Effects of input discretization, model complexity, and calibration strategy on model performance in a data‐scarce glacierized catchment in Central Asia

Glacierized high‐mountainous catchments are often the water towers for downstream region, and modeling these remote areas are often the only available tool for the assessment of water resources availability. Nevertheless, data scarcity affects different aspects of hydrological modeling in such mountainous glacierized basins. On the example of poorly gauged glacierized catchment in Central Asia, we examined the effects of input discretization, model complexity, and calibration strategy on model performance. The study was conducted with the GSM‐Socont model driven with climatic input from the corrected High Asia Reanalysis data set of two different discretizations. We analyze the effects of the use of long‐term glacier volume loss, snow cover images, and interior runoff as an additional calibration data. In glacierized catchments with winter accumulation type, where the transformation of precipitation into runoff is mainly controlled by snow and glacier melt processes, the spatial discretization of precipitation tends to have less impact on simulated runoff than a correct prediction of the integral precipitation volume. Increasing model complexity by using spatially distributed input or semidistributed parameters values does not increase model performance in the Gunt catchment, as the more complex model tends to be more sensitive to errors in the input data set. In our case, better model performance and quantification of the flow components can be achieved by additional calibration data, rather than by using a more distributed model parameters. However, a semidistributed model better predicts the spatial patterns of snow accumulation and provides more plausible runoff predictions at the interior sites. Key Points In glacierized data‐scarce regions increasing model and input complexity not necessarily lead to better performance Long‐term glacier mass balance modeled with GlabTop is proposed for multiple data set calibration Calibration data are more important than model or input complexity for model performance