Study On The Real-time Flood Forecast By Watershed Hydrological Model Coupled With Kalman Filter Technique

Real-time flood forecast is a prediction for future flood events or characteristics by adopting watershed hydrological model or methods on the basis of real-time hydrological and meteorological data. Modern real-time flood forecast usually combines with real-time calibration techniques, then the forecast system could correct or update model structures and parameters in the process of flood prediction with newly input information, therefore, the forecasted result could be close to the real value and in accord with the practicality step by step. Besides obtaining exact or reliable real-time hydrological and meteorological data in time, the improvement of flood forecast accuracy also depends on the flood forecast model, which could describe the rule of rainfall-runoff generation and movement of flood wave exactly, and effective calibration methods.Watershed hydrological models coupled with kalman-filter technique to set up a real-time flood forecast system is studied in the article. Two watershed hydrological models, i.e. a hydrological model considering spatial distribution of rainfall and TOPMODEL with comparatively well-developed theory are both applied to forecast flood. The former mainly concerns on the spatial variation of rainfall in runoff yield. Based on the joint spatial probability distribution of rainfall and infiltration, Horton model was adopted to compute the surface runoff and water infiltration. Part of the infiltration water supplement to the soil moisture, and the rest is taken as the ground water, which was computed by water storage curve of soil moisture based on the theory of saturated runoff modeling. TOPMODEL takes spatial variation of topography as its main structure, describes water flow movement by topographic information, and simulates surface and groundwater flow on the basis of varied area of runoff yield. A lag time function is adopted to simulate overland flow, while the matrix solution of Muskingum method is utilized to simulate river routing, which could constructa state equation to link the whole forecast system, and additionally, provide a computation platform for the import of real-time calibration technique. The formal kalman filter method is selected for calibration. Model noise and measure noise could be estimated on-line. According to the real input of water stage or discharge, the routing parameters could be updated, thus the accuracy of prediction could be improved.This real-time forecast system was applied to the flood forecast for the watershed above lushi hydrological station located on Luohe River, a branch of Yellow River. Results indicate that, these two hydrological models are applicable to the studied semi-arid or semi-humid region, and with the calibration by kalman filter technique, the forecast accuracy could be improved appropriately.