Evaluating the performance of eight global gridded precipitation datasets across Iran

Abstract

Listen

Translate article

Precipitation is one of the crucial variables in hydrological and water resources studies. Runoff, soil moisture, groundwater recharge, and other hydrological variables change with the amount of precipitation in a region. Thus, accurate precipitation estimation is a major issue in the field of water resources and hydrological modeling. Most hydrological studies and water management decisions in Iran are established using in situ observation of precipitation. However, due to limited spatial distribution and lack of long-term high-quality data in many stations, this information usually does not fully meet the information requirements of a given study. In this study, the accuracies of eight global gridded precipitation datasets including Global Precipitation Climatology Centre (GPCC), Climatic Research Unit (CRU), Climate Prediction Center (CPC), Climate Hazards Group InfraRed Precipitation with Station (CHIRPS), PERSIANN-Climate Data Record (PERSIANN-CDR), Global Precipitation Climatology Project (GPCP), Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA2), and ECMWF Reanalysis version 5 (ERA5) were evaluated to identify the strengths and weaknesses of each dataset in different regions and main catchments of Iran. The precipitation of 96 synoptic stations during the years 1987–2016 was used as the basis for the evaluations. The evaluations were made at monthly, seasonal, and annual scales using grid-based evaluation and areal average evaluation at the catchments scale. In addition to the amount of precipitation, the datasets were evaluated using three statistical performance indices (i.e., correlation coefficient, RMSE, and relative bias) and other metrics such as probability distribution similarity with observation data (using Anderson-darling test), empirical cumulative distribution function (ECDF), and Taylor diagrams. The results indicated that dataset performance varied by catchment. Except for central parts, GPCC was the best dataset in 44% of the evaluated grids in terms of time series pattern recognition. It showed correlation coefficients of 0.71, 0.80, and 0.85, in monthly, seasonal, and annual scales, respectively. In central areas, MERRA2 estimated the average annual precipitation in this catchment with + 2 mm difference, and CRU had a correlation coefficient greater than 0.90. Moreover, CHIRPS was the best dataset in terms of relative bias in the south and southwestern catchments, so that the relative bias was negligible in these catchments in most months. This study provides rankings and recommendations for selecting an appropriate alternative precipitation dataset, which, in turn, provides the knowledge required in monitoring systems and hydrological and water resources modeling that need high-resolution input data.