Technical specsNative resolution: 1 km2 (30 arc seconds)Spatial extent: GlobalMasked area: Non-vegetated landcover from remote sensing datasets (e.g., rock, ice, and desert habitats) and highly urban/built-up landcoverUnits: Predicted Richness = number of species / 100 m2 (for any 100 m2 area within a 1 km2 pixel, this is the predicted quantity of species); Predicted Endemism = rarity-weighted richness (normalized as a relative importance score between 0 - 10); Uncertainty = coefficient of variation (unitless); Model Extrapolation = %DescriptionThe mycorrhizal mapping data products shown here are high-resolution spatial interpolations  from ensemble machine-learning models. These predictive models are trained on mycorrhizal fungal diversity metrics from a curated global database of soil fungi, encompassing 2.8 billion fungal DNA sequences from 25,000 soil samples across 130 countries, and dozens of open-source environmental layers on climate conditions, vegetation, topography, soil properties, and human factors (e.g., percent of human-modified landcover). Models were built as k-fold cross-validated random forest regression models, with final predictions calculated as an ensemble average of the top 10 highest performing models over 100 bootstrapped runs.Each map pixel represents a prediction of mycorrhizal fungal diversity per 100 m2 — in other words, the expected mycorrhizal diversity (combined from multiple sub-samples covering a 100 m2 area) within each 1 km pixel. The ‘richness’ predictions come from sample-level calculations of the total number of unique fungal species using a CHAO rarefaction/extrapolation estimator. The ‘endemism’ predictions are based on rarity-weighted richness (RWR) calculations using a sample-level sum of species rarity. For simplicity, we have normalized the RWR data to a relative importance scale between 0 and 10, as RWR is a unitless metric. Note that mycorrhizal fungal ‘species’ here refer to 97% similar clustered Operational Taxonomic