Comprehensive Evaluation of Multi-Radar Multi-Sensor Precipitation Products with a Distributed Hydrologic Model

Chengmin Hsu (1), Rob Cifelli (2), Lynn E. Johnson (3), Robert J. Zamora (2)

Abstract
Distributed hydrologic models are ideal tools for simulating hydrologic systems at high resolutions as they possess the ability to characterize the spatial variability of forcing and the progression of horizontal flux occurring in the real world. Using the Hydrologic Laboratory – Research Distributed Hydrologic Model (HL-RDHM) as the assessment tool, this study focuses on comprehensively evaluating four Multi-Radar Multi-Sensor (MRMS) precipitation products (“Gauge-Only”, “Radar-Only”, “Radar with Gauge Correction”, & “Radar with VPR and Gauge Correction”). The study was executed for the three sub-watersheds in the Russian and Napa River basins (4949 km2) in California, USA. The initial MRMS data comparisons show that the Radar with Gauge Correction products, with a bias of -5.54%, matched best with the rain gauge observations, while the other three MRMS products (“Gauge-Only”, “Radar-Only”, & “Radar with VPR and Gauge Correction”) underestimated precipitation by 12.72%, 39.22%, and 8.23% respectively. Four parameters of the Sacramento Soil Moisture Accounting (SAC) module and two of the routing parameters were then calibrated based on the USGS streamflow observations. With the HL-RDHM calibrated, both the “Radar with Gauge Correction” and “Radar with Gauge and VPR Correction” products exhibited the most optimal hydrologic skill with average Nash–Sutcliffe coefficients (NSCs) 0.84 and 0.86, respectively, for the seven evaluation events, while the Radar-Only data produced the worst NSCs. For the sub-watersheds far away from the Radar station, the excellence of the “Radar with VPR and Gauge Correction” products in simulating hydrologic responses became especially apparent. Nonetheless, the “Radar-Only” products could produce very efficient hydrologic skills when the model parameters were recalibrated according to them. Three spatial statistics indices (“spatial autocorrelation”, “directional distribution”, and “grey level co-occurrence matrix”) were then employed to comprehend what causes this contradict result to occur and how the spatial patterns are altered between the original “Radar-Only” and the other MRMS products. The characterization of the relationships between the alteration of MRMS precipitation distribution patterns and the hydrologic simulation performance is vital information for the future effort of improving radar rectification methods. In this project, the impacts of simulation resolutions on the hydrologic skill were also evaluated. Comparing with the models at the HRAP resolution (~4.1-km at the Russian River basin), high-resolution simulations (~1-km) ordinarily result in better discharge replications, especially at sub-watersheds with complex topography. The aptitude of distributed models for simulating the hydrologic system of complex terrain at high spatial resolutions is thus highlighted. In summary, the strengths of the HL-RDHM in capturing precipitation spatial variations and simulating hydrologic responses at high resolutions confirmed the high quality of the MRMS precipitation products.