COMPARISON BETWEEN CONVENTIONAL AND ANN-BASED FLOOD FORECASTING METHODS

Abstract

Flood forecasting models have been developed for many different large rivers. Traditionally they are based on regressions of downstream with upstream gages. Recently, more physically based models have been used which consist of an input like areal averaged rainfall, gauge levels or discharges at upstream gauges measured in real time and transferred to the regional forecasting centre, where they are converted by means of suitable hydrological and hydraulic models into future water levels and discharges at some critical points. Basis of most models is a numerical model for the rainfall - runoff process, whose parameters are fitted to meet local conditions. In the hydrological context, as in many other fields, artificial neural networks (ANN) are increasingly used as black-box, simplified models. For hydrological applications, ANN models can take advantage of their capability to reproduce the unknown relationship existing between a set of input variables descriptive of the system, for example, rainfall and set of output variables, for example, river flow rate. Mahanadi is a major river basin of eastern India comprising a catchment of 141569 sq. km. Its major part is occupied by Chhatisgarh and Odisha States. The basin lying in the South-West monsoon tract is exposed to myriad hydro-climatic variations. The reservoir Hirakud which drains nearly 83000 sq. km. is the prime flood moderation structure in the basin, and the catchment of about 50000 sq. km. remains unregulated. Both the Hirakud release and intercepted catchment cause the flood damage to about 9000 sq. km of delta. As such there is not a serious flooding problem existing in the upstream of Hirakud. However, the Mahanadi delta is threatened by floods every 2 or 3 years. Out of the 19 rainfall-generated floods observed during the post-Hirakud period, 13 were due to intercepted catchment contribution and 6 due to Hirakud dam releases. During these floods five coastal districts were affected and a great loss to life and property occurred. The option for another flood moderation structure is not justifiable due to several socio reasons. Presently, inflows to Hirakud are forecast by Central Water Commission, Govt. of India, using MIKE-11, NAM, and HD (Hydrodynamic) models. To this end, data from Automatic Weather Stations and satellite images are used.