A Progressive Segmented Optimization Algorithm for Calibrating Time-Variant Parameters of the Snowmelt Runoff Model (SRM)

To capture the temporal variability of parameters of hydrological models, the segmented optimization algorithm (SOA) is usually used which subdivides the calibration period into a number of sub-periods and seeks optimal parameters for each sub-period by optimizing the objective function based on the measured and estimated data in the same sub-period. In this paper, we developed a new method that is called a progressive segmented optimization algorithm (PSOA), which seeks optimal parameters by optimizing the objective function based on both the current and all the prior sub-periods. We applied and compared the SOA and PSOA algorithms to the Snowmelt Runoff Model (SRM) in simulating snow-melt streamflow for the Manasi River basin, northwest of China, during snowmelt seasons of 2001-2012. The study showed: (1) PSOA can effectively calibrate the time-variant model parameters while avoiding too much computational time caused by a significant increase of parameter dimensionality. (2) PSOA outperforms SOA for both single-snowmelt-season and multi-snowmelt-season simulations. (3) For single-snowmelt-season simulation, the length of the sub-period has an apparent effect on model performance, the shorter the sub-period is, the better the model performance will be, when the model is calibrated using the PSOA method. (4) For multi-snowmelt-season simulation, an over-short sub-period may cause overfitting problems in some cases such as the situation of taking Nash-Sutcliffe efficiency (NSE) as the objective function. A compromised length of sub-period and objective function may have to be chosen as a trade-off among evaluation criteria and between the importance of calibration and validation.