In this paper, we present the potential of generalization for a RescueNet data-trained damage estimation model towards evaluating natural disasters which occurred in December 2024 to a French island, namely Mayotte. Indeed, more and more disasters occur over the world and affect people’s lives; notably because of climate change. Intelligent disaster evaluation models require a huge amount of representative disaster data in order to be trained efficiently. Indeed, after a massive disaster, aerial images are usually collected in order to evaluate the disaster impact at a large scale. We propose a transfer learning approach that exploits publicly available massively collected disaster data (coming from the Hurricane Michael) towards estimating the damage severity over another disaster, namely Mayotte (Cyclone Chido). In particular, data preparation and adaptation stages are presented towards permitting an estimation of damages. The generalizability potential of an efficient RescueNet model is evaluated by comparing its damage predictions over Mayotte image datasets. Both the damage evaluation architecture and the prepared Mayotte datasets have been publicly made available at: https://github.com/iyed-dhahri/Evaluating_Mayotte_disaster_damages .