Hybrid change detection for watershed impervious surface using multi-time remotely sensed data

In this paper, an approach to quantify basin impervious surfaces as an input variable for hydrological model was proposed, in which a hybrid change detection method and decision tree classifier based on data mining algorithm was employed using multi-temporal Landsat TM/ETM images in 1988, 1994 and 2002 at the same season in the lower reach of Yangtze River. The change types for 1994-2002 and 1988-1994 were extracted and validated with overlay analysis of GIS. The experimental results were shown that the overall classification accuracy is 88.1% compared with 69.3% of MLC in 2002 for six watershed types, and detection accuracy for five change types was 89.1% and 91.4% respectively for 1994-2002 and 1988-1994. It is demonstrated that the proposed approach is of capability for the change detecting, and can be achieved better accuracy at 30m resolution for distributed hydrological models with multi- temporal data.