In this study, a model based on an artificial neural network (ANN) was developed to forecast the runoff of a meso‐scale, partly glaciated (40%), Alpine catchment area in the southernmost Andes in Patagonia, Chile. The study area is located in a maritime climate with a mean annual air temperature of + 5·7 °C and about 5500 mm of precipitation per year at sea level. The multilayer feed‐forward network is designed to make use of the Levenberg‐Marquardt algorithm to increase the speed of computation (convergence). Using climate data recorded at an automatic weather station nearby as well as water level records measured simultaneously, the ANN model was trained and verified using independent training and validation datasets. Parameters and the corresponding time lags were determined by statistical methods such as partial, cross‐ and autocorrelation. The results of the simulation confirm that the proposed model was able to identify the underlying non‐linear relationships between the input parameters and the observed discharge. The correlation during validation shows a significant correlation coefficient of 0·98, and an RMSE of 0·02 m respectively. However, it is almost impossible to decipher the internal behaviour of ANN due to its black‐box character. Nevertheless, valuable insights were gained in the complex input–output relationships, and the occurrence of dependencies between different input variables were detected using global sensitivity analysis (GSA). The results of the GSA were compared with those of multiple linear regression (MLR). While the performance of the ANN is much better than the MLR, both models return similar results in terms of the dependency of the discharge upon input variables. It was found that despite the large proportion of glaciated surface area within the catchment, discharge is mainly controlled by precipitation (49%). Furthermore, the runoff is slightly influenced by temperature (19%), global radiation (15%) and wind speed (16%). While the ANN proves to be a very efficient tool for simulating runoff in glacerized, Alpine catchments from meteorological data, the GSA method, as outlined and used in this paper, offers a useful approach of analysing ANN output. Copyright © 2008 John Wiley & Sons, Ltd.

EXPLOR_STEP=$WICRI_ROOT/Wicri/Rhénanie/explor/UnivTrevesV1/Data/Main/Curation

HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000E50 | SxmlIndent | more

HfdSelect -h $EXPLOR_AREA/Data/Main/Curation/biblio.hfd -nk 000E50 | SxmlIndent | more

{{Explor lien
   |wiki=    Wicri/Rhénanie
   |area=    UnivTrevesV1
   |flux=    Main
   |étape=   Curation
   |type=    RBID
   |clé=     ISTEX:014B387FB0B0B6539CB7D33E2211B7D981C42B81
   |texte=   Simulation and analysis of runoff from a partly glaciated meso‐scale catchment area in Patagonia using an artificial neural network
}}