2.3. Environmental variables
We used bioclimate layers combined with soil properties to project current and future niches for the species and each genetic group. A total of 14 bioclimate variables were downloaded from Africlim online regional climate models (RCMs) data portal (https://webfiles.york.ac.uk/KITE/AfriClim/) (Platts et al. 2014). Current and future variables averaged between the time periods 1986-2015 (2000) and 2041-2060 (2055) were downloaded at a 30 arc-second (~1 km) spatial resolution. For future climatic conditions, predictions from the Ensemble model (Platts et al.2014) were used. This model simulates changes based on a set of scenarios. The projections were run under Representative Concentration Pathway (RCP), RCP 4.5 and RCP 8.5 for the 2055 time horizon (Meinshausen et al. 2011). In all RCPs, the climatic conditions are extreme in RCP 8.5 scenarios compared to RCP 4.5. RCP 4.5 projects temperatures to rise above industrial levels by at least 1.5°C in West Africa, with atmospheric CO2 reaching 500 ppm while in RCP 8.5 projections, temperatures are predicted to rise by 2.8°C and atmospheric CO2 to be over 550 ppm (IPCC 2013). These climate projections were statistically downscaled to match the bioclimatic variables using the delta method, (Ramirez-Villegas & Jarvis 2010).
Data related to soil characteristics were available in the World Soil Information (ISRIC) databases (Soil-property-maps-of-Africa-at-250-m-resolution)at 250 m resolution (Hengl et al.2015). These spatial predictions of soil properties were generated based on two predictive approaches such as random forests and linear regression (Hengl et al. 2015). Soil characteristics identified as relevant to M. geocarpumagricultural management included 11 variables related to the soil physical, chemical and nutritional properties. Soil data were then converted to 30 arcseconds using ArcGis software v 10.7.1 to match with bioclimate layers. Finally, using shapefile boundaries of four West African countries (Benin, Burkina Faso, Ghana and Togo) we cropped all variables to encompass the broad geographic regions that define Kersting’s groundnut global distribution.
Jackknife Procedure in Maxent 3.4.4 was used to reduce the number of variables to be included in the prediction models (Phillips et al. 2005). The six variables with highest contribution proportions were selected and were used in the final models of the species, and with genetic information.