Evaluation of Surface Runoff Prediction by A nn AGNPS Model in a Large Mediterranean Watershed Covered by Olive Groves

The distributed parameter and continuous simulation Annualized Agricultural Non‐Point Source (AnnAGNPS) model was implemented in the watershed Anzur (Spain) covered by olive groves, to assess its prediction capability of surface runoff at the large watershed scale in semi‐arid conditions. A 5‐year database reporting hydrological, geomorphological and land use characteristics of the watershed allowed model implementation. Almost 180 surface runoff events were modelled by AnnAGNPS and compared with the corresponding observations through statistical indexes and grouping the runoff events in order to evaluate the model at different temporal scales (event, monthly and seasonal). AnnAGNPS evaluation showed that, in general, runoff was estimated by the default model with low accuracy at all the investigated time scales, likely, as a result of a simple representation of spatial variability. Calibration (by reducing initial curve numbers (CN) of the olive groves) provided more accurate and satisfactory predictions of event, monthly and seasonal runoff volumes with a low effort in the parameterisation approach. The best model performance was achieved at the event scale. The runoff prediction reliability may be attributable to the AnnAGNPS inaccuracy in adjusting CN values during the continuous simulation of the soil moisture conditions, because estimations of daily evapotranspiration values are quite realistic. Copyright © 2015 John Wiley & Sons, Ltd.