Meter-Resolution Canopy Height Data Alone Can Be Used for Predicting the Conservation Value of Boreal Forests

Artikel i vetenskaplig tidskrift, 2026

We consider the problem of automated binary classification of forests as having either low or high conservation value. We demonstrate that the resolution of the data used in classification is a crucial factor. At resolutions of around 2 m or less, it is possible to distinguish individual trees, a situation that greatly improves classification accuracy: Using canopy height models with 1-m resolution our U-Net-based classifier obtains a classification accuracy of 0.956 over a test dataset covering southern Sweden. This result can be compared to earlier reported accuracies in the range 0.81–0.92 at lower resolution (10–100 m) over a similar region. We also show how the classification performance varies over a range of resolution values, and find a strong positive jump in accuracy at resolutions of 2 m or less, where individual trees can be discerned.