Parameter Estimation of a Distributed Hydrological Model Using the Adjoint Method: A Case Study in the Ibo River Watershed, Japan

Japan is currently faced with the trend of increasing frequency of short but heavy rains and the number of dry days due to climate change. The distributed hydrological model is versatile especially when the spatial variation of precipitation, geography, or others has a large impact on simulation results. The model, however, has many parameters that must be set and modified to improve its skill. An adjoint model of a distributed model was developed in this study to achieve a better simulation and obtain spatial distributions of the model parameters. The method was applied for a heavy rain and a large river runoff in the Ibo River basin, Japan on July 2015. Moreover, tangent linear and adjoint models were employed for estimating the spatial distributions of five parameters by the variational assimilation of observed river discharge. Specifically, after calculations that started from January 1 for the setup of initial conditions, the hourly observed river discharge was assimilated for 6 days from 00:00 on July 15 to 24:00 on July 20 in 2015. All assimilation cases delayed the runoff time to follow the observed pattern of the runoff. In the case of the estimation of Manning’s roughness coefficient in river channels, the impact of estimation reached 20 km upstream from the gauge station. All estimated parameters varied spatially to obtain a more accurate state of the discharge and delay the flooding.