Variable Infiltration-Capacity Model Sensitivity, Parameter Uncertainty, and Data Augmentation for the Diyala River Basin in Iraq

Abstract

To construct a valid model, hydrologists face challenges in determining sensitivity to the forcing data and uncertainty in model parameters. These require basin data and forcing data from different sources, which may be incommensurate. The study reported here calibrated the Variable Infiltration-Capacity (VIC) platform to quantify model result sensitivity to model parameters and uncertainty in those parameters. The modeled basin was the Diyala River in Iraq, above the Derbendikhan Dam. The study produced the first complete set of daily forcing data for the basin using different sources. Besides ground observations from the Iraqi Ministry of Water Resources, two additional data sources were tested: Tropical Rainfall Measurement Mission (TRMM) and Global Implemented Data (GIDAL). Several methods were implemented to adjust the data, and model sensitivity and parameter uncertainty were examined by Generalized Likelihood Uncertainty Estimation (GLUE) and Differential Evolution Adaptive Metropolis (DREAM). Neither of these techniques had been applied before in Iraq. The VIC model was then calibrated manually using Kling–Gupta efficiency (KGE). The analyses indicate that neither TRMM nor GIDAL data are adequate for gridded precipitation analysis in the study basin. TRMM tends to underestimate and GIDAL tends to overestimate actual data. Multiplicative random cascade and Schaake Shuffle were used to determine daily precipitation data. A set of correction equations was developed to adjust GIDAL temperature and wind speed. Results for the GLUE and DREAM analyses imply that the depth of the second soil layer is the parameter that causes the most sensitivity in the model. The VIC model outputs were calibrated on a daily timescale with a KGE average of 0.743.