Modeled Harvest Dates of Barley and Wheat (2005–2023) in Peninsular Spain, UTM 10×10 km Grid

Separate Generalized Additive Models (GAMs) were fitted using penalized splines with the mgcv::bam() function, optimized via fast Restricted Maximum Likelihood (fREML) to efficiently handle large datasets and incorporate random-effect smooths. Two crop-specific models—one for barley and one for wheat—were constructed to predict harvest week across mainland Spain, using a 10 × 10 km UTM grid as the spatial framework for the period 2005–2023. Each grid cell was assigned to the agricultural district occupying the largest area within that cell, and this assignment was used both for linking input data and for aggregating modeled harvest dates. Predictor data Meteorological data sets were obtained from CHELSA (www.chelsa-climate.org) and the Spanish Meteorological Agency (www.aemet.es). Cartographic bases (administrative boundaries, elevation model, UTM grid) were obtained through the services provided by the Spatial Data Infrastructure of Spain (www.idee.es). Harvest dates were provided by Agroseguro (www.agroseguro.es), the entity responsible for managing agricultural insurance on behalf of the insurance companies that form the Spanish co-insurance pool. These data were available at the agricultural district level. In addition, an average harvesting date was generated for each grid cell, taking into account the proportion of each crop sown within the cell for the period 2015–2023, which is the timeframe for which data are available. These data were provided by the Spanish Agricultural Guarantee Fund (Fondo Español de Garantía Agraria, O.A.; www.fega.gob.es) at the agricultural district level. Columns in the dataset are: utmcode year wheat_harvest_week barley_harvest_week harvest_week Harvest week follows ISO 8601-1:2019 (www.iso.org/iso-8601-date-and-time-format.html)