Deep Learning-Based Modeling of Amphibian Spatial Distribution for Assessing the Effectiveness of Protected Areas

The gap analysis based on species distribution modeling (SDMs) is a essential tool for evaluating the performance of Protected Areas and identifying conservation gaps. Improving the accuracy and reliability of SDMs can significantly aid in implementing strategies under the Global Biodiversity Framework. This study aimed to enhance gap analysis methods through deep learning-based modeling and to overcome some limitations of traditional gap analysis approaches by using a fuzzy approach to calculate habitat suitability, ultimately assessing the efficacy of gap analysis in identifying conservation gaps. The study employed the CNN-SDM deep learning-based method to identify complex relationships between species' spatial distributions and environmental conditions. WorldClim data and records of 45 amphibian species from the GBIF in Southwest Asia were used to calibrate the model. After generating habitat suitability maps, species were weighted according to their threat levels in the IUCN Red List, creating a weighted species richness map. To address the limitations of binary data, a fuzzy approach was used to represent habitat quality on a continuous scale from 0 to 1. Future predictions were made using the SSP126 and SSP585 climatic scenarios for the years 2060 and 2100. The comparison of weighted species richness maps and stacked richness revealed that 43% of threatened amphibian habitats were overlooked by the stack binary method. Results confirmed that under the worst climate scenario, while 85% of suitable habitats remain within protected areas, the average pixel value of habitat suitability within these areas significantly decreases, with fuzzy habitat suitability declining for 97% of the species. The CNN-SDM model demonstrated high accuracy, with an AUC of 0.94 and a TSS of 0.89. This study highlights the need for enhancing gap analysis methods and addressing the limitations of traditional approaches