Water quality monitoring based on multiple remote sensing imageries

Remote sensing imageries are widely applied in the field of water quality monitoring since they have the features of wide detection range and fast data acquisition. Water quality monitoring based on remote sensing imagery takes the semi-empirical method mostly. However, the resolution of the most remote sensing imageries processed by the semi-empirical method is relatively low. The semi-empirical method needs a lot of manpower for data collection. Meanwhile it has limits in time and space since it cannot process every single image from different imaging conditions adaptively. In this paper, a novel approach of integrating high spatial resolution imageries and hyperspectral imageries (HSI) to monitor water quality is presented. According to multispectral high spatial resolution image, a one class support vector data description (SVDD) classifier is first trained. Pixels with anomalous spectrum are detected with this SVDD classifier. These pixels show that there exists pollution. Abnormal pixels corresponding to HSI imagery are then uncovered by mapping the abnormal pixels in high spatial resolution imagery to HSI imagery of neighboring date. To retrieve different pollution indexes of water quality, two bands of abnormal pixel in HSI data are selected with feature selection method. Finally, different linear regression models are built with appropriate combinations of feature bands to retrieve pH, dissolved oxygen (DO), permanganate index and ammonia concentration. Experiment on the GF-1 Wide Field of View (WFV) imageries and HJ-1A HSI imageries of Lanshanzui Region in the western bank of Taihu has showed that water pollution condition can be detected automatically and accurately by applying our method.