Systematic assessment of GPM IMERG V06 precipitation products with dense rain gauge observations over Zhejiang Province, China

Abstract

This study systematically assessed the performance of the Integrated Multi‐satellitE Retrievals (IMERG) for Global Precipitation Measurement V06, including the near‐real‐time “Late Run” (IMERG‐L) and the post‐real‐time “Final Run” (IMERG‐F), over Zhejiang Province (ZJP), China. The evaluation was conducted at daily and hourly timescales for a full year and for each season, based on dense rain gauge observations and continuous and categorical validation statistics. For the full year and for each season, IMERG‐F outperformed IMERG‐L in representing the spatial pattern of multiyear mean precipitation. For regional mean of ZJP, IMERG‐F and IMERG‐L overestimated the daily/hourly precipitation for the full year by 6.51 and 4.98%, respectively. Among seasons, the regional mean relative biases for IMERG‐F were between 5.65 and 8.63%; however, for IMERG‐L, they exhibited notable variations with a maximum of 11.09% in fall and a minimum of 0 in spring. Bias composition suggested that the regional mean overestimations were largely due to false bias for the full year and for each season, except in winter, wherein it was due to hit bias. Spatially, the biases for the full year and for each season commonly arose from false and hit biases at daily timescale, and from false and miss biases at hourly timescale. Based on the remaining continuous metrics (i.e., root‐mean‐square‐error [RMSE], correlation coefficient [CC], and Kling–Gupta Efficiency [KGE]) and all categorical metrics, the IMERG daily/hourly performance was acceptable on regional and grid scales throughout the year and in all seasons. From a region‐average perspective, IMERG‐F outperformed IMERG‐L according to CC, RMSE, and KGE, but both products showed the same performance overall based on all categorical metrics; most grids also share these characteristics. This study provides a valuable reference for IMERG developers to improve product accuracy from the perspective of the final postprocessing step and for potential users in ZJP.

Bibliographic Information