Climatic suitability predictions for the cultivation of macadamia in Malawi using climate change scenarios

Global climate change is altering the suitable areas of crop species worldwide, with cascading effects on people and animals reliant upon those crop species as food sources. Macadamia is one of these essential lucrative crop species that grows in Malawi. Here, we used an ensemble model approach to determine the current distribution of macadamia production areas across Malawi in relation to climate. For future distribution of suitable areas, we used the climate outputs of 17 general circulation models based on two climate change scenarios (RCP 4.5 and RCP 8.5). The precipitation of the driest month and isothermality were the climatic variables that strongly influenced macadamia’s suitability in Malawi. We found that these climatic requirements were fulfilled across many areas in Malawi under the current conditions. Future projections indicated that vast parts of Malawi’s macadamia growing regions will remain suitable for macadamia, amounting to 36,910 km2 (39.1%) and 33,511 km2 (35.5%) of land based on RCP 4.5 and RCP 8.5, respectively. Alarmingly, suitable areas for macadamia production are predicted to shrink by −18% (17,015 km2) and −21.6% (20,414 km2) based on RCP 4.5 and RCP 8.5, respectively, with much of the suitability shifting northwards. This means that some currently productive areas will become unproductive in the future, while current unproductive areas will become productive. Notably, suitable areas will increase in Malawi’s central and northern highlands, while the southern region will lose most of its suitable areas. Our study, therefore, shows that there is potential for expanding macadamia production in Malawi. Most, importantly our future projections provide critical evidence on the potential negative impacts of climate change on the suitability of macadamia production in the country. We recommend developing area-specific adaptation strategies to build resilience in the macadamia sector in Malawi under climate change.


Occurrence data. 162
Data on macadamia tree species' occurrence was collected from smallholder macadamia producing districts 163 in 2019 through a field survey of macadamia farms in Malawi. For our analysis, we only sampled ten-year-164 old successfully established macadamia orchards under smallholder rainfed conditions. At each farm, the 165 Global Position System (GPS) coordinates (in WGS84 datum) were collected using a global position system 166 (Garmin eTrex Vista ® Cx) together with altitude. A total of 120 orchards were sampled throughout Malawi, 167 but for this study, a total of 36 points were used for the modelling purpose (Fig 3). The remaining 84 168 occurrence points were used for cross-validation to evaluate the predictive model accuracy [46]- [49].

Climate data. 172
We used bioclimatic predictors (~1970-2000) from WorldClim data set version 1.4 (http:// 173 www.worldclim.org/) at a spatial resolution of 2.5 arc-minute (4.5 km 2 at the equator) to model the current 174 areas suitable for macadamia production in Malawi. Calculated from monthly temperature and precipitation 175 climatologies, these variables reflect spatial variations in annual means, seasonality, and extreme/limiting 176 conditions (Supplementary Table S2). We used bioclimatic variables from 17 general circulation models 177 (GCMs) based on two representative concentration pathways (RCP) of climate change [42] for future 178 predictions. We selected RCP 4.5, which is an optimistic scenario that considers an intermediate GHG 179 concentration and predicts an average increase in temperature by 1.4 o C (0.9-2.0 o C) and RCP 8.5 the most 180 pessimistic scenario, which considers higher GHG emissions concentration with a 1.4-2.6 o C projected 181 increase in mean global temperature by the 2050s (period 2046-2065). The bioclimatic variables from 182 WorldClim that we used include limiting factors that are ecologically important based on temperature and 183 precipitation variation. To avoid model overfitting, we selected the least correlated variables by applying 184 the variance inflation factors (VIF) and retained those with VIF < 10 [50]. Variables with the highest 185 correlation (VIF ≥ 10) were removed, resulting in eight bioclimatic predictors for our analysis (Error! 186 Reference source not found.). The long-term ecological conditions are essential for predicting perennial 187 crop production [51] because perennial crops are in the field for more than 25 years, and productivity is 188 measured by yield quantity and quality. 189

Modelling approach. 190
We modelled the current and future distribution of macadamia species in Malawi based on an ensemble 191 suitability method implemented by the R package BiodiversityR. We used an ensemble modelling technique 192 because it combines predictions from different algorithms and can provide better accuracy in predictions 193 than relying on individual species distribution models [52]. The procedure consisted of four steps. 194 Firstly, we evaluated the predictive performance accuracy of 18 algorithms of species distributions models 195 (SDM) using a fivefold cross-validation technique. Following work by [53]  the macadamia occurrence data was coupled with 500 randomly pseudo-absence data generated throughout 214 Malawi (Fig 3). We used the pseudo-absence data as opposed to using real absences to avoid 215 underestimation issues [59]. 216 In the second step, we retained only the algorithms that contributed at least 5% to the ensemble suitability 217  Table S5). AUC values ranged between 0 and 1, and a value less than 0.5 indicated 220 that the simulated result was worse than random (1) Where the ensemble suitability (Se) is obtained as a weighted (w) average of suitabilities predicted by the 227 contributing algorithm (Si). 228

12
The third step generated the current distribution maps (probability maps and presence-absence maps) of 229 macadamia under the current climate. This was based on the weights which were generated during model 230 calibration. To generate the absence-presence layers, we used the maximum sensitivity (true positive + ) and 231 maximum specificity (true negative -) approach [62], where we reclassified the distribution maps to binary 232 maps (suitable and unsuitable areas). In [48], [59], [63], it was shown that this method is one of the most 233 reliable for choosing a reclassification method. In this analysis, sensitivity is the proportion of observed 234 presences correctly predicted and therefore is a measure of omission errors, whereas specificity represents 235 the proportion of correctly predicted absences and thus quantifies commission errors. 236 To create distribution maps for future bioclimatic conditions, we utilized the same procedure used in the 237 baseline suitability and presence-absence maps but utilized the climate information from each of the 17 238 future GCM for RCP 4.5 and RCP 8.5. Since no criteria exist to assess which of the GCMs best predict 239 future climate [64], by incorporating all 17 GCMs, we encompassed all possible changes in the distribution 240 of the macadamia species. To integrate the results of the 17 GCMs presence-absence layers into a single 241 layer, we used the criterion of likelihood scale [48], which requires at least 66% of agreement among GCMs 242 to keep the predicted presence or absence in a given grid cell. 243

Factors determining land suitability of macadamia in Malawi. 245
Our study has shown that precipitation-related variables were the most important in determining the 246 distribution and suitability of macadamia in Malawi. Precipitation of the driest month (9.69) was the 247 variable with the greatest relative influence on macadamia production. Possibly because of the sensitivity 248 of pod growth during this phase to water scarcity. Among the temperature variables, isothermality (this 249 variable is calculated by dividing mean diurnal temperature range by mean annual temperature range) was 250 the most significant, with a VIF score of 8.95 ( 251 Table 1). Based on our ensemble model, annual means did not influence macadamia suitability in Malawi. 252 13

Current suitability of macadamia in Malawi. 254
Results of the present (~1970-2000) suitability analysis showed that 53,925 km 2 (57.4%) of the surface area 255 in Malawi is suitable for macadamia production (Fig 4), while 40,524 km 2 (42.6%) is unsuitable for the 256 crop. Therefore, our findings demonstrate that currently, macadamia is grown in a broad range of

Gain and loss of suitability under future projections in Malawi. 276
Compared to current climate conditions, the extent of suitable areas for macadamia production is expected 277 to decrease in the future under both emission scenarios utilized in this analysis. Our results revealed a net 278 loss of −18% and −21.6% of potentially suitable land for macadamia production under RCP 4.5 and RCP 279 15 8.5, respectively (Fig 5). This translates to 17,015 km 2 (RCP 4.5) and 20,414 km 2 (RCP 8.5) of Malawi's 280 total cultivatable surface area. Areas located in lower altitudes (500-1000 m.a.s.l.) will suffer the greatest 281 decline in suitability due to the projected general temperature increases and reduced precipitation amounts 282 and distribution. These losses will be more pronounced in Malawi's southern region areas, especially those 283 along the shire valley. Further, some southern region areas will become marginal or even unsuitable for 284 macadamia, while others will remain suitable though less than today. Thyolo district, which is currently the 285 country's most productive and biggest macadamia growing area, is predicted to suffer significant reductions 286 in suitability areas due to climate change. This is attributed to southern Malawi's low-lying nature and high 287 risks of heatwaves, flooding, and droughts linked to the El Niño Southern Oscillation. land, translating to 476 km 2 of Malawi's total surface area. We observed that projected newer areas will be 297 more under RCP 8.5, amounting to 0.28% more than RCP 4.5. The reason being that some of the very cold 298 areas currently unsuitable for macadamia will become suitable due to the projected increased warming by 299 the scenario RCP 8.5. The newer areas are predicted to occur in Dedza (Mua and Chipansi), Mangochi 300 (Namwera and Chaponda), and Ntcheu (Tsangano and Bonga) districts based on both emission scenarios. 301 Nevertheless, these apply only to very limited areas in the country and cannot compensate for the suitability 302 decrease in the lowlands. Our analysis, therefore, shows that the results for the RCP 4.5 and RCP 8.5 models 303 are similar in direction, but the RCP 8.5 models project a greater reduction in suitable areas in warmer 304 locations and expansion of suitable areas in colder locations by the 2050s. 305

309
Our results suggest a northward shift in the location of the most suitable areas for macadamia production, a 310 reduction of highly suitable areas in the south, and an increase along the central and northern parts of 311 Malawi, dependent on the landscape topography (Fig 6). Areas projected to lose their suitability occur 312 mainly in Malawi's southern regions, including Blantyre, Chikwawa, Chiradzulu, Machinga, Mwanza, 313 Mulanje, Thyolo, and Zomba districts. The projected loss is approximately 95-100% of the currently 314 suitable areas in southern Malawi. This is attributed to the projected increases in temperature and frequency 315 of droughts in the areas. Nevertheless, some higher elevated areas (≥ 1600 m.a.s.l.) within Chitipa (Misuku 316 hills), Ntchisi (Malomo and Kalira), and Rumphi (Mphompha and Ntchenachena) districts will similarly 317 lose some of their suitable areas for macadamia production. 318  variables at a national scale determined in this study might be different from climate indicators at the 325 regional and global scale. A previous researcher showed that the annual mean temperature, the warmest 326 month's maximum temperature, minimum temperature of the coldest month, and annual precipitation were 327 viewed as climatic regionalization indices for macadamia in Nepal. In the present study, we observe that 328 precipitation-based factors are more valuable in determining the suitability of macadamia in Malawi, 329 verifying zoning studies for macadamia production done for the country [38]. In these studies, precipitation 330 distribution and quantity were identified as the most critical variables, and it is apparent when considering 331 the current macadamia belt that water supply is the most limiting factor for the crop. Consequently, 332 projections that climate change will reduce rainfall amounts making its distribution unreliable in many parts 333 of Malawi, especially the southern region [9], will drive many areas out of macadamia production. 334 The distribution of precipitation is more related to precipitation of the warmest quarter and precipitation 335 seasonality. In this regard, areas with sufficient and sustainable water supply during the drier months of the 336 season (May-November) will remain suitable for macadamia. However, the regions with low annual 337 precipitation and soils with poor water holding will lose their macadamia production suitability. Our projections that climate change will increase the number of hot days (30.5) and hot nights (40) [11] will 358 certainly reduce the number of suitable areas for macadamia production in Malawi due to increased 359 warming, which will result in increased evapotranspiration rates. Taking this into account, trees currently 360 grown in the hotter areas will require sufficient water availability to cater to the water lost through 361 evapotranspiration. In Australia, Nepal, and South Africa, studies have shown that high daytime and high 362 night-time temperatures are responsible for the reduction in yields and suitable production areas for 363 macadamia [36], [39], [66], therefore agreeing with our current findings in Malawi. Consequently, climate 364 change will have dual impacts on macadamia production by reducing suitable production areas and reducing 365 the nut yield and quality. 366 Due to its geographic location and socio-economic status, Malawi is most exposed to climate change [7], 367 Furthermore, our findings suggested the suitability of macadamia in Malawi's south-eastern parts, such as 372 in Luchenza, Katunga, and Nsabwe, which are beyond the current reported production areas and considered 373 to be too hot for the crop. This is expected as the suitability maps capture the potential production areas, 374 some of which have not yet been translated to realized areas [51]. Additionally, this illustrates the broad 375 adaptability of some macadamia cultivars that allows its production from high potential areas to marginal 376 and low input areas with several environmental constraints. Nonetheless, these areas are the most vulnerable 377 to climate change because of limited buffering potential. 378 20 Malawi is already falling outside the prescribed optimal range for macadamia production, attributing it to 379 climate change. This is evident by the 0.9 o C increase in annual mean temperature and overall drying 380 recorded in the past five decades [6], [67], [68]. As a result of the projected temperature increases, changing 381 rainfall patterns, and increased water scarcity, the suitability for macadamia production in Malawi is likely 382 to decrease in the 2050s and is expected to shift northwards. Differences in loss-gain of suitability highlight 383 which agro-ecological zones could be more vulnerable to climate change (Fig 7). According to our 384 predictions, lowland areas will be the most affected (due to inadequate rainfall), with the central and 385 northern highlands even improving capacity to sustain macadamia production in areas where this is not From our results, we observe that the extent of suitable areas for macadamia production in Malawi will 395 decrease over the next 40-50 years. Our analysis reveals that the currently suitable areas in the southern 396 region will be the most affected, while areas located along the country's central and northern parts, 397 dominated by highlands, will become more favorable for the crop. The greatest victims will be areas 398 currently experiencing a hotter and drier environment (Fig 8). Consequently, these results show the 399 sensitivity of macadamia to variations in ecological conditions. Our findings confirm and, more 400 importantly, extend the work by [38], who found an inverse relationship between increases in temperature 401 Farmers are, therefore, encouraged to start implementing adaptation measures such as the use of improved 413 macadamia varieties, agroforestry, intercropping, water conservation, and irrigation for long-term and 414 sustainable macadamia production. Nonetheless, these suitability changes are predicted to occur over the 415 next 40-50 years, so these will mostly impact the next generation of macadamia farmers rather than the 416 current generation. Therefore, there is still time for adaptation. Failure to adapt in time to the risk of 417 decreasing yields and incomes may lead to the migration of rural populations to the main cities of Blantyre, 418 Lilongwe, and Mzuzu. 419 Altitude provides an excellent climatic change comparison for health, growth, and yield of crops [74], [75]. 420 As a result, individual plants grow very well in high altitudes, whereas others can only grow in middle or 421 lower-altitude areas [76]. Comparing the current and future suitable areas for macadamia production in 422 Malawi reveals an upslope shift in suitability. Our ensemble model showed that low-lying areas at altitudes 423 ranging from 500-1000 m.a.s.l. will have a decline in macadamia suitability because of the projected general 424 temperature increases and more dryer conditions. This is primarily true to Malawi's southern districts, 425 mainly those along the lake and shire valley (Blantyre, Mwanza, Neno, Mulanje, Chikwawa, Thyolo, and 426 Zomba). This is in line with the predicted losses in land suitable for tea production in the same region 427 (Mulanje and Thyolo) due to projected increases in warming and droughts [21]. Similarly, in their global 428 study of coffee suitability, [76] reported that climate change might lead to large losses of areas suitable for 429 22 the coffee across the globe, mostly in low altitudes below 1000 m.a.sl. However, we established that some 430 higher elevated areas (≥ 1600 m.a.s.l.), such as some parts of Chitipa, Nkhatabay, Ntchisi, and Rumphi, will 431 lose suitability due to predicted cold temperatures (≤ 4 o C) and frequent and intense rainfall (≥ 1750 mm). 432 This reduced suitability is attributed to the high levels of cloud cover experienced in these areas, which 433 results in lower light intensity reaching the leaves of the trees, thus affecting the total net photosynthesis for 434 tree growth and oil accumulation. Our findings coincide with [77], who found that suitable areas for 435 macadamia production decreased after an increase in altitude of over ≥ 1400 m.a.s.l in Thailand. Despite 436 large areas losing suitability, our findings show that some areas will gain suitability for growing macadamia. 437 This will generally depend on the landscape topography and will occur in the mid-altitude areas as suitability 438 moves upslope to compensate for increased temperature. Nevertheless, this only applies to minimal areas 439 within Malawi and cannot compensate for the decrease in suitability.  [79]. 450 Therefore, the need for a thorough evaluation of adaptation approaches suggested for smallholder 451 macadamia farmers, as these may be different from those utilized by commercial growers. 452 It is known that SDM development, particularly for areas with varying topographical terrains such as that 453 of Malawi, is challenging due to the complexity of the local and regional climate gradients [82]. Hence 454 23 careful interpretation is required when utilizing our results for the local effects of the future predictions on 455 macadamia production in Malawi. For agricultural land use planning, our results must be interpreted with 456 the knowledge of soil nutrition and social-economic factors. In addition, not all macadamia cultivars may 457 be similarly affected by climate change. We recommend that further studies need to be conducted to evaluate 458 the effect of climate change on the trait combination of the various cultivars available in Malawi. This will 459 ensure that the right cultivar is grown in the right place to maximize yields. Model building is another limiting factor when considering to assess the distribution of species in an area 475 due to different forms of uncertainties that may be incurred during this process [79]. We utilized the 476 automated model calibration method for our analysis as it is embedded with novel modelling frameworks 477 [79]. Using the automated approach, we eliminated sources of uncertainty such as collinearity and model 478 overfitting, which are associated with other methods of model building, such as that of the "priori selection 479