Diagnostic calibration of a hydrological model in a mountain area by hydrograph partitioning

Abstract

Hydrological modeling can exploit informative signatures extracted from long time sequences of observed streamflow for parameter calibration and model diagnosis. In this study we explore the diagnostic potential of hydrograph partitioning for model calibration in mountain areas, where meltwater from snow and glaciers is an important source for river runoff (in addition to rainwater). We propose an index-based method to partition the hydrograph according to dominant runoff water sources, and a diagnostic approach to calibrate a mountain hydrological model. First, by accounting for the seasonal variability of precipitation and the altitudinal variability of temperature and snow/glacier coverage, we develop a set of indices to indicate the daily status of runoff generation from each type of water source (i.e., glacier meltwater, snow meltwater, rainwater, and groundwater). Second, these indices are used to partition a hydrograph into four parts associated with four different combinations of dominant water sources (i.e., groundwater, groundwater + snow meltwater, groundwater + snow meltwater + glacier meltwater, and groundwater + snow meltwater + glacier meltwater + rainwater). Third, the hydrological model parameters are grouped by the associated runoff sources, and each group is calibrated to match the corresponding hydrograph partition in a stepwise and iterative manner. Similar to use of the regime curve to diagnose seasonality of streamflow, the hydrograph partitioning curve based on a dominant runoff water source (more briefly called the partitioning curve, not necessarily continuous) can serve as a diagnostic signature that helps relate model performance to model components. The proposed methods are demonstrated via application of a semi-distributed hydrological model (THREW, Tsinghua Representative Elementary Watershed) to the Tailan River basin (TRB) (1324 km2) in the Tianshan Mountains of China. Results show that the proposed calibration approach performed reasonably well. Cross-validation and comparison to an automatic calibration method indicated its robustness.