Biodiversity loss is a pressing global crisis, and producing high-quality spatial data products at fine resolution is essential for large-scale conservation planning and decision-making. Our recent preprint demonstrates a novel deep learning segmentation approach that delivers continuous national-scale predictions of high biodiversity-conservation-value forests in Sweden at 10 m resolution with unprecedented accuracy, significantly exceeding existing methods and saving substantial field inventory effort.  In this project, we will extend and scale this deep learning framework to broader biodiversity modelling tasks that require processing millions of high-resolution raster data tiles. Using the Alvis GPU cluster, we will run repeated model training, validation, and large-scale inference workflows that are computationally intensive due to the volume of input data and the need for efficient exploration of model configurations. This will allow us to generate reproducible, high-resolution biodiversity predictions that can inform ecological research and support environmental decision-making at multiple spatial scales.