An Algorithm of Spatial Composition of Hourly Rainfall Fields for Improved High Rainfall Value Estimation

Abstract

This study proposes two variants of the traditional conditional merging (CM) method that merges the next-generation radar (NEXRAD) ground gauge precipitation data. The first method, named CM considering simple optimal estimation (SOE), employs a novel algorithm of simultaneously considering rainfall spatial intermittency and inner variability to replace the conventional semivariogram algorithms of the CM method. The second variant, called CM-GR-SOE, employs additional Ground-Radar rainfall ratio (so called the G/R ratio) to the CM-SOE method. Model performance was evaluated using the hourly rainfall data collected between 2004 and 2007 in the regions of Houston and Dallas in Texas. The leave-one-out cross-validation was conducted, and the relative mean error (RME) and coefficient of determination (R-2) were calculated for each of the methods. In areas where the rainfall intensity was low (<0.25 mm/h), NEXRAD Stage IV, and occasionally the CM method, showed lower absolute values of RME, and higher R-2 values than other variants. As rainfall intensity increased (greater than 7.6 mm/h), the CM-GR-SOE method showed the best performance. Further analysis revealed that spatial correlations of rainfall field is the primary source of seasonal variability of the model performance. The analysis also revealed that the correlation between the model seasonal performance and the rainfall spatial correlation depends on the density of ground gauges. For this reason, the CM-GR-SOE method performed better at the Dallas area.