A complete description of maps' methods, accuracy, strengths, and limitations is available in Aramburu-Merlos et al. (Nat. Food, 2024).
Briefly, we combined site-specific yield potential estimates of the Global Yield Gap Atlas with gridded environmental predictors in a machine-learning metamodel to generate global maps of yield potential at a 30-arc-second resolution for maize, wheat, and rice, separately for irrigated and rainfed conditions. Model predictions were restricted to their area of applicability and lands harvested with the given crop and water regime condition (harvested area > 0.5% according to SPAM v2.0).