Upload
grssieee
View
401
Download
0
Tags:
Embed Size (px)
Citation preview
04/12/23 1
Yunyue Yu1, Ming Chen2, Dan Tarpley3 , Jeffrey L. Privette4 ,
Miguel O. Roman5
Present by Ming Chen
1NOAA/NESDIS/STAR, 2IMSG at NOAA/NESDIS, 3Short & Associates, 4NOAA/NESDIS/NCDC, 5NASA/GSFC
IGARSS 2010, Honolulu, Hawaii, USA
Outline
04/12/23 2
• Potential Issues of Using Ground Measurement to Validate Satellite LST
• Our Goal and Proposed Approach
• Preliminary Analysis and Results
• Further Plans
IGARSS 2010, Honolulu, Hawaii, USA
Issues of LST Cal/Val
04/12/23 3IGARSS 2010, Honolulu, Hawaii, USA
Different sensor IFOVs, introducing site heterogeneities through
a) landscape b) land cover
c) soil moisture , etc.
Ground data accuracy Possible bias in pixel geo-referencing Cloud contamination
All the above may result in thermal and emissivity uncertainties
Our Goal
04/12/23 4IGARSS 2010, Honolulu, Hawaii, USA
• To evaluate potential scale mismatch uncertainties involved in direct site-to-pixel comparison for ABI/VIIRS LST validation
• To establish scaling model that would provide statistical relationship between ground station LST and the overpass ABI/VIIR pixel LST
Approach
04/12/23 5IGARSS 2010, Honolulu, Hawaii, USA
• Use well-calibrated high-resolution satellite pixels to quantitatively characterize the sub-pixel heterogeneity of a low resolution satellite pixel.
• Establish relationship between the low-resolution pixel and its sub-pixels.
Pixel-Synthesizing Scheme
04/12/23 6
Synthetic Pixels: Large circles
Sub-pixels: ASTER 90m TIR Pixels
Each synthetic pixel encloses the target ground site.
The center of synthetic pixel varies around the ground site, for counting variable over-passing satellite footprints, or small geo-referencing bias.
IGARSS 2010, Honolulu, Hawaii, USA
ASTER 90m TIR pixels Ground site
Scaling and Characterizing (1)
drrdrrTTTT occijijijijsat ,,
04/12/23 7
Aggregation of Synthetic pixel from high-resolution ASTER pixels
IGARSS 2010, Honolulu, Hawaii, USA
Tij is individual ASTER pixel observation within the synthetic pixel scope, rij is the ASTER pixel position vector, rc is position vector of synthetic pixel center, ro is the position vector of ground site.
Scaling and Characterizing (2)
04/12/23 8IGARSS 2010, Honolulu, Hawaii, USA
}{ gsat TTE }{ gsat TT
CTT gsat
)()( gccsatgsat TTTTTT
Sub-pixel heterogeneity
Quantification of the difference between the Synthetic Pixel and Ground Measurement, that is,
Evaluation: and with
the model:
Note that:
Systematic bias between measurements of different sensors
In-Situ LST and Broadband Emissivity
04/12/23 9
umumum cba 125.105.8
• SURFRAD LST need be calculated from SURFRAD radiation measurements.
Where Rup_pir, Rdown_pir are upwelling and downwelling thermal radiations. εs is surface broadband emissivity.
a=0.2122, b=0.3859,c=0.4029 (Wang, 2004)
• Regression of Broadband emissivity
IGARSS 2010, Honolulu, Hawaii, USA
4/1___ )]/()[( spirdownspirdownpirupg RRRT
Regression based on the UW-Madison Baseline Fit Emissivity Database( Seemann et al., 2008).
Enhanced Cloud Screening
04/12/23 10
Enhanced cloud filtering is needed for high quality satellite-ground match-up dataset
Cloud
yIGARSS 2010, Honolulu, Hawaii, USA
Computing Algorithms
04/12/23 11IGARSS 2010, Honolulu, Hawaii, USA
Enhanced Cloud Screening Satellite cloud mask data Spatial, temporal ,Spectral variation tests Manual control
Match-up Datasets
LST Algorithms1. Satellite LST calculation or
extraction2 SURFRAD LST calculation
Time-serial Analysis
Synthetic Datasets Spatial analysis models up-scaling models down-scaling models
Geolocation Match-up
Satellite and Ground Time Match-up
Original Datasets
Stations and Match-up Datasets
Stations Latitude-Longitude
Clear CasesBy ASTER Cloud masks
Clear CasesBy Augmented Screening
Desert Rock, Co
36.63oN 116.02oW
63 46
Bondville, IL
40.05oN 105.24oW
115 51
PennState,PA
40.72oN 77.93oW
61 20
Boulder, CO
40.13oN 105.24oW
35 13
Fort Peck, MT
48.31oN 105.10oW
12 8
04/12/23 12IGARSS 2010, Honolulu, Hawaii, USA
In addition to the candidate stations, shown in this map is the occurrence frequency of completely clear sky over US continent, which would affect the data size.
Sample Scatter Plots
04/12/23 13IGARSS 2010, Honolulu, Hawaii, USA
DesertRoc
k
)()( gccsatgsat TTTTTT
Sample Statistics of Desert Rock Site
04/12/23 14IGARSS 2010, Honolulu, Hawaii, USA
Pixel ID Deg E{Tsat-Tc} σ{Tsat-Tc} E{Tsat-Tg} σ{Tsat-Tg}
0 C -0.04 0.69 1.78 2.131 0 0.01 0.60 1.82 2.262 45 -0.08 0.61 1.74 2.203 90 -0.20 0.92 1.61 1.994 135 -0.06 0.96 1.75 2.035 180 0.24 0.98 2.05 2.186 225 0.34 0.80 2.15 2.307 270 0.26 0.65 2.07 2.408 315 0.16 0.60 1.97 2.37
Note that:E{Tc-Tg}: 1.81 K σ{Tc-Tg}: 2.46 K
Topographic Feature and Mean Satellite-In Situ Difference
04/12/23 15IGARSS 2010, Honolulu, Hawaii, USA
Tsat-Tg is vectorized, where the modes are the absolute values of Tsat-Tg (Column 7 in Table) , and the directions are the surface site to the synthetic pixel centers. The contour shapes reflect the spatial structure of Tsat-Tg.
Aggregated Results
04/12/23 IGARSS 2010, Honolulu, Hawaii, USA
Station E{Tsat-Tc} σ{Tsat-Tc} E{Tc-Tg} σ {Tc-Tg} E{Tsat-Tg} σ {Tsat-Tg}
DesertRock, NV 0.07 0.76 1.81 2.46 1.88 2.21
Boulder, CO 0.13 0.72 0.77 2.60 0.90 2.58
Fort Peck, MT 0.06 0.99 -0.20 3.36 -0.15 2.62
Bondville, IL 0.09 1.05 0.59 2.01 0.68 2.08
Penn State, PA 0.08 1.14 -0.25 2.09 -0.17 1.99
18
Discussion and Conclusive Remarks
04/12/23 17IGARSS 2010, Honolulu, Hawaii, USA
• The mean difference between the synthetic pixel and the collocated central ASTER pixel is less than 1K, and the standard deviation of such mean difference is around 1K, which indicates that a) SURFRAD stations in this study may be treated as homogenous or slightly heterogeneous sites(within 1Km) for VIIRS/ABI LST cal/val. b) LST of SURFRAD measurements may be used as good references for VIIRS/ABI LST cal/val if the measurements are of high-quality and the in-situ estimation of LST is accurate enough.
• Directional variations are small, which indicates that a) Small geo-referencing (within 1Km) bias may not be an issue affecting VIIRS/ABI LST cal/val at the above SURFRAD sites. b) Directional variation of the potential sub-pixel heterogeneity is found to be consistent with the physical topographic features, even if it is small.
Discussion and Conclusive Remarks
04/12/23 18IGARSS 2010, Honolulu, Hawaii, USA
•The mean difference between the synthetic LST and SURFRAD LST(Tsat-Tg) is generally around 1K and the standard deviation of such difference is around 2K , which is relatively larger than those between the synthetic LST and the collocated ASTER pixel (Tsat-Tc). Systematic bias between measurements of different sensors (Tc-Tg ) is the major source of Tsat-Tg.
•The limited datasets doesn’t allow us to characterize the seasonal variation of heterogeneities, which is more desirable than a simple mean difference. More datasets are expected. And about 1K difference seems unavoidable in practice.
Further Efforts
04/12/23 19IGARSS 2010, Honolulu, Hawaii, USA
• The limited data made it impossible to perform fine analysis over time scales of interest, e.g., seasonal variation. More match-up data will be analyzed.
• Analysis over sites of different surface types
• Up/down-scaling models
• Emissivity uncertainties
04/12/23 20
Thank you for your attention !
IGARSS 2010, Honolulu, Hawaii, USA