GlobalGeoTree: A Multi-Granular Vision-Language Dataset for Global Tree Species Classification

TL;DR

GlobalGeoTree provides a globally representative, multimodal dataset pairing Sentinel-2 time-series with extensive environmental context to enable large-scale tree species classification across a 21,001-species taxonomic spectrum. It introduces GeoTreeCLIP, a vision-language model pretrained on 6M samples and evaluated on a 10k-species benchmark, demonstrating strong zero-shot and few-shot performance by leveraging hierarchical taxonomic text. The dataset design, validation framework, and evaluation protocol address long-tail distributions and geographic domain shifts, establishing a robust benchmark for biodiversity mapping. The work offers open data, code, and baselines to advance practical ecological monitoring and dynamic forest management applications.

Abstract

Global tree species mapping using remote sensing data is vital for biodiversity monitoring, forest management, and ecological research. However, progress in this field has been constrained by the scarcity of large-scale, labeled datasets. To address this, we introduce GlobalGeoTree, a comprehensive global dataset for tree species classification. GlobalGeoTree comprises 6.3 million geolocated tree occurrences, spanning 275 families, 2,734 genera, and 21,001 species across the hierarchical taxonomic levels. Each sample is paired with Sentinel-2 image time series and 27 auxiliary environmental variables, encompassing bioclimatic, geographic, and soil data. The dataset is partitioned into GlobalGeoTree-6M for model pretraining and curated evaluation subsets, primarily GlobalGeoTree-10kEval for zero-shot and few-shot benchmarking. To demonstrate the utility of the dataset, we introduce a baseline model, GeoTreeCLIP, which leverages paired remote sensing data and taxonomic text labels within a vision-language framework pretrained on GlobalGeoTree-6M. Experimental results show that GeoTreeCLIP achieves substantial improvements in zero- and few-shot classification on GlobalGeoTree-10kEval over existing advanced models. By making the dataset, models, and code publicly available, we aim to establish a benchmark to advance tree species classification and foster innovation in biodiversity research and ecological applications.

Figures (10)

• Figure 1: Overview of the GlobalGeoTree dataset, which includes 6.3 million samples spanning 21,001 tree species across 221 countries/regions. The map illustrates the geographic coverage, with color intensity representing the number of samples in each 1° × 1° latitude/longitude grid. Each sample is paired with remote sensing data, including Sentinel-2 time series, auxiliary environmental variables, and hierarchical taxonomic labels spanning from functional type to species level.
• Figure 2: The taxonomic hierarchy of the GlobalGeoTree dataset. The visualization shows the nested relationships, branching from the four functional types (e.g., Deciduous Broadleaf) at the center, through families (e.g., Fabaceae), genera, and out to the 21,001 species at the outermost ring. This multi-level structure is a core feature that enables multi-granular classification tasks.
• Figure 3: Species in GlobalGeoTree are categorized into Frequent, Common and Rare groups based on the number of samples per species.
• Figure 4: Geographic distribution of GlobalGeoTree-10kEval. This benchmark includes species selected from Frequent, Common, and Rare categories, as described in the text.
• Figure 5: The GlobalGeoTree Explorer App. The interface allows users to visualize the global distribution of tree samples (bottom map) and inspect individual points against high-resolution satellite imagery (top panels a-f). Detailed attributes for selected samples are displayed in the sidebar, facilitating transparent community validation.