Reliability and discrepancies of rainfall and temperatures from remote sensing and Brazilian ground weather stations

Abstract

Insufficient ground meteorological stations limit agricultural research in wide geographic areas, but high-quality data from remote sensing may decrease information gaps, when surface stations are scarce. This study compared meteorological datasets, estimated from satellite and ground meteorological stations in latitudes from 0 to 33 °S, to support agricultural research in Brazil. The dataset comprised 3600 records of monthly temperatures and rainfall from 01 Jan 2004 to 31 Dec 2014 in 30 Brazilian municipalities distributed in six regions, labeled according to their precipitation homogeneity. Climatic records from NASA's Prediction of Worldwide Energy Resource (POWER) online database were compared with data from Brazilian surface stations managed by National Institute of Meteorology (INMET). Monthly rainfall data showed satisfactory correlation coefficients for almost all locations, between 0.75 and 0.95 (p < 0.01), and simple linear models were fit for estimated (satellite) and observed (ground) rainfall relationship (p < 0.001). Complimentary accuracy and precision tests endorsed rainfall satellite-estimated data according to the root mean square error (RMSE) and the modified index of agreement. Maximum and minimum temperatures estimated by remote sensing in the Brazilian South Region were also statistically supported, but unsuitable results were found especially in lower latitudes, based on higher RMSE. The Pearson's correlation coefficient for temperatures increased proportionally with latitude, while rainfall did not show this correlation. These results showed satellite-data quality varies regionally and is affected by seasonal variation. Remote sensors may not detect extreme climatic events such as heavy rainfall or draught and, therefore, need to be appraised carefully.