Assessment of lumped hydrological balance models in peninsular Spain

Abstract. The assessment of inflows in a water resources system is essential for the appropriate analysis of its management. These inflows can be obtained from hydrological balance models. In this paper, we intend to perform a comparative study of six lumped hydrological balance models in several basins with different climatic conditions within Spain. Lumped models enable the estimation of catchment resources without using spatially distributed information that would not be available in many cases. We have selected basins where long time series of climatic and hydrological data are available (more than 30 years) to calibrate parsimonious models, taking into account the stochastic behaviour of the natural streamflow and the climatic variables. The study period comprises 34 years (1977–2010). The explored models are Témez, ABCD, GR2M, the Australian water balance model (AWBM), GUO-5 parameters (Guo-5p) and Thornthwaite-Mather. Six statistical indices are applied to compare the results of the models: Nash–Sutcliffe model efficiency coefficient (NSE), root-mean-square deviation (RMSE), Pearson’s correlation coefficient (R), percent bias (PBIAS), RMSE-observations standard deviation ratio (RSR) and the relative error between observed and simulated runoff volumes (REV). The results show that although lumped models can be employed in humid and sub-humid regions, the more humid the catchments are, the better the results obtained. Témez models provide the worst results in dry sub-humid and semi-arid regions. Guo-5p estimates runoff volumes with errors below 10 % despite the unsatisfactory results provided according to the Bressiani classification. The Bressiani classification takes into account different comparison criteria to help in the decision-making process when selecting a model. Nevertheless, the assessment of the margin of error in total runoff volume using REV is also a key index. The usefulness of Pearson’s correlation when selecting a model is quite low but can be helpful in the analysis of models’ weaknesses.