Land Cover Classification from Very High-Resolution UAS Data for Flood Risk Mapping.

Area of Science:

• Environmental Monitoring
• Remote Sensing
• Disaster Risk Reduction

Background:

• Global increase in natural hazards necessitates effective disaster risk reduction tools.
• Accurate land cover (LC) data is crucial for assessing vulnerability and potential damage in hazard-prone regions.
• Rural areas, particularly those susceptible to flooding, require high-resolution mapping for targeted interventions.

Purpose of the Study:

• To generate a high-resolution land cover map of flood-prone rural villages in southwest Niger.
• To develop a detailed land cover classification system relevant for estimating flood-induced damages.
• To utilize multispectral drone imagery for advanced land cover mapping.

Main Methods:

• Two photogrammetric flights using fixed-wing unmanned aerial systems (UAS) with RGB and Near-Infrared (NIR) sensors.
• Structure from Motion (SfM) workflow to generate orthomosaics and a Digital Surface Model (DSM).
• Object-oriented supervised classification with a Random Forest (RF) classifier, incorporating textural and elevation features.

Main Results:

• Generated a land cover map with nine detailed classes, including houses and production areas.
• Achieved an F1_score of 0.70 and a median Jaccard index of 0.88 during segmentation.
• The Random Forest model demonstrated an overall accuracy of 0.94 for land cover classification.

Conclusions:

• High-resolution land cover mapping using drone imagery is feasible and effective for flood-prone areas.
• The developed classification system and methods provide valuable data for disaster risk assessment.
• Further refinement is needed for specific classes like grasslands and rocky areas to improve mapping accuracy.