Simulated impacts of climate on hydrology can vary greatly as a function of the scale of the input data, model assumptions, and model structure. We chose three models that have been used to simulate current and future streamflow and to estimate the impacts of climate change on the water cycle in the Pacific Northwest, USA (PNW): the MC1 Dynamic Global Vegetation Model, the Regional Hydro-Ecologic Simulation System (RHESSys) model and the Variable Infiltration Capacity (VIC) model. To better understand the differences between the models representations of hydrological dynamics, we compared results between these three models and observed streamflow data for the HJ Andrews Experimental Forest (HJA) experimental forest in the Oregon’s western Cascades. To better characterize the hydrology and make comparisons between models, we calculated runoff and Nash-Sutcliffe model efficiency coefficients.
Results/Conclusions
We documented model needs (soil and climate inputs, initial conditions, spinup protocols, etc.), compared existing model results with available streamflow observations and documented model strengths and weaknesses as a way of providing the most relevant information for regional and local managers answering the next set of questions: at what scale are each of these models most relevant? Can the models inform each other and improve the reliability of their overall projections? Are there key differences between the models? Calculated runoff ratios and Nash-Sutcliffe model efficiency coefficients indicated all three models performed reasonably well, with a few exceptions, in simulating various parameters of the hydrological cycle but each could benefit from additional calibration and tuning. Timing and duration of flows were generally well portrayed by the three models as were seasonal low flows. All three models, however, were unable to accurately capture many of the highest flow events. Investigations utilizing modeled hydrology need to include detailed understanding and description of the model as all three models arrived at similar results based on largely dissimilar inputs and/or model processes.