MiTREE: Multi-input Transformer Ecoregion Encoder for Species Distribution Modelling

Climate change poses an extreme threat to biodiversity, making it imperative to efficiently model species' habitats, movements, and ranges for effective conservation planning. The availability of large-scale remote sensing images and environmental data has facilitated the use of machine learning in Species Distribution Models (SDMs). The aim of SDMs is, for any spatial location of interest, to be able to predict the bird species that will be present. Previous models either do not leverage the relationship between environmental data and satellite imagery or do not account for differences in resolution between images from various sources. Additionally, location information and ecological characteristics at the location play a crucial role in predicting species distribution models, but these aspects have not yet been incorporated into state-of-The-Art approaches. We introduce MiTREE: A multi-input vision-Transformer-based model with an ecoregion encoder that embeds the ecological classification, and subsequently the location, of the region into the representation. We evaluate our model on the SatBird Summer and Winter datasets, in which the goal is to predict bird species encounter rates, and find that our approach improves upon state-of-The-Art baselines.