View in EDS

Mapping forest tree species and their uncertainty using Earth observation and National Forest Inventory data: towards operational monitoring in Sweden.

Saved in:
Bibliographic Details
Title: Mapping forest tree species and their uncertainty using Earth observation and National Forest Inventory data: towards operational monitoring in Sweden.
Authors: Abdi, Abdulhakim M.1 (AUTHOR) hakim.abdi@mgeo.lu.se, Wang, Fan2 (AUTHOR)
Source: International Journal of Remote Sensing. Apr2026, Vol. 47 Issue 7, p2912-2943. 32p.
Subjects: Forest mapping, Remote sensing, Classification algorithms, Forest surveys, Spatial resolution, Forests & forestry
Geographic Terms: Sweden
Abstract: Timely, detailed information on forest composition is essential for effective management, biodiversity protection, and understanding ecosystem dynamics. This study maps the distribution of seven dominant tree species in Swedish forests and produces spatially explicit, pixel-level estimates of classification uncertainty. The mapping framework integrates multitemporal Sentinel-1 radar and Sentinel-2 optical observations with field data from the Swedish National Forest Inventory and auxiliary predictors describing topography and canopy height. We trained a Bayesian-optimized extreme gradient boosting model on spatiotemporal metrics derived from these datasets and quantified classification confidence through entropy computed from the class-probability outputs. We applied a spatial block partitioning approach to limit the effects of spatial autocorrelation between optimization and validation data and ensure a more realistic assessment of the model's generalization capacity. Model overall accuracy reached 85% (F1 = 0.82) using a 60 m spatial block validation. Under a more conservative 200 m block configuration, performance decreased to F1 = 0.63, reflecting reduced training data availability. The county-level species coverage derived from the classification aligned closely with published figures from the Swedish Forest Agency (Spearman's ρ = 0.94, 95% CI: 0.89 – 0.96, p < 0.001). Variable importance analysis showed that Sentinel-2 spectral bands, particularly shortwave-infrared and red-edge captured during spring and summer, contributed most to species discrimination, while Sentinel-1 backscatter provided complementary structural information. The integration of forest inventory data, Earth observation, and machine learning to produce tree species maps and a spatially explicit measure of prediction uncertainty yields a robust and reproducible framework for large-area forest mapping. The results provide detailed, spatially continuous information on species composition along with an accompanying confidence surface. This offers practical value for ecological assessments, regional planning, and emerging legislative and environmental goals. The data are freely available for download and the maps can be interactively visualized using this link: . [ABSTRACT FROM AUTHOR]
Copyright of International Journal of Remote Sensing is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Engineering Source
Description
Abstract:Timely, detailed information on forest composition is essential for effective management, biodiversity protection, and understanding ecosystem dynamics. This study maps the distribution of seven dominant tree species in Swedish forests and produces spatially explicit, pixel-level estimates of classification uncertainty. The mapping framework integrates multitemporal Sentinel-1 radar and Sentinel-2 optical observations with field data from the Swedish National Forest Inventory and auxiliary predictors describing topography and canopy height. We trained a Bayesian-optimized extreme gradient boosting model on spatiotemporal metrics derived from these datasets and quantified classification confidence through entropy computed from the class-probability outputs. We applied a spatial block partitioning approach to limit the effects of spatial autocorrelation between optimization and validation data and ensure a more realistic assessment of the model's generalization capacity. Model overall accuracy reached 85% (F1 = 0.82) using a 60 m spatial block validation. Under a more conservative 200 m block configuration, performance decreased to F1 = 0.63, reflecting reduced training data availability. The county-level species coverage derived from the classification aligned closely with published figures from the Swedish Forest Agency (Spearman's ρ = 0.94, 95% CI: 0.89 – 0.96, p < 0.001). Variable importance analysis showed that Sentinel-2 spectral bands, particularly shortwave-infrared and red-edge captured during spring and summer, contributed most to species discrimination, while Sentinel-1 backscatter provided complementary structural information. The integration of forest inventory data, Earth observation, and machine learning to produce tree species maps and a spatially explicit measure of prediction uncertainty yields a robust and reproducible framework for large-area forest mapping. The results provide detailed, spatially continuous information on species composition along with an accompanying confidence surface. This offers practical value for ecological assessments, regional planning, and emerging legislative and environmental goals. The data are freely available for download and the maps can be interactively visualized using this link: . [ABSTRACT FROM AUTHOR]
ISSN:01431161
DOI:10.1080/01431161.2026.2625513