Which Salinity Data are Best for You?

SMAP Land Correction - Part 1

The data processing approaches used by RSS and JPL result in differences in sea surface salinity (SSS) coverage along coasts.The processes employed, known as "Land Correction," are outlined in this section.
RSS (V4) JPL (V5)
Level 1B (processed at sensor level)
  • Noise for SMAP Effective Field of View is 2 PSU
    • Useless for most ocean applications
  • Far sidelobe correction is applied
    • This is comparable to a "land correction" 
  • Pre-processes collocations of model sea surface salinity, sea surface temperature, significant wave height and SMAP data
  • Version 5 is "based on the newly released SMAP V5 Level-1 Brightness Temperatures (TB). An enhanced calibration methodology has been applied… which improves absolute radiometric calibration and reduces the biases between ascending and descending passes."
  • Look-Up Table Approach 
    • This is consistent with other empirical data-driven calibration done at Level 1B
Level 2 (swath)
  • Resample using Backus-Gilbert Optimum Interpolation (BG-OI) onto fixed ¼-degree Earth grid
  • 40-km product
    • BG-OI reduces noise to 1 PSU over the open ocean; however, BG-OI gives poor results close to land
    • No Q/C applied but quality info is included in L2 files 
  • 70-km product
    • Employ next-neighbor averaging of the 40-km product to reduce noise to 0.5 PSU
    • This "smearing out" is done instead of BG-OI resampling of the 70-km footprint
    • Extensive Q/C filters out questionable data
  • Resample onto fixed ¼-degree Earth grid
  • 60-km product
    • JPL spatially averages the native 40-km retrieval to a larger footprint to reduce data noise
Level 3 (gridded) 
  • For each grid size (i.e., 40-km and 70-km), two products are created: 8-day and monthly averages
    • 40-km products
      • Some basic Q/C (sun-glint, high galaxy, high wind, etc.) is applied
    • 70-km products
      • Extensive Q/C at Level 2 (see above) has filtered out questionable data
      • Monthly averaging reduces noise to 0.2 PSU (science goal)
  • Time-averaged over one month to reduce noise to 0.2 PSU (mission requirement)
Next Steps
  • Our next version (RSS V5) will have formal uncertainty estimates including errors from land contamination in Level-2C and all Level-3 products.
  • We are working to improve our data screening… that is filtering implemented between Level-2 and Level-3 products.

Key Definitions, including differences between RSS and JPL:

Land fraction
RSS: Provided two land fractions: (1) Percentage of land 3-dB of antenna footprint ("fland"); and (2) Land fraction weighted by antenna gain pattern ("gland").
JPL: Ratio of gain-weighted solid angle over land to that in the visible disk ("fland"). Note: JPL's fland is equivalent to RSS's gland.
Geophysical Model Function (GMF) – An empirical relationship between geophysical factors, the observation geometry, and radiometer wavelength (e.g., L-band)
RSS: The measured SMAP antenna temperature – which we record in Level-2C files – depends on look direction (i.e., forward-looking or aft-looking). That means at a specific location, there is a brightness temperature (TB) record for the forward-looking and another TB record for the aft. Both differ because of galaxy reflection, Faraday rotation, and wind direction. The GMF used in our algorithm accounts for all of this and thus depends on look direction. Moreover, when you get close to land, fland and gland also depend on look direction, because SMAP's footprint is not circular.
JPL: We have a model function, which is a big table of expected observations based on given geometry, salinity, wave height, sea surface temperature, and wind speed/direction. Our Maximum Likelihood Estimator (MLE) process – also called the "objective function" – solves for the best salinity that makes the measurements most match the GMF. This overall equation is what we minimize. How good that fit is helps to quantify the uncertainty; this also comes out of the algorithm in the retrieval process.