Multiple timescale assessment of wet season precipitation estimation over Taiwan using the PERSIANN family products

Abstract

This study evaluates the performance of five satellite precipitation products (SPPs) from the Precipitation Remotely Sensed Information using Artificial Neural Networks (PERSIANN) family for depicting precipitation changes in Taiwan over multiple timescales. Rain gauge data provided by the Central Weather Bureau (CWB) of Taiwan were used as a reference for evaluation, which focused on the wet seasons (May to October) during the period 2003–2019. All SPPs were found to have good ability in expressing the temporal phase changes over most of Taiwan on all the timescales examined, with significant temporal correlation coefficients (TCC) observed between the SPPs and the CWB data. We further evaluated the performance of the SPPs in making quantitative precipitation estimates based on the root mean square error (RMSE). For all examined timescales, the comparison between the best and worst performance shows greater normalized differences in quantitative estimates (i.e., RMSE) than in temporal phase depiction (i.e., TCC). In general, all SPPs tend to underestimate precipitation over most of Taiwan; however, two relatively new products (PDIR-Now and PERSIANN-CCS-CDR) have better RMSE performance than other SPPs on different timescales. PDIR-Now is the best product for quantitatively estimating precipitation on interannual, annual, and seasonal timescales, while PERSIANN-CCS-CDR is superior for daily and diurnal timescales. The findings also highlight that the performance of the PERSIANN-family in quantitatively estimating Taiwan precipitation does not depend primarily on the spatiotemporal resolution of SPPs, but may be related to the use of the cloud patch approach and the inclusion of weather station information in producing PDIR-Now and PERSIANN-CCS-CDR.