Intercomparison of Climate Data Sets as a Measure of Observational Uncertainty

摘要

Uncertainties in climate observations are revealed when alternate observationally based data sets are compared. General circulation model-based 'reanalyses' of meteorological observations will yield different results from different models, even if identical sets of raw unanalyzed data form their starting points. We have examined 25 longitude-latitude fields (including selected levels for three-dimensional quantities) encompassing atmospheric climate variables for which the PCMDI observational data base contains two or more high-quality sources. For the most part we compare ECMWF with NCEP reanalysis. In some cases, we compare in situ and/or satellite-derived data with reanalysis. To obtain an overview of the differences for all 25 fields, we use a graphical technique developed for climate model diagnosis: a 'portrait diagram' displaying root-mean-square differences between the alternate data sources. With a few exceptions (arising from the requirement that RMS differences be normalized to accommodate different units of variables) the portrait diagrams indicate areas of agreement and disagreement that can be confirmed by examining traditional graphics such as zonal mean plots. In accord with conventional wisdom, the greatest agreement between alternate data sets - hence the smallest implied observational uncertainty - occurs for upper tropospheric zonal wind. We also find fairly good agreement between reanalysis and more direct measures of precipitation, suggesting that modern observational systems are resolving some longstanding problems with its measurement.