This proposal requests computational resources to develop deep learning methods for evaluating forest naturalness using Canopy Height Models (CHM). Assessing forest naturalness is crucial for understanding biodiversity, ecosystem stability, and resilience. Traditional methods are labor-intensive and prone to error, while advancements in remote sensing and deep learning offer accurate and scalable alternatives. The project aims to create a deep learning framework to analyze high-resolution CHM data for naturalness evaluation. CHMs, derived from LiDAR data, provide 3D representations of forest canopy structures, capturing variations in canopy height indicative of different forest conditions. The approach involves data preprocessing, model development, training, and validation. High-resolution CHM data - provided by Skogsstyrelsen as open data - will be processed to enhance quality, followed by employing deep learning architectures like Convolutional Neural Networks (CNNs) and Transformer-based models to extract features. These models will be trained on annotated datasets with known naturalness levels. Due to the computational demands of training deep learning models on large-scale CHM data, we request access to high-performance computing facilities with GPUs. These resources will facilitate efficient model training and hyperparameter tuning, reducing training time and improving performance. The expected outcomes include a highly accurate deep learning model for forest naturalness evaluation, a comprehensive dataset of annotated CHM data, and best practices for applying deep learning to remote sensing data. This project will provide a valuable tool for forest management, conservation planning, and ecological research, aiding in monitoring forest health, detecting disturbances, and guiding restoration efforts. Integrating deep learning with CHM data represents a significant advancement in forest naturalness evaluation, contributing to sustainable forest management.