RGP geolocation analysis
The geolocation problem
• We don’t have all the necessary information:
– Optical model needs tuning• Can prob. do this now but not sufficient because…..
– Require spin axis misalignment details– Start of line accuracy out of spec.
• Have per image correction derived from SEVIRI by column to column jitter remains
The geolocation problem
• What does geolocation accuracy mean for the data?EXAMPLE 1: clear sky coast– Land point contaminated with ocean and ocean with
land (ignoring unfiltering error which exacerbates the problem)
Ocean SW radiance 20Wm-2 sr-1 and land 70Wm-2sr-1
0.5 pixel error implies: Ocean and land 45Wm-2sr-1
If this occurs 25% or time average radiances become:
Ocean: 26.25 (31% bias)
Land: 85Wm-2sr-1 (21% bias)
0.1 pixel error 25% of time reduces this to 6% and 4% biases resp.
The geolocation problem
• What does geolocation accuracy mean for the data?EXAMPLE 2: Clear ocean and cloud– Cloud point contaminated with ocean and ocean with
cloud (ignoring unfiltering error which exacerbates the problem)
Ocean SW radiance 20Wm-2 sr-1 and cloud 150Wm-2sr-1
0.5 pixel error implies: Ocean and land 85Wm-2sr-1
If this occurs 5% or time average radiances become:
Ocean: 23.25 (16% bias)
Land: 147Wm-2sr-1 (2% bias)
0.1 pixel error 25% of time reduces this to 3.2% and 0.4% biases resp.
NOTE cloud forcing calculations: errors compound
How good is the reprocessing geo?Reprocessing
Compared to optimal
How does this compare to NRT geoReprocessed
compare to NRT
One pixel geolocation difference
Dark blue 5%
Light blue 10%
Cyan 20%
Green 30%
Yellow 40%
Reed 50%
White 100%
• Dark blue 5%• Light blue 10%• Cyan 20%• Green 30%• Yellow 40%• Reed 50%• White 100%
Dark blue 5Wm-2
Light blue 10Wm-2
Cyan 20Wm-2
Green 30Wm-2
Yellow 40Wm-2
Reed 50Wm-2
White 100Wm-2
DERIVING the BANANA
• Pixel azimuth and elevation derived from lon-lat determined by RMIB reprocessed geolocation matching and NANRG satellite position
DERIVING the BANANA
• Pixel azimuth and elevation derived from lon-lat determined by RMIB reprocessed geolocation matching and NANRG satellite position
• Probability distribution of pixel azimuth and elevation built up from the full dataset
Distribution of pixel positionWe can then look at the proportion of the time the pixel is a given distance from the most probable position
1% < Purple < 5%5% < Blue < 10% 10% < Cyan < 25%25% < Green < 50%50% < Orange < 60%60% < Red
Distribution of pixel positionWe can then look at the proportion of the time the pixel is a given distance from the most probable position
1% < Purple < 5%5% < Blue < 10% 10% < Cyan < 25%25% < Green < 50%50% < Orange < 60%60% < Red
Summary• Reprocessing geo very close to optimal
– Within 0.25 pixel except towards disk edge
– Not possible in NRT • Cost 32,000€, or slower than real time archive or more work solution (still
non-ideal as level 1.5 and level 2 geo disconected)
• NRT geo often more than 0.5 pixel different from reprocessing
• Updated optical model in level 1.5 NANRG + current paramters
• with per column azimuth and elevation correction better than 0.2 pixel to reprocessing 90% of time
• With per image azimuth and elevation correction better than 0.3 pixel to reprocessing 90% of time
• Need to asses what final decision means on products