Sea surface temperature metrics for evaluating OSI SAF sea ice concentration products
In the context of the OSI SAF Visiting Scientist Program, Sandra Castro from the University of Colorado Boulder, USA, worked on sea surface temperature metrics for evaluating OSI SAF sea ice concentration products. This work took place from May 2019 to September 2019 and was supervised by Rasmus Tonboe and Jacob Høyer from DMI.
The aim of this investigation is to do additional quality control on the EUMETSAT OSISAF and ESA CCI sea-ice concentration climate data records (CDRs), with emphasis on the 25 km OSI-450 and SICCI-25 km, gridded sea ice concentration (SIC) products for the Arctic region. This post-processing quality assessment is based on SIC comparisons relative to an independent SST product that has been vetted for quality. The reference SST product of choice is the level 4, 5-km OSTIA SST. To facilitate comparisons between the SIC and SST products, both OSI-450 and SICCI-25 have been re-gridded to the same 5 km-pixel grid of the SST product. SIC pixels with corresponding SST >= 3 °C, are flagged for additional quality screening in both products. This set, termed the “inconsistency set,” is used to identify spurious SIC, and their sources, and to design quality control checks to remove these sources of error from the SIC CDRs.
It is found that the vast majority of spurious retrievals are located along the sea ice – open water edge, coastal boundaries and subpolar marginal seas. They result from uncorrected atmospheric effects impacting low concentration retrievals over water, land contaminated microwave retrievals along coastlines, and representation errors resulting from gridding/remapping data to different spatial resolutions. Interpolation procedures associated with changes in grid resolution can smear the ice edge over neighboring pixels resulting in proportionally larger areas of underestimated retrievals and/or erroneous retrievals spreading over wider areas away from the coasts and into the open waters.
A new filter with multilevel thresholding is evaluated for its skill to eliminate SST-flagged spurious SICs from the inconsistency set. The filter chain is based on multiple thresholds that check for conservation of the SIC-SST dependence within mixed pixels (i.e., SST = ƒ(SIC), maximum SST and maximum SIC), a minimum distance from land requirement to eliminate false/underestimated SIC retrievals due to land-spillovers, and a valid range of standard smearing uncertainties (SSE), as this variable was found to be the most sensitive to blurring edge effects introduced by interpolation incurred when remapping/regridding to finer spatial resolutions.
Each SIC product reports an estimated SSE for each pixel. Evaluation of the SSE suggested that the OSISAF uncertainties are underestimated, requiring a more aggressive thresholding. A new “energy” conservation metric, measuring the continuity between SIC and SST within a pixel, is proposed to evaluate the impact of the noise corrections on subsequent truncations of the inconsistency set after each threshold eliminated a fraction of the data. The new metric appears to be highly sensitive to small changes resulting from masking out the SIC outliers. According to the metric, OSI-450 SIC retrievals are strongly impacted by corrections addressing noise resulting from land spillovers, followed by residual atmospheric effects. The reverse order is true for SICCI. Although the filtering methodology proposed here rejected 98% of the initial inconsistency set, the remaining de-noised set has important contributions to the variability of the ice edge, with standard deviations of~ 12 – 15%.
Despite common objections for using an ancillary data set for independent quality assessments, a synergistic SIC – SST retrieval appears to be beneficial for improving the accuracy of both SIC and SST products. Since both products have non-complementary sources of error, this synergism can be exploited for mutually identifying/removing residual noise in the other product. Moreover, SSTs for the de-noised SIC appear to have standard deviations ≈ 0, suggesting that they are not introducing trends of their own.
Report on this study: Sea surface temperature metrics for evaluating OSI SAF sea ice concentration products