Multiscale tree species composition, classification uncertainty, and diversity metrics for Swedish forests based on National Forest Inventory data and satellite remote sensing

This repository contains the data and documentation for the data paper: "Multiscale tree species composition, classification uncertainty, and diversity for southern Swedish forests” by Abdulhakim M. Abdi and Fan Wang. The paper is currently under preparation for submission. The repository contains a tree species classification raster, a pixel-level entropy raster representing classification uncertainty, entropy-weighted tree species fractional cover, and a suite of forest diversity metrics. Together, these resources enable large-area analyses of forest composition, uncertainty propagation, ecological monitoring, biodiversity assessment, sustainable forest management, and the spatial characterisation of forest stands across Götaland. Contents Tree_Species_Classification.tif — Discrete raster (8-bit unsigned integer) of dominant tree species predicted by an XGBoost model trained on Sentinel-1/-2, topographic, and canopy-height data. Raster values: 1 = Norway spruce, 2 = Scots pine, 3 = Birch, 4 = Beech, 5 = Oak, 6 = Alder, 7 = Aspen, 8 = Larch, 9 = Other species. Pixels outside productive forest carry a no-data value of 0. Accompanied by Tree_Species_Classification.clr, a colour map defining the legend in QGIS. Classification_Uncertainty.tif — Continuous raster (16-bit unsigned integer) of per-pixel Shannon entropy quantifying classification uncertainty. Values are base-2 entropy (bits) scaled by 10,000; divide by 10,000 to recover entropy (0 to log₂9 ≈ 3.17). Species_Fractions_30m.zip — Nine continuous rasters (Float32, 0–100%) giving the entropy-weighted fractional cover of each tree species at 30 m resolution, aggregated from the 10 m classification in disjoint 3 × 3 blocks with each pixel weighted by its classification confidence (1 − normalized entropy). One file per class (Norway spruce, Scots pine, Birch, Beech, Oak, Alder, Aspen, Larch, Other species). All pixels, including non-forest, enter the denominator, so the nine layers sum to ≤ 100% per cell; non-forest cells are 0. Diversity_Metrics_100m.zip — Six continuous rasters (Float32) of forest diversity metrics at 1-hectare (100 × 100 m) resolution — Shannon's diversity, its Hill-number transformation, Simpson's D, the Gini–Simpson index, Pielou's evenness, and species richness — computed from the species proportions of the classified forest pixels within each cell. Includes forest_fraction_100m.tif, giving the proportion of each cell classified as productive forest (0–1), provided so users can restrict or down-weight the metrics in sparsely forested cells. Fractions_Description.docx — Documentation for the species fractional-cover layers: derivation method, entropy weighting, and interpretation of the 30 m fractions. fractions_code.R — R script that generates the 30 m entropy-weighted species fractional-cover layers from the 10 m classification and entropy rasters. Diversity_Description.docx — Documentation for the diversity metrics: within-cell species proportions, the six metric formulas, and the accompanying forest-fraction layer. diversity_code.R — R script that computes the six 1-hectare diversity metrics and the forest-fraction layer from the 10 m classification raster. Spatial characteristics Property Specification Geographic coverage Region Götaland (Skåne, Blekinge, Halland, Kronoberg, Jönköping, Kalmar, Västra Götaland, Östergötland, Gotland counties) Extent 267320, 6133440 : 757330, 6571770 (EPSG:3006 – SWEREF99 TM) Projection Projected (UTM), units in meters Spatial resolution 10 × 10 m (classification, uncertainty) · 30 × 30 m (species fractions) · 100 × 100 m (diversity metrics) Raster dimensions 49,001 × 43,833 pixels (10 m grid) Origin 267320, 6571770