Reliable precipitation estimates are a crucial backbone for supporting hydrologic modeling and other geophysical applications. However, watersheds that extend across international boundaries or those that contain large bodies of water pose particular challenges to acquiring consistent and accurate precipitation estimates. The North American Great Lakes basin is characterized by both of these features, which has led to long-standing challenges to water budget analysis and hydrologic prediction. In order to provide optimal conditions for hydrologic model calibration, retrospective analyses, and real-time forecasting, this study comprehensively evaluates four gridded datasets over the Great Lakes basin, including the Analysis of Record for Calibration (AORC), Canadian Precipitation Analysis (CaPA), Multi-sensor Precipitation Estimate (MPE), and a merged CaPA-MPE data set, in which these products are analyzed at multiple spatial and temporal scales using station observations and a statistical water balance model. In comparison with gauge observations from the Global Historical Climatology Network Daily (GHCN-D), gridded datasets generally agree with ground observations, however the international border clearly delineates a decrease in gridded precipitation accuracy over the Canadian portion of the basin. Analysis reveals that rank in gridded precipitation accuracy differs for overland and overlake regions, and between colder and warmer months. Overall, the AORC has the lowest variance compared to gauge observations and has greater performance over temporal and spatial scales. While CaPA and AORC may better capture atmospheric dynamics between land and lake regions, comparison with a statistical water balance model suggests that AORC and MPE provide the best estimates of monthly overlake precipitation.

In this study we adapted WRF-Hydro to the Cape Fear River basin (CFRB) to assess its performance during Hurricane Florence (2018). The model was first calibrated with a strategy of mixture of automatic and manual calibration during Florence and then evaluated with an independent hurricane event. With satisfactory NSE values (>0.4) achieved at all gages for hourly simulation, the model demonstrates its potential in simulating the flood response at both basin and sub-basin scale during hurricane events. The model’s capability in reproducing rainfall and properly translating it to hydrological response was further evaluated. The analysis suggests that the calibrated WRF-Hydro in combination with a series of WRF simulation using different microphysics schemes can provide reasonable flood simulations. The model reproduced peak streamflow observed at gage stations with acceptable errors in timing and amplitude. Meanwhile, positive(negative) bias in rainfall input is likely to be amplified (reduced) in streamflow forecast when simulated rainfall volume is larger than the “model true”. And the timing bias mostly inherited from rainfall simulation and calibration process.