Abstract: For under-developed regions where the rivers have no or scarce hydrological gauging datasets, it is significant to explore the remote sensing techniques to determine the dynamic variation of up-/downstream water levels and to alert the potential flood inundation. In this work, the Nilwara Ganga in southern Sri Lanka, which is prone to floods, was taken as an example. A total of 14 scene Sentinel-1A SAR images from 2015 to 2017 were chosen to determine the up-/downstream flood peak levels. Based on the derived datasets, the prediction models of downstream flood peak levels were established, as well as the forecasting model on the maximum flood extent of the Nilwara Ganga. Consequently,the accuracy of the prediction model was evaluated, and an experiment of the predicted flood inundation was validated using 4 scene Sentinel-1A SAR images in 2018. The primary conclusions are summarized as follows:1) The fluctuation of the up-/downstream flood peak levels can be accurately and efficiently extracted by remote sensing technique;2) Among the established models, including quadratic polynomial, liner, power function, and exponential regression models, the exponential regression model under the ASTER GDEMV2 data is the optimal one, with R² of 0.79 and RMSE of 0.4, which means a consistent fluctuation between the upstream and downstream flood peak levels; 3) The validation results indicated that the overall accuracy of the predicted maximum flood extent is not less than 0.71. The method proposed in this paper aims to provide a new perspective for the flood early warning methodology using remote sensing techniques in the drainage area with less or no gauging datasets.

Key words: remote sensing technology, flood early warning, flood peak level, maximum flood extent