Constructing sampling and measurement error models for ICOADS SST from ships based on ESA CCI SST analysis

Short abstract

Taking advantage of high resolution, reliable uncertainty estimates, and in situ data independence of daily fields of the European Space Agency (ESA) Climate Change Initiative (CCI) SST Analysis product (hereinafter, CCI SST), versions 1, errors of 1^ox1^o  monthly bin averages of ship SST observations from International Comprehensive Ocean-Atmosphere Data Set (ICOADS), Release 3.0, were modelled as a sum of random effects, once their systematic biases were approximated and removed as their climatologically-averaged differences from similarly binned CCI SST.  For 1992-2010 period, in more than 66% (50%) of locations with temporal coverage exceeding 50%(66%) for 1^ox1^o monthly bins containing more than one observation, the error magnitude agrees within 20%(10%) with the estimates, based on the random error model.  These error estimates were also split into sampling and measurement error components. Seasonal variations in the total error magnitude were traced to the sampling error component, which is driven by seasonal changes in the intra-bin SST variability, while the seasonality of measurement error estimates appears not significant (by Levene’s test for variance).  Random measurement error estimates for different measurement methods used on ICOADS ships compared well with previously published estimates. Improved error estimates were constructed by recombining all-season measurement error estimates with sampling error estimates based on the full data sample from the CCI SST data set.