Dargana: fine-tuning EarthPT for dynamic tree canopy mapping from space

2025-04-24

AI Summary

This study addresses the need for efficient natural capital management by proposing a low-overhead, high-accuracy method for dynamic canopy cover monitoring. Methodologically, it leverages the EarthPT temporal foundation model, applying lightweight supervised fine-tuning using <3% of its pretraining data and only 5% of the original computational budget, while integrating multi-temporal Sentinel-2 imagery to achieve 10-m pixel-level semantic segmentation. The framework supports coniferous/deciduous forest classification, as well as fine-grained structural identification (e.g., hedgerows, shrubs) and detection of dynamic changes such as new afforestation. Its key contribution lies in being the first to specialize EarthPT for fine-grained vegetation dynamics monitoring under extreme data and compute constraints—enabling sub-pixel object recognition and quantitative time-series analysis despite limited annotations. Evaluated in Cornwall, UK, it achieves ROC-AUC = 0.98 and PR-AUC = 0.83, and generalizes successfully to unseen fine structures and novel change types.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

We present Dargana, a fine-tuned variant of the EarthPT time-series foundation model that achieves specialisation using<3% of its pre-training data volume and 5% of its pre-training compute. Dargana is fine-tuned to generate regularly updated classification of tree canopy cover at 10m resolution, distinguishing conifer and broadleaved tree types. Using Cornwall, UK, as a test case, the model achieves a pixel-level ROC-AUC of 0.98 and a PR-AUC of 0.83 on unseen satellite imagery. Dargana can identify fine structures like hedgerows and coppice below the training sample limit, and can track temporal changes to canopy cover such as new woodland establishment. Our results demonstrate how pre-trained Large Observation Models like EarthPT can be specialised for granular, dynamic land cover monitoring from space, providing a valuable, scalable tool for natural capital management and conservation.

Problem

• Specializing EarthPT for dynamic tree canopy mapping
• Classifying tree types at 10m resolution from space
• Tracking temporal changes in canopy cover efficiently

Innovation

• Fine-tuned EarthPT for dynamic canopy mapping
• Uses minimal data and compute for specialization
• Achieves high accuracy in tree classification