Multispectral remote sensing inversion for city landscape water eutrophication based on Genetic Algorithm-Support Vector Machine

Abstract

Eutrophication has become the primary water quality issue for many urban landscape waters in the world. It is a focus in this paper which analyzes Enhanced Thematic Mapper images and quality observation data for 12 consecutive years in 20 parts of the urban landscape water in Xi'an City, China. A water quality model for urban landscape water based on Support Vector Machine (SVM) was established. Based on in situ monitoring data, the model is compared with water quality retrieving methods of multiple regression and back propagation neural network. Results show that the Genetic Algorithm-SVM (GA-SVM) method has better prediction accuracy than the inversion results of the neural network and the traditional statistical regression method. In short, GA-SVM provides a new method for remote sensing monitoring of urban water eutrophication and has more accurate predictions in inversion results [such as chlorophyll a (Chl-a)] in the Xi'an area. Additionally, remote sensing results highly agreed with in situ monitoring data, indicating that the technology is effective and less costly than in situ monitoring. The technology also can be used to evaluate large lake eutrophication.