Detection of flood zones using machine learning algorithms and remote sensing to determine the area of land under flood damage (case study: March 2019 flood in Aqqla city)
Year: 2024
Type: article
Abstract: Objective: The growth of settlements and the increase in human activities in floodplains, especially riverbanks and flood-prone areas, have increased the amount of damage caused by this hazard.Therefore, determining the extent of flooding in line with risk reduction, preparedness and response, and recovery and rehabilitation planning after this hazard is of great importance.Methods: Research Method: In the present study, common machine learning algorithms and Sentinel 2 image classification were used to produce land cover maps, detect flood zones, and determine the area of land damaged by the April 2019 flood in Aqqala County.Also, in order to examine and increase the accuracy of the algorithms, three normalized spectral indices of vegetation (NDVI), water areas (MNDWI), and built-up areas (NDBI) were combined with the images used.Different parameters of each algorithm were evaluated using spatial cross-validation to determine their effect on classification accuracy and to prevent optimistic results due to spatial correlation between the training and experimental samples.Results: The results showed that the combination of spectral indices led to an increase in the overall accuracy of the algorithms and in order to produce land cover maps, the random forest algorithm with an accuracy of 83.08% due to the use of the ensemble learning method has higher accuracy and generalizability than other support vector machine and neural network algorithms with an accuracy of 11.79 and 44.75%, respectively.After determining the most accurate algorithm, the floodplain map was produced using the random forest algorithm in two classes of irrigated and non-irrigated lands, and the overall accuracy of the algorithm was 40.93% in the most optimal parameters and with the combination of the spectral index (MNDWI).Then, by overlapping the land cover and floodplain maps, the area of built-up land, cropland, and green spaces under flood damage was estimated to be 2008.4and 0772.41 square kilometers, respectively.Conclusions: Based on the results obtained, the generated land cover and floodplain maps provide reliable results for estimating and understanding the effects of flood risk.
Authors: Omid Ashkriz, Fatemeh Falahati, Amir Garakani
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4408528837