The connection between runoff predicting models efficiency and basin aridity is investigated by using data from five representative basins in the mountainous central and northern Greece. Two substantially different models were selected, a monthly water balance model and an empirically fitted rainfall runoff model of the Kalman filter type. Both models were calibrated and validated in each basin and their overall efficiency was evaluated for each basin. The five basins were classified by using a humidity index and were found to range from humid to semi-arid. The efficiency of each model was then tested in terms of decreasing aridity and the two models exhibited a clearly contrasting behaviour; a decreasing efficiency of the Kalman filter model was observed, whereas the water balance model efficiency showed to increase with decreasing aridity. This behaviour is explained by considering the autoregressive characteristics of the corresponding five runoff time series, which are imparted in them by the filtering action of the basin. This factor becomes more important the more arid the basin is. Conclusively, the model of the Kalman filter type, designed to take account of the runoff autoregression, is shown to perform better with increasing aridity while in contrast, the water balance type of model where runoff memory is not accounted for, performs better the more humid the basin is, hence the runoff autoregressive characteristics become less significant.

This content is only available as a PDF.