The validation of the GEWEX SRB surface longwave flux data products using BSRN measurements

Abstract

The longwave downward fluxes at the Earth׳s surface are a significant part of the products of the NASA GEWEX SRB (Surface Radiation Budget) project which has produced and archived a 24.5-year continuous record from July 1983 to December 2007 of global shortwave (SW) and longwave (LW) radiation fluxes at TOA and the surface from satellite measurements. The data are generated on a system of grid boxes ranging from 1° latitude by 1° longitude at lower latitudes to 1° latitude by 120° longitude next to the poles. The LW datasets, which are available as 3-hourly, 3-hourly–monthly, daily and monthly means, are produced from two sets of algorithms, the GEWEX LW (GLW) algorithm which is designated as primary and the Langley Parameterized LW (LPLA) algorithm which is designated as quality-check. The inputs of the latest versions, GLW (V3.1) and LPLA (V3.0), include the Geostationary Satellite system (GEOS) Version 4.0.3 meteorological information and cloud properties derived from the International Satellite Cloud Climatology Project (ISCCP) DX data. In this paper, we compare the LW downward fluxes at the Earth׳s surface from both algorithms against over 4000 site-months of the Baseline Surface Radiation Network (BSRN) data from among the 59 BSRN sites. The comparisons are made for the 3-hourly, daily and monthly means each for the entire record, and on a month-by-month basis as well as a site-by-site basis. It is found that the overall daily mean bias/RMS for the GLW (V3.1) and LPLA (V3.0) algorithms are, respectively, 1.1/22.1 and 4.6/22.8Wm−2, their monthly counterparts are, respectively, 0.9/11.1 and 4.5/12.9Wm−2. Anomaly time series for a subset of more continuous BSRN measurement data sets show a standard deviation of 2.3Wm−2 and a correlation of 0.82 indicating the accurate replication of month-to-month variability. Clusters of similar surface types are analyzed showing that the uncertainties are largest over the polar regions. Finally, Kolmogorov–Smirnov (KS) two-sample test and Cramér–von Mises (CvM) two-sample test are used to show that the GLW is able to replicate the cumulative frequency distribution of the measurements at the 0.01 significance level.