Natural disaster is one of the major global issues, and it is an important premise for disaster prevention and reduction to monitor it. In the present paper, multi-temporal HJ-1 images pre- and posto-typhoon Morakot were used. First, radiometric calibration and registration were done, then, decorrelation stretch (DS) was applied, and finally, maximum likelihood classification (MLC) was adopted to extract water body and monitor change of water body caused by typhoon Morakot. The results show that after DS spectral enhancement, the correlations among bands decrease and spectral differences increase, and it is helpful to identify surface features. The accuracy assessment demonstrates that the overall accuracies and Kappa coefficients of four phases are higher, above 96.0% and 0.94 respectively, than that of direct MLC without DS spectral enhancement. It is beneficial to scheduling flood discharge and ensuring the safety of reservoir downstream by comparing extracted multi-temporal water bodies.