Sea surface temperature can be estimated from infrared satellite radiances. The NOAA-AVHRR radiometer has been used to produce SST fields during the last decades. The aim of this study is to examine, with Pathfinder Oceans Matchup Database, the SST accuracy deviations that occur when we use in situ data from different latitudinal and longitudinal bands to compute, by linear regression, the NLSST algorithm coefficients. We have restricted the in situ SST to 8
latitudinal bands and 18 longitudinal bands. Applying the resulting coefficients of the equatorial region (20N to 20S) and from the polar zone to the global case, we have found that the mean of residuals (SST in situ minus estimated SST) is greater than -0.6°C and the standard deviation near to 1°C in both cases. When we use the coefficients from longitudinal bands in global data set, the residuals show a bias lower than -0.1°C in about the 80% of cases. We conclude that the mean errors for the longitudinal algorithms are small when compared to the latitudinal, but in both cases have a substantial dependence on the SST in situ.