Possibilities and Limitations of IMERG Datasets for Estimating Probable Maximum Precipitation

Abstract

The probable maximum precipitation (PMP), which is conventionally derived based on precipitation gauge data, is an important input for hydrological models useful for accurate storm predictions, flood modeling, and general decision making. However, precipitation data are not readily available in most regions due to limited and failing gauges. This study assessed the possibilities and limitations of deriving PMP using precipitation records from the Integrated Multi-satellite Retrievals for GPM (IMERG). PMPs for six durations were calculated with IMERG precipitation records (2000-2020) using Hershfield statistical technique built in R. The IMERG PMPs were further evaluated with NOAA-Atlas-14 station PMPs at 55 rain gauge locations in Kansas, USA, using coefficient of correlation, root mean square error, and relative bias as statistical metrics. Result of PMP evaluation for the six durations showed that, as durations increase from 30-mins to 24-hr, CC decreased from 0.956 to 0.854, RMSE increased from 4.41 mm to 10.28mm, and RB increased from -3.77% to 7.82%. RB estimates ranged between -3.77% to -7.96% at shorter durations (30-min to 12-hr) depicting underestimation in IMERG PMP. Relationship between precipitation amount and PMP errors at the six durations showed that increase durations reduces PMP error. At the 24-hr durations, the most R-squared were estimated at 51.5% and 50.3% for total accumulated, and maximum precipitation respectively. Further exploration showed that IMERG has an average percentage missing value of 74.21% with a 2-hr average length of missing value. The relationship between Missing values and PMP errors were statistically significant at the 30-mins durations with p-values of 0.00287 and 0.00127 for both percentage missing values and length of missing values respectively, while the longer intervals showed decreasing relationship between the variables with p-values of 0.665 and 0.4213 respectively. Overall assessment showed that IMERG estimate PMP better in wetter areas and longer intervals (e.g., 24-hr), while the estimated PMP errors may be influenced by missing values in IMERG datasets, with the greater influence at shorter durations (e.g., 30-min).