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

increase in temperature by 1.4 oC (0.9–2.0 oC) and RCP 8.5, the most pessimistic scenario, which considers higher GHG emissions concentration with a 1.4–2.6 oC projected increase in mean global temperature by the 2050s (period 2040–2060). For this study, we did not consider scenario 2.6 because it represents the most efficient and effective mitigation scenario, i.e., keeping the temperature below 2 oC. At present, this scenario is not feasible with projections of current policies (expected temperature increase of 3.3–3.9 oC) [35]. Furthermore, to achieve this scenario, emissions would need to be 25% lower than in 2018 (GHG emissions rose by 1.5% in 2018) [36]. Emission scenario 6.0 was also not consid- ered for our analysis because its projections are between the ranges of RCP 4.5 and RCP 8.5 [37]. Further, RCP 6.0 only has 42% of the GCM outputs, meaning that the scenario has fewer outputs than the rest of the emission scenarios [38]. 2.4. Variable selection In species distribution models, multicollinearity (multiple correlations between variables) among the bioclimatic predictors may result in overfitting or bias in the resulting suitability model [39,40]. To avoid these challenges, variable quality evaluation criterion using a multicol- linearity degree was employed through the variance inflation factor analysis (VIF). The VIF indicates the degree to which the standard errors have been inflated due to the levels of multi- collinearity among the independent variables used in running the model [41]. VIF is directly

calculated from a linear regression model with the focal numeric variable as a response, as shown in Eq (1). VIF ¼ 1 1 R2 i ð1Þ Where R2 is the regression coefficient of determination of the linear model. In our study, the "ensemble.test" function inherent in the "BiodiversityR" package available in R [42] was used to eliminate correlated variables. Following the recommendation made by Ranjitkar et al. [43], we retained variables that had a VIF of less than 10 (Table 1). 2.5. Modelling approach We modelled macadamia’s current and future distribution in Malawi based on an ensemble suitability method implemented by the R package "BiodiversityR" [44]. We used an ensemble modelling technique because it combines predictions from various algorithms and can provide better accuracy in predictions than relying on individual species distribution models [45]. The procedure consisted of four steps. Table 1. Bioclimatic variables used in the final suitability model and their variance inflation factor (VIF). Variable name Bioclimatic variable Unit VIF Score Bio 14 Precipitation of driest month mm 2.96 Bio 3 Isothermality (Bio2/Bio7) x 100 - 1.51 Bio 15 Precipitation seasonality (cv x 100) - 3.25 Bio 2 Mean diurnal range oC 6.05 Bio 18 Precipitation of warmest quarter mm 5.23 Bio 13 Precipitation of wettest month mm 2.12 Bio 6 Minimum temperature of the coldest month oC 2.02 Bio 4 Temperature seasonality (Standard deviation x 100) - 1.61 https://doi.org/10.1371/journal.pone.0257007.t001 PLOS ONE Macadamia integrifolia: Climate suitability of Malawi

PLOS ONE | https://doi.org/10.1371/journal.pone.0257007 September 9, 2021 7 / 20