Multi-objective Optimization Based Calibration of Hydrologic Model and Ensemble Hydrologic Forecast for Java Island, Indonesia

This study explores the benefits of multi-objective optimization of Variable Infiltration Capacity (VIC) model for five watersheds in Java, the most populous island in Indonesia. Six objective functions: Nash Sutcliffe Efficiency (NSE), percent bias (PBIAS), logarithmic function of root mean square error (Log-RMSE), predictive efficiency (Pe), percent errors in peak (PEP) and slope of flow duration curve error (SFDCE) were selected as evaluation metrics. These metrics were optimized by tuning four VIC model parameters: infiltration shape parameter (b), fraction of maximum baseflow where nonlinear baseflow begin (Ds), thickness of soil layer 2 (thick2) and thickness of soil layer 3 (thick3). We employed Borg Multiobjective Evolutionary Algorithm (Borg MOEA), an automatic simulation-optimization algorithm, to search for non-dominated solutions. We identified the redundancy between NSE and Log-RMSE, Pe, and PEP through visual inspection of their sensitivity to parameters b and Ds of VIC model and to baseflow index (BFI). Accordingly, we proposed NSE, PBIAS and SFDCE as critical objective functions to represent hydrologic processes in tropical region of Java, Indonesia. Using these three objective functions, we culled the objective functions based on at least - NSE > 0.75, PBIAS < 15% and SFDCE < 15% - and showed that the number of optimized objective functions as well as model parameters in the ensemble are reduced but the value ranges are maintained. We used the culled model parameters, to run the VIC model using an ensemble of conditioned seasonal climate forecast to generate an ensemble streamflow prediction in the period 2001 - 2010, the time window when the seasonal climate forecasts and observed streamflow records overlaps. We measured the skill of this seasonal forecast by computing the rank probability skill score (RPSS) of seasonal total flows and extremes at three different thresholds, for the dry and wet seasons. We showed that the RPSS of seasonal flows and the extremes are very good for both seasons. This study, for the first time, demonstrates the utility of the multiobjective based calibration of hydrologic model in tropical regions and its applications in generating skillful seasonal ensemble hydrologic forecasts which are important for short and long term water resources planning and management.