Inter-comparison of high and low microwave frequency sea ice concentration algorithms
In the context of the OSI SAF Visiting Scientist Program, Carolina Gabarro from the BEC/ Institut de Ciències del Mar (ICM - CSIC) in Barcelona, worked on the inter-comparison of high and low microwave frequency sea ice concentration algorithms during winter and autumn. This work took place in from November 2017 to October 2018 and was supervised by Rasmus T. Tonboe (DMI) and Thomas Lavergne (MetNO).
Objectives and framework of the study:
The objective of this associated scientist project is to inter-compare the sea ice concentration from microwave radiometers, for winter and autumn period, at low microwave frequency to current OSI SAF sea ice concentration products based on AMSR data. New algorithms recently developed in the ESA CCI project (Clime Change Initiative) and published in the OSI-450 CDR has been considered also in the study.
Files from the RRDP SICCI-RRDP-V1_1 have been used for year 2007-2011, and 2013- 2015, which uses AMSR family instruments.
An analysis on random errors of the SIC products on closed ice regions (selected from SAR images) for several models has been performed. Figure 1 show the bias and std of the analysed SIC product (ground truth is 100%). We can observe that LTP637, SICCI2-50km, LowFreq and the Emis6.9V are the models with better stability (less STD) during winter for regions with high concentration of sea ice. An important observation is that the models with lowest STD use the low frequency bands (6 or 10GHz), which present more precise Tb values as already stated in the conclusions of the AVS- 16-03.
The systematic errors (biases) of different SIC algorithms for pixels with high sea ice concentration have also been estimated. AMSR-2 TB maps at different bands have been used for winter and autumn 2014. The figure 2 maps show the biases present during the whole winter period with respect 100%.
The algorithms using low frequency bands (figures 2.a and 2.b) show much lower bias and also very low temporal STD, except the SMOS product. Among all, the LowFreq (6V, 10V, and 18V GHz) and SICCI2 (6V and 37H, V) SIC maps show less systematic error and also less temporal variability in high SIC regions.
Algorithms based on the 19/37GHz frequency (Bootstrap F, Bristol) show high positive bias in the thick ice regions in the Canada Basin. See figure 2.c and 2.d.
Algorithms based on the polarization only (Bootstrap P, TwoChan) show higher systematic error in the coast, ice edge, and thin ice regions (Russian coast). TwoChan seems to delineate information on the ice type characterization at 10 GHz. See figure 2.e.
Correlations between SIC products and snow-ice interface temperature obtained from IMB buoys and with snow depth and SIT from OIB measurements have been assessed. Results are not conclusive, since they vary with the time period, and correlations are not high enough (lower than 0.5).
With all this, we conclude that the SICCI2 is the algorithm showing better performances from the analyzed algorithms, but one should take into account that this model is the only one from the analysed algorithms which uses dynamic tie points. Furthermore, we emphasize that the algorithms using low frequency bands are the ones showing less random and systematic SIC errors in regions of high SIC values during winter.
Benefits for the SAF: The activity is also of interest to the OSI SAF climate data records (CDR) and (ICDR): OSI-409/OSI-409-a, OSI-430 and OSI-450.
Report on this study : Inter-comparison of high and low microwave frequency sea ice concentration algorithms