Flood risk mapping using artificial intelligence “application to the east Algerian region”

The increased needs of the actors in land management mean that static maps no longer meet the requirements of scientists and decision-makers. Access is needed to the data, methods and tools to produce complex maps in response to the different stages of risk evaluation and response. The availability of high spatial resolution remote sensing data makes it possible to detect objects close to human size and, therefore, is of interest for studying anthropogenic activities. The development of new methods and knowledge using detailed spatial data, coupled with the use of Geographic Information System (GIS), naturally becomes beneficial to the risks analysis. Indeed, the extraction of information from specific processes, such as vegetation indices, can be used as variables such as water heights, flow velocities, flow rates and submersion to predict the potential consequences of a flood. The functionalities of GIS for cartographic overlay and make it possible to identify the flood zones according to the level of risk from the flood, thus making it a useful decision-making tool. This study was carried out on the territory of watersheds in the Annaba region, East of Algeria. The choice was guided by the availability of data (satellites images, maps, hydrology, etc.) and hydrological specificities (proximity to an urban area). The adopted model is divided into two parts. The first part is to establish a methodology for the preservation of wetland biodiversity and the protection of urban areas against floods. The second part of the model consisted of the integration of cadastral information with the flood risk map obtained in the first part of our research. In other words, through the use of remote sensing and machine learning algorithms, in particular decision trees and AdaBoost, we have generated a flood risk map for the specified catchment area of the Annaba region. The results showed that AdaBoost was satisfactory compared with the field reality and the most optimal model with an area under the curve (AUC) value of 0.90. While the decision tree had a value of 0.68. The findings of this study are used for planning and implementing flood mitigation strategies in the region.