Flood Prediction Modelling

Abstract

Natural Disasters results in great damage or loss of life. Flooding has been one of the most costly natural disaster and results in property damage, causalities, and environmental degradations. Prediction of water is of great interest for flood control and water management. The need for flood prediction is still increasing. Artificial intelligence and knowledge discovery advances offer techniques for the modelling and simulation of such complex phenomenon. In this project an attempt has been made to predict the discharge of a river using different data driven techniques based on historical datasets of rainfall, temperature and evaporation and discharge. Different training parameters were used to achieve the best result. Principal component analysis technique is used to reduce the number of inputs. Two approaches are used in this project, Artificial Neural Network (ANN) and Support Vector Machine (SVM). Data for training and testing purpose of these approaches is obtained from Pakistan Meteorological Department (PMD) for the years 1990-2007.Principal component analysis was also performed to find the importance of each input. Mean squared error was used to evaluate the performance of each approach. 75% of data was used for training purpose and 25% of the data was used for testing purpose. The approaches use some variables like rainfall, temperature and evaporation which require in-depth knowledge of hydrological properties but advantage of these techniques required such knowledge. The findings show that Artificial Neural networks can be used to extract information from data and can be used to predict the river discharge. Comparative study of different data driven techniques is also presented. The results enables the weather forecast centers and flood managers and regional weather survey centers for better decision making and benefits public with the aim of finding the risk of flood. This approach gives the acceptable output for the neural network architecture.