OUTLIER DETECTION IN GRIDDED SHIP'S DATASETS

we discuss robust statistical methods to detect « local errors » in gridded ship’s datasets. More precisely, we attack the problem of outlying areal averages in gridded ship's datasets such as Comprehensive Ocean-Atmosphere Data Set (COADS; Woodruff et al. 1987) 2° lat × 2° long monthly summaries and how to test the statistical significance of them. Since the majority of climate researchers use gridded ship's datasets instead of individual ship reports, we suggest that these datasets must be checked for the presence of doubtful raw monthly means in the same manner as individual ship reports are quality controlled before being integrated in ship reports databases.