Evaluation of a distributed streamflow forecast model at multiple watershed scales

摘要

The demand for reliable estimates of streamflow has increased as society becomes more susceptible to climatic extremes such as droughts and flooding, especially at small scales where local population centers and infrastructure can be affected by rapidly occurring events. With critical hydrologic observation networks in decline worldwide, future expansion of existing networks into current ungauged locations seem unlikely. Spatially distributed models can help improve hydrologic predictions in ungauged basins because of their ability to model hydrologic processes at small scales, thus providing estimates at multiple subbasin locations. The Hydrology Laboratory-Research Distributed Hydrologic Model (HL-RDHM) is used to explore the accuracy of a distributed hydrologic model to simulate discharge at interior points representing various watershed scales. Basin sizes range from 20 – 2500 km2, with subbasins nested in three National Weather Service (NWS) forecast basins in the upper Midwest. The model is calibrated and validated using USGS observed discharge data at the basin outlets, and subbasin discharge is then evaluated. Two different precipitation products, NLDAS-2 with a nominal 12.5 km resolution and Stage IV with an approximate 4 km resolution, were tested to characterize the role of input uncertainty and resolution on the discharge simulations at the various scales. In general, across study basins, model performance decreased as basin size decreased, where correlation coefficients for NLDAS-2 and Stage IV simulations were 0.65 and 0.04, respectively. Once basin area was less than 250 km2 or 30% of the total watershed area, model performance became unreliable. Nash-Sutcliffe efficiency (NSE) scores were highest using the NLDAS-2 product, where basin outlets ranged from 0.50 to 0.75 during calibration and subbasins less than 250 km2 ranged from 0.11 to 0.40. Subbasins located further away from the watershed outlet had an increased chance of poorer model performance, especially for the Stage IV product (correlation = 0.35). The lower resolution NLDAS-2 data tended to improve discharge simulations during the verification period based on NSE and Percent bias (Pbias) scores compared to the higher resolution Stage IV. However, simulated discharge using Stage IV performed better for low flow periods leading to better Mean Absolute Error (MAE) scores, but the relative influence of errors versus spatial scale was difficult to characterize.