Comments on “Reanalyses Suitable for Characterizing Long-Term Trends”

Abstract

recent article by Thorne and Vose (2010, hereafter TV) concerns the use of reanalysis data for characterizing long-term climate trends. The article raises legitimate questions about the ability to extract accurate climate information from time-varying observational datasets by means of model-based data assimilation techniques. TV predict that current approaches adopted by producers of reanalysis data are unlikely to result in climate-quality datasets. They make several recommendations for improving reanalysis methodology, and also propose a definition of “climate-quality” that is based on robust accuracy requirements. The purpose of this note is to explain our views on these issues and to address some of TV’s specific recommendations. To derive accurate and complete representations of climate variability and trends from observations is an ambitious goal. The difficulties involved, of course, are not specific to reanalysis. Fundamentally, the climate system is incompletely and inaccurately observed; data coverage, measurement techniques, and associated uncertainties are continually changing. Any such change can generate or modulate systematic errors in estimates of climate parameters. The idea behind reanalysis is to try to combine the observations by making optimal use of all available information. This includes metadata pertaining to data quality, as well as information about the physics of the climate system that can help us interpret and compare different pieces of data. In this way it becomes possible to expose the underlying uncertainties and to reduce their impact on the representation of climate parameters.