SST Reprocessing 2014.0


The R2014.0 processing of MODIS Sea Surface Temperature (SST) data by the OBPG is the first major update to the MODIS SST algorithms in over a decade. The primary changes incorporated into this reprocessing were to the derivation and application of the correction factors for response versus scan angle (RVS) and mirror side. Algorithm coefficients are now latitude based. These changes, combined with minor adjustments to quality level definitions and thresholds for some SST flags, reduce the the global uncertainty of the SST product by ~0.1. In addition, seasonal and regional biases are reduced. Details are available in the white paper "Implementation of Version 6 AQUA and TERRA SST processing, K. Kilpatrick , G. Podesta, S. Walsh, R. Evans, P. Minnett".

Summary of Changes

Changes to the SST algorithms

  • A total of 3 additional correction terms were added to both the LWIR SST and MWIR SST4 algorithm formulations; 2 terms are related to a satellite zenith angle correction and a single term relating to a mirror side correction
  • Coefficients were estimated and are applied by latitude band and month of generic year.
    • The LWIR SST algorithm no longer selects coefficients based on brightness temperature difference as a proxy for water vapor.

Changes to SST Flag thresholds

  • SSTREFDIFF changed to a cold only tests SST - REFSST >= -3.0 to prevent flagging, as bad, good pixels which may be warmer than reference as a result of the diurnal heating of the skin surface at low wind speeds during the day.
  • SSTREFDIFF modified to include a more stringent cold threshold (SST - REFSST) >= -1.25, in regions likely to be contaminated by dust where retrievals are generally colder. Dust Region is defined as falling within a latitude <= 30N and > 10S and longitude of and between 105 E and 105W.
  • SSTF CLOUD binary decision trees added to identify pixels with contaminated atmospheres (dust absorbing aerosols ect) not captured by uniformity tests.

Quality level definition changes

  • Quality 0 and Quality 1 definitions differ only by the SST Flag HISENZ and BTNONUNIF flag.
  • Quality level of daytime pixels in glint regions, that are otherwise clear in all SST flags, can be no better than 1 due to visible band tests not being valid in the glint region
  • Quality levels of pixels with the BTVNONUNIF set is changed to be no better than 3.
  • Quality level of pixels with the SSTF CLOUD flag set can be no better than a 3. An inherited coding error was found in the version V5 code at OBPG, and traced to the original MODAPS code, whereby the SST CLOUD bit was being set but not evaluated in regard to the final quality level.
  • Pixels failing the SSTREFDIFF are now assigned to Quality level 2
  • Quality level 3 and better will be binned in global maps, previously only quality 2's and better were included.

Addendum R2014.0.1

The MODIS Aqua and Terra R2014.0 daytime SST was found to have a higher false positive cloud misclassification rate than expected. The problem was traced to an issue with a few of the visible band tests of the cloud classification algorithm. Thresholds based on reflectance were being compared against units of radiance, and on occasion this resulted in large areas of clear pixels being misclassified as cloud. Misclassification was dependent on the single path a pixel took through a series of tests in the Classification Tree. In addition to correcting the issue in the SST cloud classifier, L3 map products for the R2014.0.1 use improved mapping software, L3mapgen, based on weighted spatial averaging rather than the previous nearest neighbor methodology of the SMI maps. Users may find minor differences in the mapped values due to changes in the mapping method, in addition to any change due to the increased number of best or good quality retrievals in some L3 bins in the ISEAG grid.