Tropical Dry Forest climatic edaphic and vegetation Database

Selection of sample points
We created a 1 km2 grid for each of the five biogeographic regions and identified the centroid of each cell of the grid, which we used as a sample point. We selected 100 sample points in a random block design from the full population of centroids for a given ecoregion and repeated this for each of the 80 studied ecoregions. We then used the HILDA+ Global Land Use Change layer (HILDAplus_vGLOB-1.0; Winkler, Fuchs, Rounsevell, & Herold, 2021) to evaluate the land use types for the resulting grid of 8,000 points. The HILDA+ layer contains land cover information from 1960 to 2019, which allowed us to mask points defined as urban, cropland, ocean, water, or as missing data. We used the remaining unmasked points, defined as forest, shrubland, pasture, rangeland, or unmanaged grassland, in our analyses.
We used a randomized block design to assess relationships between NDVI and climatic and edaphic variations found across TDL. All spatial analyses were conducted in ArcGIS 10.2.1.

Normalized difference vegetation index
We used 12 years (2005 to 2016) of NDVI data (250 m pixel resolution) collected with Moderate Resolution Imaging Spectroradiometer (MODIS) instruments aboard US NASA’s Terra and Aqua satellites. The MODIS product was used because of its longevity, resolution, and frequent observations. Data analyses relied on 12-yr mean NDVI conditions for each of the 80 ecoregions sampled here. While data are collected sub-weekly, we used averaged monthly data to calculate mean annual NDVI for the 12-yr period. NDVI is dimensionless and ranges from -0.1 to 0.9, with higher values associated with denser vegetation and lower values associated with sparser vegetation. The lowest values represent bare ground.

Climatic and edaphic metrics
For each sample point, we obtained rainfall, temperature, and evapotranspiration data from repositories including: NASA Earth Observations (https://neo.gsfc.nasa.gov/); Climatic Hazards Center - UC Santa Barbara (www.chc.ucsb.edu/about); Numerical Terradynamic Simulation Group (NTSG)-University of Montana (www.ntsg.umt.edu/); Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC) for Biogeochemical Dynamics (www.daac.ornl.gov/); and SOILGRID (www.soilgrids.org/). We examined a total of 17 climate variables: four temperature metrics, 10 precipitation metrics, an evapotranspiration metric, and two aridity metrics that integrate annual precipitation and temperature measures. All climate variables were summarized into mean annual averages representing from 1 up to 12 years of available data (Table S3). We also considered 17 edaphic metrics involving physical, chemical and biological properties of soils. These were all represented by a single point in time measurement, and so have no temporal averaging component. Each of the 34 metrics was processed and normalized to standard units.