Wildfires have intensified in frequency, scale, and destructiveness due to accelerating climate change, posing critical threats to ecosystems, biodiversity, and human societies. Rising global temperatures, shifting precipitation patterns, and prolonged droughts have created increasingly fire-prone environments across many regions. At the same time, wildfire-driven habitat loss is becoming a major factor in species endangerment and extinction. In this context, there is an urgent need for accurate, spatially explicit wildfire prediction models that can inform ecological conservation strategies under future climate scenarios. This project proposes to develop a deep learning framework based on convolutional neural networks (CNNs) to predict wildfire burned areas using multi-source spatiotemporal data. Our approach integrates historical wildfire burned area records (GFED4), high-resolution climate reanalysis datasets (e.g., ERA5), vegetation indices (such as NDVI and LAI), and land cover classifications. The CNN model will be trained to learn complex spatial-temporal patterns in past fire events and generalize them to predict future fire-prone areas under climate projections from scenarios such as SSP2-4.5 and SSP5-8.5. Beyond fire prediction, this research uniquely focuses on assessing the ecological consequences of increasing wildfire activity. By overlaying predicted burned areas with species habitat maps from databases like the IUCN Red List, we aim to identify biodiversity hotspots at high risk of future fire exposure. This will allow us to quantify potential species loss and prioritize regions for conservation intervention. The integration of AI-based forecasting with biodiversity impact analysis represents an innovative step toward more holistic, data-driven environmental planning. The expected outcomes of this project include: (1) a robust CNN-based wildfire prediction model trained on diverse environmental datasets and (2) future wildfire burned area datasets under different climate scenarios. These results will support policy decisions in climate adaptation, fire management, and biodiversity conservation. To realize this work, we request access to GPU-enabled high-performance computing resources. Training deep neural networks on large-scale spatial-temporal datasets requires extensive computational power to achieve reliable accuracy and efficiency. With adequate GPU resources, this research will contribute to a deeper understanding of climate-driven wildfire dynamics and their ecological consequences, ultimately informing long-term strategies to mitigate biodiversity loss.