Evaluation of global impact models’ ability to reproduce runoff characteristics over the central United States

The central United States experiences a wide array of hydrological extremes, with the 1993, 2008, 2013, and 2014 flooding events and the 1988 and 2012 droughts representing some of the most recent extremes, and is an area where water availability is critical for agricultural production. This study aims to evaluate the ability of a set of global impact models (GIMs) from the Water Model Intercomparison Project to reproduce the regional hydrology of the central United States for the period 1963-2001. Hydrological indices describing annual daily maximum, medium and minimum flow, and their timing are extracted from both modeled daily runoff data by nine GIMs and from observed daily streamflow measured at 252 river gauges. We compare trend patterns for these indices, and their ability to capture runoff volume differences for the 1988 drought and 1993 flood. In addition, we use a subset of 128 gauges and corresponding grid cells to perform a detailed evaluation of the models on a gauge-to-grid cell basis. Results indicate that these GIMs capture the overall trends in high, medium, and low flows well. However, the models differ from observations with respect to the timing of high and medium flows. More specifically, GIMs that only include water balance tend to be closer to the observations than GIMs that also include the energy balance. In general, as it would be expected, the performance of the GIMs is the best when describing medium flows, as opposed to the two ends of the runoff spectrum. With regards to low flows, some of the GIMs have considerably large pools of zeros or low values in their time series, undermining their ability in capturing low flow characteristics and weakening the ensemble’s output. Overall, this study provides a valuable examination of the capability of GIMs to reproduce observed regional hydrology over a range of quantities for the central United States.