Climate suitability predictions for the cultivation of macadamia (Macadamia integrifolia) in Malawi using climate change scenarios

predicted binary values for each pixel were extracted. Finally, the total number of pixels for each predicted class was used to estimate the total coverage of the predicted suitable area against the unsuitable area within Malawi. Following recommendations by Chemura et al. [55], we divided the two suitability classes (suitable/unsuitable) into five classes (unsuitable, marginal, moderate, optimal, and highly suitable). The final visualization maps for the suitability classes of macadamia were developed using Arc GIS Pro software version 2.5 (https://arcgis.pro/). In the fourth stage, we applied the derived baseline suitability model to each of the 17 downscaled GCMs to predict the future distribution of suitable areas for macadamia by the 2050s. Finally, the results of the 17 GCMs probability layers were integrated into a single layer, using the criterion of likelihood scale [56,57], which requires at least 66% of agreement among GCMs to keep the predicted presence or absence in a given grid cell. The final visualization maps for the future suitability classes of macadamia were developed using Arc GIS Pro soft- ware version 2.5 (https://arcgis.pro/). 3. Results 3.1. Model performance evaluation Our results show that the ensemble model performance (AUC = 0.9) was sufficient for our modelling activity when measured using the AUC. The model’s evaluation revealed that the modelling of macadamia areas in Malawi was based on model competence rather than chance (Table 2). Importantly, the high AUC value provides confidence to apply the ensemble model

for examining the areas suitable for macadamia under current and future climatic conditions. 3.2. Contribution of variables to the suitability of macadamia The importance of climatic factors driving the suitability of macadamia production in Malawi is shown in Fig 4. Precipitation-related variables are the most important in determining Table 2. Performance evaluation of the ensemble model. Algorithm Method AUC Envelope model BIOCLIM 0.86 Multivariate distance DOMAIN 0.90 Additive models: Generalized additive models GAM 0.89 Regression: Multivariate adaptive regression splines MARS 0.93 Stepwise GAM GAMSTEP 0.85 Mahalanobis distance MAHAL 0.99 Maximum entropy MAXENT - Boosted regression models: Generalized boosted regression models GBM - Generalized linear models GLM 0.98 Support vector machines SVM 0.86 Stepwise boosted regression tree models GBMSTEP 0.94 Artificial neural networks NNET 0.96 Random Forest RF 0.94 Multivariate Adaptive Regression Splines EARTH - Stepwise generalized linear models GLMSTEP 0.82 Mixed GAM Computation Vehicle MGCV 0.85 Support vector machines SVM 0.98 Flexible discriminant analysis FDA - Ensemble ENSEMBLE 0.90 https://doi.org/10.1371/journal.pone.0257007.t002 PLOS ONE Macadamia integrifolia: Climate suitability of Malawi PLOS ONE | https://doi.org/10.1371/journal.pone.0257007 September 9, 2021 9 / 20