RELIABILITY OF CONVENTIONAL FLOOD FORECASTING

Abstract

There are a number of complex catchment models involving very detailed functions, but such models generally have elaborate requirement for the input data and parameters. All the desired input data for a complex catchment model are, in many cases, not available on real time basis. Availability of data on real time basis is to be given due consideration in the choice of the model [CWC,1989]. In this context, reliability of conventional flood forecasting techniques has been analyzed in the present study. Conventional methods of flood forecasting as used in India consist of: (i) Methods based on statistical approach: These consist of development and use of empirical relations (usually linear) between upstream and downstream river gauge, travel time, peak discharge etc. (ii) Methods based on formation and propagation of flood: These consist of analysis of effective rainfall (direct runoff) and application of linear unit hydrograph and channel routing techniques. Usually a combination of statistical approach and linear unit hydrograph theory is adopted for flood forecasting in India. Statistical approach is illustrated for Sone basin and combined approach is illustrated for Baitarani basin. (i) Linear regression analysis of peak flood gauges to forecast peak flood gauge at Koelwar (G) from known peak flood gauge at Japlan (G) on river Sone. ii) Linear and non-linear regression analysis for forecasting travel time of flood from Japlan to Koelwar site on river Sone. Non-linear model has lesser Bias and Variance and is therefore recommended for use. However the model can be made more reliable by checking the data particularly observed gauges during nighttime. Runoff production function study of Baitarani basin is given in Chapter-3. Direct runoff was obtained by judiciously separating base flow from observed 28 flood hydrographs. The direct runoff was related to areal average storm rainfall by; iii Reliability of Conventiona I Flood Forecasting (i) Linear regression analysis (ii) Non-linear regression analysis and (iii) Linear multiple equation between direct runoff (Y) in mm and two independent variables areal average rainfall (X) in mm and base flow (Y) in cumecs. It is recommended that Multiple Linear Regression may be used to estimate direct runoff (excess rainfall) for the Baitarani basin. Four different Unit Hydrographs are obtained for Baitarani using different methods. Peak discharge as well as time to peak differs significantly. Errors in forecasted flood in terms of estimation of peak discharge and time to peak due to use of different Unit Hydrograph when rainfall is not uniform and when rainfall assumed to be uniform over the catchment have been worked out. Reliability in terms of Coefficient of determination (R2), Bias (B) and Variance (V) have been computed for following forecasting models, which have been studied in previous Chapters. (i) Gauge forecast using linear regression equation (ii) Travel time forecasting using linear and non-linear regression equation (iii) Collin's Unit Hydrograph (complete shape, rising portion only) with reference to average UH (from Isolated flood, IUH and Synthetic Unit Hydrograph) (iv) Synthetic Unit Hydrograph (complete shape, rising portion only with reference to average Unit Hydrograph (from Isolated flood, IUH and Coffin's Unit Hydrograph). A procedure based on auto regression has been suggested for updating of flood forecast in Baitarani basin. Lag one auto regression model has been used to update forecast three hours ahead. Accuracy of forecast with updating technique improves over forecast without updating as demonstrated in the case study. The study demonstrates that conventional methods of flood forecasting could be made more reliable by making these computers based for development of appropriate models.