Chapter 10 - Short-term flood forecasting using artificial neural networks, extreme learning machines, and M5 model tree

Abstract

Flood forecasting is the estimation of future water levels or flows at a single or multiple sites of a river system for different lead times. Precise and reliable forecasting is important for flood warning, flood control planning, and rehabilitation. Of the several models available for flood forecasting, soft computing technique–based models are often better in terms of accuracy and reliability for operational flood forecasting systems, irrespective of data scarcity issues. The artificial neural network (ANN), a soft computing technique, has seen wide applications in flood forecasting studies. Recently, extreme learning machines (ELMs) and M5 model trees have also gained popularity in hydrological forecasting as improved artificial intelligence approaches that require significantly less computational time than the classical ANN model. This chapter provides an overview of the ANN, ELM, and M5 model tree and presents a case study where all three models are used to forecast floods at 1-, 5-, and 10-h lead times. A comparative analysis is carried out among different approaches applied for model development. For 1- and 5-h lead time forecasting, performance of the ELM and M5 model tree is better than that of the ANN model, whereas for 10-h lead flood forecasting, performance of all three models is quite similar.