REAL-TIME WATER LEVEL PREDICTION USING ADJOINT SENSITIVITY ANALYSIS (2)

Abstract

For real-time prediction of river water level, a numerical method of adjoint sensitivity analysis in river flow simulation was studied. First, a numerical method of river water level prediction with the one-dimensional unsteady river flow model is presented. The model’s initial condition, water level and discharge of all the model grids, is obtained by data assimilation using adjoint sensitivity analysis. Next, the model verification was carried out on the actual flood event of the Tama River. As a result, this method provided practically sufficient accuracy in 1 hour predicted water level prediction. Furthermore, the prediction performance was compared by the different prediction conditions of number of water gauge, assimilation time period and upstream boundary condition. As one of the results, this method was confirmed to be able to predict the water level within propagation time of flood even in the absence of the predicted discharge.