HPC-Accelerated Deep Learning for Early Bark Beetle Outbreak Detection

The business experiment addresses the challenge of late detection of spruce bark beetle outbreaks, which cause significant economic and ecological damage across European forests. PRIOT will develop an HPC-accelerated deep learning pipeline that analyses Sentinel-2 NDII time series, additional vegetation indices, and spectral bands to identify early signs of canopy stress before visible symptoms appear. The work includes data cleaning, cloud filtering, gap filling, label generation using CUSUM and PELT methods, training and benchmarking LSTM, GRU, Transformer, and CNN models, hyperparameter optimisation, cross-region validation, and the development of a prototype service. The goal is to transform PRIOT’s research prototype into a validated early-warning Software-as-a-Service (SaaS) solution for forest owners, forest managers, and public forestry agencies, enabling earlier intervention and more effective forest protection.

Organisations involved: 

End-User: PRIOT d.o.o.

Technology expert: University of Maribor, Laboratory for Heterogeneous Computer Systems

Domain Expert: Slovenian Forest Service