# Accuracy assessment of the Global Forest Change tree cover loss by visual interpretation and comparison with forestry logging records in Slovakia

## TL;DR

The Global Forest Change dataset is mostly accurate for tracking tree cover loss in Slovakia but has some limitations in small areas and edge pixels.

## Contribution

This study evaluates the accuracy of the Global Forest Change dataset in Slovakia using visual interpretation and forestry records.

## Key findings

- The GFC dataset achieved 95% producer’s accuracy and 78% user’s accuracy in detecting tree cover loss.
- GFC and forestry logging records showed similar total areas of loss and spatial patterns with a median difference of 6%.
- Disagreements in GFC data were partly due to marginal pixels and operator interpretation challenges.

## Abstract

Global Forest Change (GFC) provides a widely accessible tool for monitoring tree cover loss and is frequently used by both professionals and the public in discussions of forest dynamics. However, its reliability in specific regions has not been sufficiently evaluated. We assessed its accuracy in part of Slovakia through visual interpretation of Sentinel 2 imagery conducted independently by three operators. The GFC dataset achieved 95% producer’s accuracy and 78% user’s accuracy in detecting tree cover loss. Therefore, it could slightly overestimate loss in the study area, especially in small patches and edge pixels, as the marginality of pixels explained a large part of the disagreement (AUC = 0.76). However, we could also underestimate the loss during its visual identification, as it proved more challenging than anticipated (average agreement between operators was 85%). To complement this analysis, we compared GFC tree cover loss with official forestry logging records across Slovakia. Both datasets reported similar total areas of loss and largely consistent spatial patterns, with a median difference of 6% in their ratio within forestry mapping units. Nonetheless, we identified regions where GFC reported substantially more loss, potentially reflecting errors in forestry records, as well as regions where forestry records indicated more loss, possibly due to temporal misclassification. Overall, the GFC dataset represents an appropriate source for monitoring tree cover loss in Slovakia, provided that users are aware of its limitations.