Adaptive neuro-fuzzy inference system for flood forecasting in a large river system

Abstract

Large river systems provide essential water resources and economic benefits for a significant number of people. However, large floods that occur less frequently (e.g., 25 to 100-year floods) can cause significant loss of lives and property. Therefore, flow forecasting, which aims to provide accurate forecasts of flow and/or level of a river is an essential tool for flood plain management and flood mitigation purposes. New approaches in computational intelligence tools are now being applied to build effective forecasting models. This research investigated the application of the data driven modeling approach for flood forecasting in the Lower Mekong, which was chosen as a case study. The Adaptive Neuron-Fuzzy Inference System or ANFIS was chosen for this purpose. ANFIS is a modern approach in Computational Intelligence which combines the learning and reasoning capabilities of the Artificial Neural Network and Fuzzy Inference System techniques, respectively. In order to develop a flood forecasting model based on ANFIS, it was important to clarify the link between hydraulic/hydrologic factors (such as flood travel times and the catchment response to rainfall) and ANFIS model performance, as this would provide important information for model input selection. River flows at various stations along a large river can vary in terms of the flow contributions from upstream stations, rainfall and lateral inflows from sub-basins immediately upstream of the station of interest. Therefore, the present study provided an analysis on how these factors influence ANFIS model performance by developing a series of ANFIS models for three stations along the Lower Mekong where the flows at the stations selected differed in terms of the contributions by flow from upstream stations along the main river, laterals and rainfall from the local sub-basin. In such a scenario, the choice of input variables can significantly affect ANFIS model performance. The results obtain showed that the inclusion of water levels from the station of interest and the immediate upstream station were sufficient for 1-and 3lead day water level forecast. The addition of rainfall information beside water levels was not necessary. The ANFIS models developed in the study performed well for 1-and 3-lead-day water level forecasts when compared to the benchmark adopted by Mekong River Commission (MRC). In addition, ANFIS model results CHAPTER 1: INTRODUCTION 1 CHAPTER 1 INTRODUCTION 1.1. Flood Forecasting for Large River Systems Floods in rivers are natural phenomena that occur when water levels in river channels lead to overspill of natural banks or artificial embankments and subsequently inundate the surrounding flood plains (White, 2000). River floods usually occur after heavy rainfalls over river basins during the wet season. This causes increased runoff into upstream reaches of rivers, and when downstream water levels are increased to very high levels, the rivers may overspill the natural banks or artificial embankments or even break dike systems at some locations, re...

Application of Artificial Neural Network and Climate Indices to Drought Forecasting in South-Central Vietnam

“…is a data-driven model such as ANN; IC i is formed by a number of input candidate pools X n (t -1, t -2,..., t -i) where X n generally indicates previous observation of SPEI, rainfall or CIs, and t -i indicates lagged time of the above-mentioned variables. We selected i = 3 as maximum lagged time for local candidates [31] and i = 12 as maximum lagged time for CIs [32].…”

Drought index prediction using advanced fuzzy logic model: Regional case study over Kumaon in India

“…The outcome of the study found the ANN model capable of predicting the SPEI and SPI over the considered area. Nguyen et al [11] focused on the prediction of drought at short-term and long term basis in Cai river basin, Vietnam using the ANFIS model; the prediction was based on SPI & SPEI. From the result, SPI and SPEI were found useful in predicting short-and long-term drought, respectively.…”

Comparing Machine-Learning Models for Drought Forecasting in Vietnam’s Cai River Basin

“…Due to the impact of drought, an effective and timely monitoring system is required, the effective monitoring of drought can aid in the development of an early warning system [17]. At present a lot of models have been accepted as efficient tools for the modeling of complex hydrologic systems widely employed for forecasting, such as artificial neural networks (ANNs) and autoregressive moving average approach (ARMA) [18][19][20].…”