FLOOD ESTIMATION AND FORECASTING IN MAHANADI RIVER BASIN USING SOFT COMPUTING TECHNIQUES

Abstract

Water is known as the most precious gift of nature for growth of civilization as well as a destructive element causing mass devastation. Flood hazards have become ever increasing natural disasters resulting in the highest economic damage among all kinds of natural disasters around the world. The country India is full of rivers and rainfall patterns are heavily influenced by monsoon. Thus occurrence of flood remains an inevitable feature in most parts of the country. The large river systems like Ganga, Brahmaputra, Godavari and Mahanadi influence the flood scenario of the country. Mahanadi is the 6l largest river system in India. The river is also known for its huge water potential and frequent flood devastations. Chhatisgarh and Orissa states of India cover almost 99% of the catchment area of Mahanadi basin. Currently a number of developmental projects are going on in these two states. For these projects well defined flood estimate formulae are required. The lower reach of Mahanadi basin is in the state of Orissa and flood is a permanent threat to this reach. Hirakud reservoir is the only major flood controlling structure in the basin. The downstream area of Hirakud is around 58000 km2. It remains uncontrolled and experiences frequent floods. Flood damages can be reduced drastically by adopting various non-structural measures such as flood frequency prediction and flood forecasting. In the present study efforts have been made to develop regional flood formulae for the entire Mahanadi basin using L-moment and prioritized variables based approach. For the lower reach of Mahanadi basin (downstream of Hirakud dam) flood forecasting models have been developed using soft computing techniques like ANN and Fuzzy logic. The performance of soft computing models has been compared with conventional and conceptual models. BROAD OBJECTIVES In the present study the flood problem of Mahanadi basin has been addressed by developing regional flood formulae for the uncontrolled portion of the basin and by developing a flood forecasting model using ANN and fuzzy logic for the lower reach. The objectives are summarized as follows: i. Development of regional flood formulae for Mahanadi basin. ii. Development of a flood forecasting model for the reach downstream of Hirakud, and iii. Development of a key raingauge network for Kantamal sub-basin of lower Mahanadi basin for flood forecasting. DEVELOPMENT OF REGIONAL FLOOD FORMULAE FOR MAHANADI BASIN Mahanadi basin has been divided into homogeneous regions by applying different clustering techniques, as the entire basin is not hydro-meteorologically homogeneous. Principal component analysis (PCA) has been used to prioritise the site characteristics. Clustering techniques like Hierarchical Clustering (HC), K-mean (KM), Fuzzy C-mean (FCM), Kohonen Self Organization Map (SOM) and Andrews Plot (AP) are applied on prioritised variables to verify the results of clustering. The entire basin is divided into two homogeneous clusters based on the results of Fuzzy C-mean (FCM) technique. L-moment based methods are used to test the homogeneity of the clusters and to identify a best fitting underlying frequency distribution. The Generalised Pareto (GP) distribution holds good for cluster-1 and it contains the areas which can contribute substantially towards runoff generation due to high slope and drainage density characteristics. The cluster-2 contains areas with low runoff generation capacity as compared to cluster-1. Generalised Extreme Value (GEV) is the robust distribution for this cluster. DEVELOPMENT OF A FLOOD FORECASTING MODEL FOR THE REACH DOWNSTREAM OF HIRAKUD In this reach the discharge data are available at 3 hour interval during monsoon period. However due to various reasons, sometimes only peak discharge data are available. Three hourly and peak discharge data of three different gauge and discharge (G&D) stations located downstream of Hirakud dam have been used to develop various flood forecasting models based on soft computing methods like Multi Layered Feed Forward-Artificial Neural Network (MLFF-ANN), Radial Basis Function-Artificial Neural Network (RBFANN) and Takagi Sugeno (TS)-Fuzzy inference system. The forecasting results of these models are compared with statistical and time lag methods exercised by the Department of Water Resources, Government of Orissa. The TS-fuzzy models perform better than other models for peak flood as well as 3-hour flood forecasting. The TS-fuzzy model ii gives an efficiency of 86.7% for a lead time of 42-hours. This is expected to improve the existing operational flood forecasting system quite significantly. DEVELOPMENT OF A KEY RAINGAUGE NETWORK FOR KANTAMAL SUBBASIN OF LOWER MAHANADI BASIN FOR FLOOD FORECASTING Sometimes it becomes difficult to collect data from all the rain gauges either due to instrumental disorder, difficulty in transmission, inability to take readings and many other operational difficulties during flood times. Therefore, the network of key raingauge stations is designed. The performance of key rain gauge network in flood forecasting is discussed and demonstrated through a case study of Kantamal sub-catchment of Mahanadi basin. This sub-catchment significantly contributes to the downstream floods at Khairmal, Barmul and Mundali. The Fuzzy logic applied on the network developed through AHP has shown the best result for flood forecasting at Kantamal gauge and discharge site with efficiency of 82.74% and RMSE value of500.2 m3/s for 1-day lead period forecast. Analytic Hierarchy Process (AHP) has been successfully introduced for the first time in this study for establishing the key rain gauge network. The contents of these have been presented in seven chapters namely (i) Introduction, (ii) Description of study area and data used, (iii) Review of literature, (iv) Regional flood frequency analysis of Mahanadi basin using prioritized variables, (v) Development of flood forecasting model for downstream of Hirakud reservoir, (vi) Rain gauge network design of Kantamal sub-basin of lower Mahanadi for flood forecasting, and (vii) Conclusions and scope for further work.