Multispectral imaging (MSI) is an efficient tool for obtaining and processing spectral information for remote sensing. It is desirable to be able to monitor the effect of flash floods; by using the proper spectral wavelengths we can observe the presence and aftermath of flash floods. The data we are examining is satellite data from the Negev in Israel. We used set of Sentinel-2 images before and after a recorded flash flood in Paran wadi. Sentinel-2 images include 12 spectral bands that cover the region of visible, Near infra-red and short-wave infra-red. Sentinel-2 revisit time of five days is not enough to capture all flash-floods because of the short duration of flash floods in arid regions. Therefore, an attempt to capture the soil moisture signal after the flood was examined.
In our project, we tested two potential solution approaches: matched-filter-based flood detection, and temporal change detection. The latter was based on the RX algorithm, along with Hyperbolic Change Detection.
Results showed that the matched-filter approach could not distinguish between wet and dry soil in the flood channel and detected the area on dry and on wet days alike. Also, the RX algorithm alone was not encouraging. However, combining it with hyperbolic change detection, we were able to obtain unambiguously precise detections.
To conclude, the combined approach was able to detect temporal changes which directly correlated with the flooding. A further study needs to be done on the matched filter approach.