The environmental consequences of climate-driven agricultural frontiers

RCPs 4.5 and 8.5) (S1 Table). S1 Fig presents a flowchart showing the details of GCMs and other data used and their analysis. Current and future monthly climate grids were obtained from WorldClim Global Climate Data (www.worldclim.org). All data obtained was downscaled to 30 arc-second (approx. 1km) horizontal resolution following the methods of Hijmans et al. 2005 [21]. Variables available for download at this resolution and used in the simulations reported here were mean daily maxi- mum temperature of each month (Tmax), mean daily minimum temperature of each month (Tmin), mean total precipitation of each month (Precip) and a suite of 19 bioclimatic variables [S2 Table]. Current climate represents the mean monthly climate over the period 1950–2000 whereas future climate projections cover 20-year averages over the periods 2040–2060 and 2060–2080. Crop suitability models Climatic suitability for twelve globally important crops is determined using three discrete model- ing methods 1) EcoCrop; 2) Maxent; 3) frequency of daily critical maximum and minimum tem- peratures. The list of crops is listed in Table 1 and parameters used is depicted in S3 Table. EcoCrop is a generalized physiological model of crop suitability based on known ranges of optimal temperature and precipitation as well as climatic limits where production of that crop would be impossible [22, 23]. Parameters for optimal temperature and precipitation are avail- able from the EcoCrop database maintained by the FAO [24]. Maximum entropy (Maxent) is widely used in modeling species distributions under climate

change and increasingly applied to domesticated crop species [25]. Crop occurrence points were generated from spatially weighted random points representing the agreement of four dis- crete gridded crop databases (S4 Table). One thousand points were generated for each crop, with a greater percentage of the distribution given where there is greater agreement among the four databases. Ten variables were selected by hierarchical clustering to create the Maxent model for each crop. For each crop, all 19 bioclimatic variables were sampled at the crop obser- vation points. Variables were standardized then partitioned into clusters using hclust() base R function. The resulting cluster dendrogram was partitioned into 10 clusters using cutree() base R function where k = 10. The first variable was selected from each of the resulting 10 groups. See S2 and S5 Tables for a description of the bioclimatic variables used. For all models, 30% of the occurrence points were reserved to test and validate the model. Model performance (AUC), logistic threshold and top four most important variables for each crop are shown in S6 Table. Table 1. List of crops modeled. Crop Name Species Corn Zea mays Sugar Saccharum officinarum Oil Palm Elaeis guineensis Cassava Manihot esculenta Peanuts Arachis hypogaea Cotton Gossypium hirsutum Millet Pennisetum glaucum Sorghum Sorghum halepense Rice Oryza sativa Potato Solanum tuberosum Wheat Triticum aestivum Soy Glycine max https://doi.org/10.1371/journal.pone.0228305.t001 The environmental consequences of climate-driven agricultural frontiers

PLOS ONE | https://doi.org/10.1371/journal.pone.0228305 February 12, 2020 3 / 19