Species Distribution Models

Climate change and human activity are rapidly transforming forest ecosystems, reshaping where tree species can survive and thrive. Shifts in temperature, precipitation, and disturbance regimes are already altering species distributions across North America, with far-reaching consequences for biodiversity, ecosystem resilience, and forest management. As natural regeneration becomes increasingly influenced by future climate conditions, there is a growing need to identify which species are best suited for the landscapes of tomorrow.

This work begins with understanding the environmental conditions that support species today. The models developed in this project focus on habitat suitability—not just observed presence—allowing for a more flexible and forward-looking approach to predicting where species could thrive under future scenarios. These models can support a wide range of applications: guiding seed collection zones, delineating ecosystem boundaries, informing assisted migration strategies (as demonstrated in this web tool developed by our lab), and, when combined with remotely sensed land cover data, enhancing the accuracy of large-scale forest inventory efforts.

Deep Nueral Networks

Traditional methods of forecasting species distributions often rely on static ecosystem classifications or outdated range maps, which can lead to inaccuracies in a rapidly changing climate. These approaches struggle to account for the complex interactions between climate, topography, and species ecology, especially at fine spatial scales. As a result, they may misinform conservation priorities or regeneration strategies in both ecological restoration and commercial forestry contexts.

Machine learning approaches, particularly deep learning methods, offer powerful alternatives for modeling species distributions. Deep Neural Networks (DNNs) can learn complex, nonlinear relationships between environmental variables and species presence, capturing subtleties that traditional models may overlook (LeCun et al., 2015). By training these models on historical occurrence records and high-resolution environmental data, we can generate more accurate and scalable predictions of where tree species are likely to occur under current and future conditions.

By combining remote sensing, ecological field data, and deep learning, this work contributes to the growing field of climate-informed forest modeling. It provides a tool to support adaptive forest management, helping practitioners select the most appropriate species for productive forestry and restoration efforts in the face of ongoing environmental change.