 5.73
5 0.26 6.04 6.05
DA algorithms
6 -0.38 5.40 5.42
Averaged values from a 4x4 grid of AATSR pixels
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 23 of 26
RescallingRescalling ofof forwardforward AATSR AATSR pixelspixelsThe higher uncertainties observed in the evaluation of the DA algorithmsare supposed to be a problem of the different footprint associated to the
AATSR nadir and forward pixels.
This effect is currently under study.
Nadir pixels: 1km x 1kmThese pixels are regridded onto a 1x1km grid. A process of pixel make-up is carried out.
Forward pixels: 1.5km x 2km
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 24 of 26
4. 4. Conclusions
SW and DA algorithms have been proposed
DA are better than SW in the simulation process, confirmed in homogeneous areas
In heterogeneous areas DA are worse than SW algorithms, confirmed in Marrakech and Barrax campaigns
Due to different footprint between nadir and forward images
Additional information is necessary to consider this effect
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 25 of 26
MERIS – (A)ATSR WORKSHOP ESRIN, Frascati, Italy 26-30 September 2005 26 of 26
-3
-2
-1
0
1
2
3
4
5
6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
NDVI nadir
Tnad
ir - T
forw
ard
(ºC
)
-3
-2
-1
0
1
2
3
4
5
6
7
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
NDVI nadir - NDVI forward
Tnad
ir - T
forw
ard
(ºC
)