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

Description

This repository contains the data, scripts, and documentation supporting the study "Mapping forest tree species and their uncertainty using Earth observation and National Forest Inventory data: towards operational monitoring in Sweden” by Abdulhakim M. Abdi and Fan Wang. The study is published in the International Journal of Remote Sensing. In addition to citing the data, please also cite the study if you use these data products:

Abdi, A. M., & Wang, F. (2026). Mapping forest tree species and their uncertainty using Earth observation and National Forest Inventory data: towards operational monitoring in Sweden. International Journal of Remote Sensing, 1–32. https://doi.org/10.1080/01431161.2026.2625513

The materials include tree species classification raster, a pixel-level entropy raster representing classification uncertainty, and spatially continuous, entropy-weighted tree species fractions, as well as the R implementation for deriving these fractions. Together, these resources enable large-area analyses of forest composition, uncertainty propagation, and the spatial characterization of forest stands across southern Sweden (Skåne, Blekinge, Halland, Kronoberg, Jönköping, and Kalmar counties).

Interactive maps of tree species and model uncertainty are available at: https://ee-treespec.projects.earthengine.app/view/treespec  

Contents

• TreeSpecies_Classification_XGB.tif — Discrete raster (8-bit unsigned integer) of dominant tree species predicted by an XGBoost model trained on Sentinel-1/2 and topographic data. The classes correspond to: 1-Norway spruce, 2-Scots pine, 3-Birch, 4-Beech, 5-Oak, 6-Alder, 7-Aspen, 8-Other species.

• TreeSpecies_Entropy_XGB.tif — Continuous raster (16-bit unsigned integer) containing per-pixel Shannon entropy values (0–Hmax), quantifying classification uncertainty.

• Weighted_Fraction_XXXX.tif (XXXX = Tree species/class) — Continuous rasters (Float32, 0–100%) representing local, entropy-weighted fractional cover of each tree species computed within a moving Gaussian window. Each file corresponds to one of the eight dominant species (Norway spruce, Scots pine, Birch, Beech, Oak, Alder, Aspen, and Other species). An additional “Weighted_Fraction_Unknown.tif” layer represents the proportion of unclassified or masked pixels within the window.

• Convert_classification_to_entropy-weighted_tree_species_fractions.R — R script that computes local (moving-window) species fractions with optional entropy weighting, producing 0–100% fractional cover maps for each species and an additional “Unknown” fraction layer.

• Derivation of entropy-weighted tree species fractions.docx — Document detailing the steps taken to derive the entropy-weighted tree species fractions.

Spatial characteristics