Deep Learning Integration of Remote Sensing Data for Threatened Mammal Conservation

Abstract

Introduction: Threatened mammals face challenges related to habitat destruction, forest fragmentation, lack of water, wildfires, agricultural encroachment, and human disturbances. Field investigations are key to providing biological data, but they cannot constantly monitor environmental changes over large areas that are difficult to access. Materials and Methods: This paper introduces an ecological evaluation framework based on deep learning that leverages satellite data, land cover information, vegetation indices, surface water data, terrain data, and verified mammalian presence. In a proof-of-concept dataset, there are 14,400 spatial records of mammals in forest, grassland, and wetland habitats. Ecological variables along with convolutional spatial features were used to identify suitability and threat level classification. Results and Discussion: The integrated model has an accuracy of 95.2%, precision of 94.7%, recall of 94.1%, and F1-score of 94.4%. Vegetation depletion, increasing distance to water, closeness to roads, and decreasing connectivity of habitat were the key factors of habitat threat. As a result of analysis, the habitat was classified into stable, transitional and fragmented habitats. Conclusion: The use of remote sensing in combination with field verification will increase the effectiveness of threatened mammal protection through early habitat change monitoring, selection of priority areas for surveys, corridors preservation, and adaptive management