Evaluation of a watershed scale forest hydrologic model

A watershed scale hydrologic model (DRAINWAT) for drained forested lands was developed by coupling DRAINLOB, a field scale forestry version of DRAINMOD and the ditch and channel routing model section of FLD and STRM. The simulation model was tested with 5 years (1988-92) of data collected on a 340 ha watershed located near Beaufort in eastern North Carolina, Testing of the model included comparison of observed and simulated daily, monthly, and annual outflows and hourly event hydrographs by three different evapotranspiration (ET) methods. Two of which (Teskey form and GS_HR form) are based on the Penman-Monteith method and the third one on the Thornthwaite method. The average absolute deviation in observed and predicted daily outflows for a 5 year period was 0.94 mm day(-1), when the Penman-Monteith methods were used to predict ET. The average absolute deviation in cumulative outflow when ET was predicted by the Thornthwaite method was, respectively, 23% and 50% higher compared with the values obtained with both forms of the Penman-Monteith method. Based on coefficient of determination (R-2), coefficient of efficiency (E), and root mean square error (RMSE), Teskey and GS_HR forms of the Penman-Monteith method performed better than the Thornthwaite method in predicting both daily and monthly outflows, However, the average daily deviations by all three methods were not significantly different at 5% level, Prediction errors in simulating monthly outflows were reduced compared with daily outflows, The predicted mean annual outflow volumes when the GS_HR and Thornthwaite methods were used for ET were in closest agreement with observed data. Statistics showed that errors resulting from use of the Thornthwaite method, with correction factors, were usually within acceptable limits given the large input data required by the Penman-Monteith ET methods, Model prediction of event hydrographs was satisfactory based on different statistical and graphical comparisons. Deviations in predicted and observed results are attributed to errors in both. Errors in the measured outflows occurred for some larger events due to weir submergence. Errors in the simulations resulted from errors in rainfall inputs, and from uncertainties in drainable porosity, hydraulic conductivity and estimates of ET due to a number of factors including approximations of leaf area index (LAI) and stomatal conductance parameters. The model performance as a whole was satisfactory given the complexity of the model, limitations of input data for the watershed, measurement errors in outflow and rainfall, and the fact that the model was not calibrated. (C) 1997 Elsevier Science B.V.