Automatic Calibration of Initial State Variables for Flood Forecasting Using a Distributed Model

Abstract The hydrological conceptual distributed model, called TETIS, divides river basins into regular cells, all of which are connected according to the network drainage. The rainfall-runoff process is modelled using linked tanks with different outflow relationships. The vertical movement is based on soil properties. Routing along the channel network has been coupled using geomorphologic basin characteristics and the kinematic wave procedure. The sensitivity analysis carried out using the Generalised Likelihood Uncertainty Estimation (GLUE) methodology demonstrates the importance of some of the initial state variables during flood events in the TETIS model. Therefore, these variables must be estimated. For a real-time automatic calibration of these initial values, the shuffled complex evolution (SCE-UA) algorithm was selected. The flood forecasting has been divided in two phases: the “warming period” which focuses on a period of time previous to forecasting using real-time data, and the “forecasting” phase. The main goal during the warming period is to estimate some initial state variables of the TETIS model using automatic calibration. In this way, it is possible to get a better accordance among observed and simulated discharges during the warming period and to increase the forecasting reliability. The results indicated that SCE-UA is robust and efficient. It was highlighted that calibrating initial state variables allowed adjusting properly the observed and simulated discharges during warming period. This warming period strategy can be coupled to a Quantitative Precipitation Forecast (QPF) scheme in order to produce better results and reduce the uncertainty.