Evaluation of model-derived and remotely sensed precipitation products for continental South America

This paper investigates the reliability of some of the more important remotely sensed daily precipitation products available for South America as a precursor to the possible implementation of a South America Land Data Assimilation System. Precipitation data fields calculated as 6 hour predictions by the CPTEC Eta model and three different satellite-derived estimates of precipitation (Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), National Environmental Satellite, Data and Information Service (NESDIS), and Tropical Rainfall Measuring Mission (TRMM)) are compared with the available observations of daily total rainfall across South America. To make this comparison, the threat score, fractional-covered area, and relative volumetric bias of the model-calculated and remotely sensed estimates are computed for the year 2000. The results show that the Eta model-calculated data and the NESDIS product capture the area without precipitation within the domain reasonably well, while the TRMM and PERSIANN products tend to underestimate the area without precipitation and to heavily overestimate the area with a small amount of precipitation. In terms of precipitation amount the NESDIS product significantly overestimates and the TRMM product significantly underestimates precipitation, while the Eta model-calculated data and PERSIANN product broadly match the domain average observations. However, both tend to bias the zonal location of precipitation more heavily toward the equator than the observations. In general, the Eta model-calculated data outperform the several remotely sensed data products currently available and evaluated in the present study.