Performance of CHIRPS dataset for monthly and annual rainfall-indices in Northern Argentina

Abstract

In this study, we analyze the performance of CHIRPS in comparison with data from 50 rain gauges (OBS) in Northern Argentina (NA) for the 1982–2019 period. The methodology consists in a point-to-pixel comparison using the correlation coefficient (RHO), the mean relative error (MRE), the mean absolute error (MAE) and the Nash-Sutcliffe efficiency (NSE). We analyze monthly rainfall, annual rainfall indices and their trends, differentiating between anchor and no-anchor stations. A comparative analysis is performed further between two NA sub-regions: Northwestern Argentina (NWA) and Northeastern Argentina (NEA). Results indicate that CHIRPS dataset better represents the interannual variability in wetter (drier) months in NWA (NEA). For all months, RHO values are higher in NEA than NWA. For annual rainfall indices, RHO values in most of the stations of NA are non-significant or low for some number-days (with threshold of 1 mm) and very extreme indices, with the exception of the eastern extreme of NA. The less extreme indices (PRCPTOT, R95pad and R99pad) are observed to have higher RHO values (> 0.5 in all cases) in NA, as well as better MRE, MAE and NSE values. Monthly values and annual indices are underestimated in general, especially in NWA no-anchor stations. Most of the significant linear trends observed in rainfall indices are not detected with CHIRPS. As an exception, a relatively better performance for the maximum number of consecutive dry days (CDD) is observed in the sense that CHIRPS detect the positive linear trends in NWA but do not locate them with precision in comparison with OBS data. CHIRPS is not recommended for studies in NA related with the aspects (mean values, interannual variability, linear trends) of rainfall analyzed in this work, especially for the extreme rainfall.