Validation of ISCCP Cloud Detections

The International Satellite Cloud Climatology Project (ISCCP) began in 1983 to collect and analyze weather satellite datasets to produce a new global cloud climatology as part of the World Climate Research Programme. The first step of the analysis is detection of the presence of clouds at each location and time by a series of tests on the space / time variations of infrared and visible radiances. This paper describes the validation of the ISCCP cloud detections by verifying the accuracy of the inferred clear-sky radiance. Comparison of retrieved surface temperatures to other measurements shows that bias errors are <2 K and random errors are about 2 K for sea surface (monthly means at 280-km scales) and that bias errors are <2 K and random errors are about 4 K for land surfaces (3 hourly at 280-km scale). Bias errors over a few persistently cloudy locations are sometimes −(2–4) K and over winter sea ice may be about +2 K. Surface reflectances are confirmed to be within 3% of other measurements and models for ocean, except for sun glint geometries, and to be within 3%–5% for land surfaces. Sufficiently accurate validation data are not available for visible reflectances of sea ice and snow-covered land, but some tests of specific cases suggest that errors are ∼10%. These errors in clear-sky radiances suggest uncertainties in the ISCCP cloud detections of about 10% with a small (3%–6%) negative bias over land. Some specific regions exhibit both larger rms uncertainties and somewhat larger biases in cloud amount approaching 10%. ISCCP cloud detections are more in error over the polar regions than anywhere else. Based on comparisons with an analysis of radiances measured at other wavelengths, the ISCCP analysis appears to miss 15%–25% of the clouds in summer but only 5%–10% of the winter clouds.