Performance of the Mixed-Pixel Retrieval Algorithm for Coastal Path Delay Shannon Brown Jet Propulsion Laboratory Shannon.T.Brown@jpl.nasa.gov 1
Mixed Pixel PD Algorithm Mixed-pixel algorithm is statistical PD retrieval algorithm valid for mixed land/ocean scenes Relies on inherent inter-channel correlation for TLand and land fraction Sensitivity of TB to PD decreases as land fraction increases Retrieval algorithm trained as a function of land fraction Merged RaOb profiles, radiometer beam-weighted land fraction map and measured land brightness temperatures to form simulated coastal TB database LF=0 TB PD TB PD LF=0.5 11/04/2008 2
Product Availability Algorithm first applied to the AMR on Jason-2 and available operationally as AMR enhanced product since April 2009 ftp://podaac.jpl.nasa.gov/pub/sea_surface_height/ostm/preview/amr/ Enhanced product algorithms implemented on Jason-2 GDR-C processing Algorithm error is estimated to be less than 0.8 cm at 15 km from land, 1.0 cm at 10 km, 1.2 cm within 5 km, and less than 1.5 cm up to the coastline [Brown 2010, TGARS] Algorithm recently adapted to Jason-1 JMR and made available on JMR enhanced product released Oct 2010 Product available via restricted assess on PO.DAAC Will be released unrestricted after OSTST evaluation Note: Land flag bug impacts cycles 280-310 for current release Land flag set to 0 over land, but correct elsewhere Corrected product will be available by 10/30/2010 3
Objectives of Current Study Assessment of algorithm performance as applied to the JMR Comparison of JMR to AMR mixed pixel retrievals in the coastal zone Assess regional and seasonal algorithm performance Coefficients for current algorithm stratified by PD, but not regionally or temporally dependent Land brightness correlation a function of region and season Potential for regional or time dependent biases 4
Performance of JMR MP Algorithm from ECMWF JMR MP algorithm unbiased near land compared to open ocean Excess error less than 1.2 cm at 25km from land; error less than 1.5 cm within 10 km from land agrees with simulations 2cm reduction in error compared to GDR Performance degrades rapidly within 10km from land Errors near coast larger for JMR compared to AMR Mean JMR Model [cm] JMR - Model RMS Excess Error [cm] MP-alg ΔPD ( D) = ΔPD ( D) 2 ΔPD ( D 150km ) 2 ALG TOT TOT > GDR Distance to Land Distance to Land 5
Comparison of AMR and JMR Near Land During Tandem Phase JMR and AMR unbiased near land using mixed pixel algorithm Inter-sensor error < 1cm to 10km from land 11/04/2008 6
Algorithm Seasonal Performance 7
Mean AMR-Model vs Month For each coastal crossing, found double-difference of AMR-Model PD 75-100km from land and 10-30 km from land No significant seasonal dependence observed PD DD AMR Model = PD PD 10 30km 75 100km AMR Model AMR Model < 1mm month-tomonth variation 11/04/2008 8
Algorithm Regional Performance 9
AMR-ECMWF vs Distance to Land Stratified by Latitude 0 < Lat < 10 10 < Lat < 20 20 < Lat < 30 30 < Lat < 40 40 < Lat < 50 50 < Lat < 60 10
Regional AMR-Model Mean Double Difference No systematic regional biases observed with coastal algorithm 11
AMR-JMR vs Distance to Land Stratified by Latitude 0 < Lat < 10 10 < Lat < 20 20 < Lat < 30 30 < Lat < 40 40 < Lat < 50 50 < Lat < 60 11/04/2008 12
AMR-JMR MP Algorithm Bias 10-50km from Land No systematic regional biases observed between AMR and JMR MP algorithms with the exception of some inland water bodies 13
AMR-JMR MP Algorithm Bias 10-50km from Land 14
AMR-JMR MP Algorithm Bias 10-50km from Land 15
AMR-JMR MP Algorithm RMS Difference 10-50km from Land 11/04/2008 16
Summary Mixed pixel algorithm now applied to both Jason-1 and Jason-2 radiometers Both available on PO.DAAC Performance for JMR unbiased near land with PD error less than 1.2 cm up to 25 km from land Nearly 2cm performance improvement from GDR Seasonal and regional assessment revealed no significant geographically or temporally correlated systematic biases with algorithm Suggested some improvement could be obtained in areas with a locally high error Jason-2 GDR-C will have near-land processing Algorithm will next be applied to TMR data and made available as enhanced product 17