| description abstract | Radar-based quantitative precipitation estimate (QPE) serves as input for flood forecasting, and its importance gets magnified during catastrophic storms, e.g., Hurricane Harvey in 2017. The record-breaking rainfall from Hurricane Harvey covered vast spatial extents and lasted for a 5-day period, providing a unique chance for evaluating radar errors, especially their spatiotemporal dependence. Using the rainfall data of Hurricane Harvey, the authors utilize a new method for sampling ground-based rainfall measurements over radar pixels (i.e., spatial reference rainfall) based on subpixel rainfall variability. The new method aims to enlarge the sample size and allow for compressively evaluating the QPE. Two hourly QPE products, the Next Generation Weather Radar (NEXRAD) Stage IV and Multiradar Multisensor (MRMS), are chosen for the evaluation due to their roles in major flood forecasting activities; and a dense rain gauge network covering the whole of Harris County, Texas, provides the spatial rainfall reference in this analysis. Comparative analyses are conducted based on Hurricane Harvey and other two flood-inducing storms occurring in 2015 and 2016 over Harris County. The results imply that the Stage IV and MRMS overestimate and underestimate, respectively, the total rainfall by a small factor, while both QPEs tend to overestimate very light precipitation. In addition, the study suggests that the spatial correlation of radar error from both QPEs be described as powered exponential functions of interpixel distance. This study also includes hydrologic simulations for an urban watershed, demonstrating the importance of both the accuracy and spatial resolution of QPE in representing the mean areal precipitation (MAP) over catchments. The insight gained from this study provides guidance for further improving the QPE performance, and the new sampling approach for spatial reference rainfall can be applied to comprehensively evaluate long-term radar rainfall data. | |
