Ground validation and bias correction of GPM-IMERG V6 satellite precipitation product over Sweden

Abstract: This study attempts to assess and correct the performance of Integrated Multi-satellite Retrievals for Global Precipitation Measurement (GPM-IMERG) final run daily precipitation product across Sweden. The performance of GPM-IMERG version 6 final run was evaluated against 677 rain gauges in a period from 12 March 2014 to 31 May 2019. Continuous and categorical performance measures were used to characterise different attributes of performance that included correlation coefficient (CC), root mean square error (RMSE), Nash-Sutcliffe efficiency (NSE), relative bias (rBias), modified relative bias (rBias_ε), false alarm detection (FAR), probability of detection (POD), critical success index (CSI), Heidke skill score (HSS), Kling-Gupta efficiency (KGE), and Willmott index of agreement (WIA). In addition to general evaluations, the impacts of temporal and spatial variability, elevation, and precipitation intensity on satellite performance were also studied. The results showed that CC, RMSE, and rBias is 0.70, 3.65 mm, and +13.65% (overestimation), respectively, with the best performance observed in spring and autumn followed by summer and winter. The performance of GPM-IMERG spatially varies: it overestimates precipitation in regions below 60°N, close to coastlines, and lowlands. The performance is also not consistent for different precipitation intensities. Precipitation events over 20 mm/day are substantially underestimated while light precipitation (< 1 mm) is overestimated. A framework for bias correction was developed that uses the results of ground-validation for recognising the driving factors on satellite bias and a Monte Carlo Cross Validation approach for calibrating bias correction models. Using the framework, bias correction models for each month and each precipitation intensity categories were developed to correct GPM-IMERG product. The corrected dataset shows overall improvement over entire domain, confirming the utility of the developed framework. This study could contribute to both practical use and further development of GPM-IMERG by providing an insight about the performance and the improved data over Sweden.