Rainfall-runoff modelling in dryland catchments, Sauce Grande, Argentina

Resumen

The poor understanding of the hydrological functioning of many dryland catchments challenges hydrological modeling on both a discrete and a continuous basis. This paper implements a simple yet robust conceptual rainfall-runoff model, GR2M, to predict long-term monthly runoff in the Sauce Grande catchment (Argentina). It aims at determining whether (i) simple rainfall-runoff models perform satisfactorily on dryland catchments, and (ii) the parameter transfer from calibration to validation works in the context of climate-driven flow variability. Two model versions are evaluated and compared considering similar and contrasting catchment conditions along the period of record. Calibration results showed from 88 to 90 % efficiency on runoff predictions (on average), with variations along calibration periods linked to prevailing flow conditions (magnitude, variability, and constancy). From both, the model version separating the part of direct runoff from subsurface flow showed greater sensitivity to extreme flow conditions and greater structure adaptability to the full range of flows. Efficiency losses from calibration to validation were yet marked (22 %, on average), and responded primarily to runoff overestimations on periods of low flow. Parameters were allowed to evolve along with hydroclimatic conditions based on decision tree learning. Through this modification, the predictive efficiency of GR2M improved by 97 %. In addition to validating the robustness of simple rainfall-runoff models on drylands once parameters may evolve, this paper yields new hydrological data that constitutes an important platform to underpin further water resources planning and management in this highly regulated catchment.