Trimming Problems and Remedies in COADS

Abstract

The Comprehensive Ocean?Atmosphere Data Set (COADS) offers the most complete marine surface data collection (1854?1992) currently available for global climate research. In the context of the long-term monitoring of global climate, COADS constitutes an important in situ baseline dataset, considerably extending the presently available decade-long satellite-based datasets. Marine observations of a given ocean location originate from a variable mixture of vessels with different instrumentations, resulting in nontrivial sampling errors. Therefore, careful quality control (QC) is paramount. The final portion of COADS QC?flagging statistical outliers and removing them from the computation of areal averages?is referred to as ?trimming? and is of particular concern here. This paper reviews evidence showing that the trimming of COADS in Release 1 and interim releases was unnecessarily restrictive. Trimming not only eliminated obvious outliers but also a relatively large fraction of realistic observations, where climate anomalies were large and available observations were few, especially common for tropical Pacific sea surface temperatures (SST). In poorly sampled regions, trimming may thus have removed the most prominent signal, the ?bull?s-eye,? of climate anomalies. For instance, during the strong El Niño events of 1877?78 and 1982?83, up to two-thirds (one-third) of central Pacific SST observations were erroneously excluded from the trimmed record, reducing the peak trimmed SST anomaly by up to 1°C. Excessive trimming occurred often enough for all observed variables to prompt a review of trimming procedures for COADS Release 1a (1980?92). In its enhanced version, trimming limits were moderately widened (by nearly 30%). This appears to have reduced erroneous trimming losses by over 50%, without noticeably increasing the number of unwanted statistical outliers. Given the present state of knowledge about COADS QC problems, the paper includes recommendations for the optimal use of the COADS climate record.