Modelling habitat suitability for Moringa oleifera and Moringa stenopetala under current and future climate change scenarios

Moringa oleifera Lam and Moringa stenopetala (Baker f.) Cufod are being widely promoted as multipurpose trees across the tropics for their nutritional, medicinal and soil health benefits. Different parts of these species are edible, have therapeutic values and their seeds are used for water purification. Although the two species are similar in many ways, they have contrasting distributions. However, their current promotion is not guided by adequate knowledge of the suitability of the target areas. Information is also scanty on the suitability of habitats for these species under the current and future climate change scenarios. Therefore, the objective of this study was to predict the habitat suitability of M. oleifera and M. stenopetala under current and future climate change scenarios using an ensemble of models assuming four shared socio-economic pathways, namely, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5 for 2050 and 2070. The results suggest that areas that are highly suitable for M. oleifera will increase by 0.1% and 3.2% under SSP1-2.6 to SSP5-8.5 by 2050, respectively. By 2070, the area suitable for M. oleifera would likely decrease by 5.4 and 10.6% under SSP1-2.6 and SSP5-8.5 scenarios, respectively. The habitat that is highly suitable for M. stenopetala was predicted to increase by 85–98% under SSP3-7.0 and SSP5-8.5 scenarios by 2050 and by 2070, while suitable areas could increase by up to 143.6% under SSP5-8.5. The most influential bioclimatic variables for both species were mean diurnal temperature range, mean temperature of driest quarter, precipitation of wettest month, and isothermality. Additionally, soil pH, elevation and water holding capacity were influential variables in the distribution of M. oleifera, while soil pH, soil salinity and slope were influential in M. stenopetala distribution. This study has provided baseline information on the current distribution and possible future habitat suitability, which will be helpful to guide formulation of good policies and practices for promoting Moringa species outside their current range.

Moringa oleifera Lam. and Moringa stenopetala (Baker f.) Cufod.are among the 13 or more known species in the monogeneric family Moringaceae 1 .Although the two species are similar in many ways, they have contrasting distributions, which makes them interesting both from a theoretical and practical perspective.How some species have wide distribution, while others are restricted in their distribution has fascinated biogeographers and ecologists for years.While M. oleifera is native to the Indian subcontinent, it is extensively cultivated in countries like the Philippines, Cambodia, and the Caribbean Islands 2 .On the other hand, the native range of M. stenopetala is restricted to a small part of East Africa including Ethiopia and northern Kenya [3][4][5] .Indeed, M. stenopetala is only known in the wild from northern Kenya, but it is widely cultivated in Ethiopia 4 .Reports from Djibouti, Malawi, Senegal, Somalia, Sudan and Uganda are probably based on recent introductions 4 .
Both species are promoted in the tropics as multipurpose species in agroforestry systems to provide animal fodder, human food and climate change mitigation 6,7 .For example, the Food and Agriculture Organization (FAO) has been promoting Moringa in agroforestry programs since the 1990s 8 .Moringa trees have several uses

Relative variable importance
In the ensemble model, the mean temperature of the driest quarter (Bio 9), precipitation of the wettest month (Bio 13), and mean diurnal temperature range (Bio 2) were the most influential bioclimatic variables in the potential distribution of M. oleifera.Isothermality (Bio 3), annual temperature range (Bio 7), and temperature seasonality (Bio 4) were highly influential in the case of M. stenopetala (Fig. 9).Among soil variables, soil pH, elevation and soil water holding capacity had higher influence on the distribution of M. oleifera, whereas soil salinity, slope, and soil pH were more influential for the current distribution of M. stenopetala.The clay and sand contents had the least contribution to the distribution of both Moringa species (Fig. 9).

Discussion
This study has established the current and future distribution of M. stenopetala and M. oleifera.The strength of this study lies in the use of an ensemble modelling approach to increase the accuracy of the predictions and the use of different performance metrics to assess the models' performance 35 .The results have also pointed out areas which are likely to be affected in the event that they become invasive.In such cases, practitioners need to be prepared to manage invasions in areas identified as highly suitable.
The distribution of M. oleifera and M. stenopetala is mainly concentrated in the West and East African countries like Kenya, Uganda, Ethiopia, parts of Sudan, and in parts of North American countries like south of Mexico, Guatemala.In contrast, the distribution of the Moringa tree is limited in countries like Botswana, Namibia, South Africa, Swaziland, and Zambia 36 , although the extent of suitable areas identified in these countries by our modelling was large.The larger areas covered by M. oleifera may be explained by its fast growth and adaptation to a wide range of climatic and soil conditions.M. oleifera takes short time to first flowering, which is about 11 months from planting 37 compared to the 2-2.5 years taken by M. stenopetala from planting to first flowering 37 .In addition, M. oleifera has been shown to withstand a wide range of precipitation conditions (annual rainfall of 250-3000 mm) 17 .On the other hand, the ideal range of rainfall required for the growth of M. stenopetala is between 500 and mm 38 .M. oleifera is also adapted to a wide range of elevations (0-2000 masl) 6 compared to M. stenopetala   (400-1200 m) 39 .Our models have predicted that M. oleifera can occur at the elevation of 100-3000 masl.Furthermore, M. oleifera can tolerate and grow well under temperatures ranging between 12.6 and 40 °C, while M. stenopetala grows better where temperatures are 24-30 °C.M. oleifera grows better in well-drained clay or clay loam soils without prolonged waterlogging 38 .While M. oleifera can grow in soil with a wide pH range (5-9), M. stenopetala mostly grows in soils with neutral reaction 38,40 .
Our model predicted that the overall suitable area of M. oleifera will be reduced, although some expansion may occur by 2070.The habitat suitability of M. stenopetala will decrease under SSP1-2.6,and only Ethiopia and parts of Kenya will be highly suitable for its future distribution.This is because some regions are projected to experience a decrease in temperature and increases in precipitation.These changes may result in a reduction in the area suitable for some species, particularly those adapted to warmer and drier conditions.This is because the cooler and wetter conditions will become more favourable for other species to thrive, displacing species adapted to warmer and drier conditions.As a result, suitable areas for some species may shrink.M. stenopetala appears to be favoured by future climate change scenarios as the highly suitable area tended to increase substantially by 2050 and 2070 under SSP5-8.5.SSP5-8.5 is expected to lead to increased variability and extreme events.These changes may result in an expansion of suitable areas for some species, particularly those adapted to warmer conditions.Similarly, a study on the habitat suitability of Ficus squamosa and Ficus heterostyla revealed a decline in their overlapping areas.However, among the two species, F. heterostyla demonstrated a greater potential for climate change adaptability.This confirms that many species may detect climatic changes and respond in various ways 41 .
Our results showed relatively high current and future overlap in the distributions of M. oleifera and M. stenopetala.Several factors favour the overlapping of both the Moringa species.Slightly saline soil of 1-2 dS m −1 , water holding capacity of 20-40%, a slightly sloppy area between 10°-30°, isothermality between 65-90% and mean temperature of driest quarter of 20-35 °C are some of the favourable factors for the growth of both M. oleifera and M. stenopetala.As a consequence, the overlapping area between M. oleifera and M. stenopetala may tend to increase under current and future climate conditions.
Our model predicted that the distribution of both species is influenced by multiple factors like soil pH, soil salinity, elevation etc., which is in agreement with previous reports.A recent study has reported that elevation is one of the factors affecting the distribution of Moringa species, as it directly or indirectly influences the temperature and soil characteristics 42 .Additionally, the slope of a site was a very influential factor in the distribution of Moringa species.The majority of Moringa habitat was found on gravelly slopes and rocky mountainous areas with a slope of 40-60% indicating that they prefer well-drained soils 42 .Soil texture, bulk density, and soil porosity are the controlling factors affecting the WHC of the soil 43 , and the WHC of soil can affect the growth of plant species.Although M. oleifera is a drought tolerant tree, WHC was one of the significant factors affecting its growth.M. oleifera can grow well on soil having more than 70% WHC compared to soil having a lower WHC 44 .Our result revealed that WHC between 20-50% can favour the growth of M. oleifera.In changing climatic conditions, temperature and precipitation may severely alter the habitat suitability of these species.
The current distribution of M. oleifera and M. stenopetala was also highly influenced by soil pH and soil salinity.Soil pH in the natural environment has an enormous influence on the soil biological, chemical and physical properties that influences plant growth 45 .In our analysis, soil having a pH of 6.5-8 had the highest contribution in the current distribution of M. oleifera.However, previous studies have suggested that the Moringa tree can grow well in soil with pH between 4.5 to 8.5; and pH between 6.3 to 7 can improve the growth of the Moringa tree 33,46,47 .Soil salinity hinders plant growth by lowering leaf water potential, causing physiological and morphological alterations, producing reactive oxygen species, raising osmotic stress and ion toxicity, and changing biochemical processes 48 .However, it was reported that the Moringa trees can grow well and germinate at lower soil salinity 49 of 5 dS m −1 .
Our results indicate that currently suitable areas may likely become unsuitable for the growth of M. oleifera and M. stenopetala in future climate change scenarios.A similar finding was also reported for a economically important plant Pinus gerardiana.Where the models predicted a remarkable decline in the potential habitat suitability in the future climate change scenarios 50 .Climate change influences the physiochemical, and biological properties of soil which relates to the functional properties of soil.Drivers of climate change affects the organic matter status, carbon and nutrient cycling, plant available water, and hence productivity, which in turn affect soil pH 51 .Furthermore, due to climate change, there will be spatial and temporal changes in temperature and rainfall 52 .These temperature changes may affect evapotranspiration, including the evaporation of water from soils.As a result, the salinity of the soil will increase that will hinder the growth of plants 53 .Additionally, erratic rainfall can greatly affect the groundwater table and soil salinity.Lower level of groundwater table can instigate capillary rise which causes the upward movement of salts from the water table to the soil surface, resulted in the accumulation of salinity at or near the soil surface, and may cause salinity stress in the plants 42 .Moreover, extreme events of anthropogenic climate change can also alter the WHC of soil; the WHC may decrease in warmer climate in the future 54 .This may possess a severe impact on the distribution of the species because WHC is an important factor for plant growth and can compensate for a lack of precipitation in dry years 43 .
The results of this study have significant implications for research and development.The baseline information on the current distribution will be helpful to guide future research on the two species.In terms of development, the information provided on habitat suitability would be useful in guiding the targeting of the species to areas where they can be successfully introduce.Several international organizations, governments, and NGOs are making efforts to popularise these species in different agroforestry interventions to improve productivity and human nutrition 55 .Governments of several developing countries like Ghana, Cuba, and India have focused on Moringa for combating malnutrition and encouraging its cultivation 56 .However, identifying suitable areas for the plantation and growth is crucial.Output of the present study can be considered for a more effective implementation of these policies through promotion of these species in in areas suitable for their planting.The Moringa trees can act as a good carbon sink, as they produces heavy flushes even during the dry season, and can help in reducing the level of atmospheric carbon dioxide (CO 2 ) 57 .Furthermore, the Moringa tree can absorb fifty times more CO 2 compared to the Japanese cedar tree dominated vegetation and twenty times more CO 2 than other vegetation 58 .In future climate change scenarios, if the highly suitable area for Moringa trees increases, it will also help in the adaptation of these trees for their various beneficial properties.However, to achieve the different goals, these Moringa species need to be planted in suitable niches with careful consideration of their potential for invasivenes.We recommend concerned bodies to promote the species in suitable areas to guarantee success.

Conclusions
Using an ensemble model, we predict the current distribution and future habitat suitable for M. oleifera and M. stenopetala.The analysis has provided evidence that ensemble models can accurately predict the distributions of both M. oleifera and M. stenopetala.It is concluded that the habitat that is highly suitable for M. stenopetala will increase under future climate change scenarios.In contrast, the overall suitable area for M. oleifera will be reduced.Nevertheless, some expansion in the suitable area is likely to occur by 2070.It is also concluded that  soil pH and water holding capacity were the major predictors of the current distribution of M. oleifera, whereas soil salinity, elevation and slope are key predictors of the current distribution of M. stenopetala.This study has provided baseline information on the current distribution and possible future habitat suitability.The results are hoped to help researchers, policymakers and practitioners to make informed choices when selecting areas for the promotion of these species under current and future climate change scenarios.Future research should aim to produce fine-scale climate projections that account for regional changes in temperature, precipitation, and extreme weather events.This would enable more precise estimation of viable habitats under various climate change scenarios.In addition, dynamic modelling and multi-species modelling of Moringa species with other species will also be beneficial in understanding the migration and adaptation patterns of Moringa species and how their habitat suitability interacts with other plant species.The maps we provided are also hoped to help in identifying appropriate niches for the cultivation of Moringa species for greater production.

Study area
The present study covers tropical regions (from the Tropic of Cancer to the Tropic of Capricorn), comprising 122 countries.Of these, 51 are from Africa, 19 are Asian, 27 are North American, 13 are Oceanian, and 12 are from South America.This region accounts for about 36% of the earth's landmass, and the mean annual temperature in the tropics ranges from 25 to 28 °C59 .

The target species
Moringa oleifera is commonly known as a 'drumstick tree' or 'horse-radish tree.' Moringa oleifera is a fast-growing, drought-resistant, deciduous, dicotyledonous tree with a height of 5-10 m 60 .Moringa oleifera can grow well in the humid tropics and hot, dry lands 11 and endure a range of rainfall from 250-3000 mm and a pH of 5-9 61 .www.nature.com/scientificreports/This species can be found at 0-1000 masl elevation and can be adapted to various soil types 38 .The tree has a soft trunk, gummy bark, and a tripinnately compound leaf 62 .
Moringa stenopetala is popularly known as the African cabbage tree, a strongly branched tree with a thick base with white to pale grey or silvery bark.Its trunk can grow up to 60 cm in diameter at its breast, and the tree has smooth wood and soft leaves 63 .In Africa, Moringa stenopetala naturally grows with the Acacia tortilis-Delonix elata-Commiphora spp.vegetation complex and can be found at an altitude of 400-2100 m.M. stenopetala has no specific soil requirement for its growth 38 .

Geo-location data
The data on the geo-locations of M. oleifera and M. stenopetala was collected from Global Biodiversity Information Facility (GBIF) 64 and published in peer-reviewed literature (Fig. 1a; Supplementary Tables S1 and S2).The GBIF is an international organisation that focuses on making scientific data on biodiversity available via the Internet using web services.
The literature survey was carried out using the search engine "Google Scholar".Search strings were created comprising the keywords "Moringa oleifera, " "Moringa stenopetala, " "drumstick tree, " "horse-reddish tree, " individually or in various combinations such as "Moringa oleifera in Asia" or "Distribution of Moringa stenopetala".Deka et al. 65 used similar data curation methods to examine the possible effects of climate change on the distribution of the endangered white-winged wood duck (Asarcornis scutulata, 1882) in the Indian Eastern Himalayan region.A total of 1739 (n = 133 literature, n = 1606 GBIF) locations were collected, out of which 1692 were for M. oleifera, and 47 were for M. stenopetala which were considered for niche modelling.However, to prevent the  www.nature.com/scientificreports/model from over-fitting, the numerous present points were removed from one grid cell 66 .The spThin 67 package was used in the R software version 4.1.3(R Core Team, 2022).Eventually, 652 occurrence records of M. oleifera (n = 100 literature, n = 552 GBIF) and 43 occurrence records of M. stenopetala (n = 33 literature, n = 10 GBIF) were considered to build the models.These are listed in Supplementary Table S1.

Environmental data
The standard 19 bioclimatic variables and elevations were downloaded from WorldClim version 2.1 68 with 30 Arc seconds, which are the average for the years 1970-2000 69 .These bioclimatic factors have often been utilized in SDMs for climate prediction based on tree species 70 .Moreover, 14 soil data layers were accessed from the International Soil Reference and Information Centre (https:// www.isric.org) 71 .Slope data was created from the DEM in the ArcGIS spatial extension.The spatial resolution of all the environmental variables was 30 Arc seconds (~ 1 km).The complete list of variables are presented in Supplementary Table S2.
Among 35 variables (Table S2), a few (18 in M. oleifera and 24 in M. stenopetala) were filtered using variance inflation factor (VIF) to avoid the effect of multicollinearity (Supplementary Table S3).We used R package called "usdm" to calculate VIF.Values greater than VIF 10 signify that an unacceptably large quantity of collinearity has been eliminated 70 .Therefore, the modelling was performed using 17 variables for M. oleifera and 11 for M. stenopetala.Some variables were calculated from others.For example, mean diurnal temperature range was calculated as the difference between mean of monthly maximum temperature and minimum temperature.

Future climate scenario data
To predict habitat suitability under future climate scenarios, we used the Coupled Model Intercomparison Project Phase 6 (CMIP6) downloaded from the WorldClim dataset.The CMIP6 is a recent climate projections data with the same spatial resolution of current period data from WorldClim for 2050 (average for 2041-2060), and 2070 (average for 2061-2080), under four shared socio-economic pathway (SSP) scenarios (i.e., SSP1-2.6,SSP2-4.5, SSP3-7.0, and SSP5-8.5).Among these scenarios, SSP3-7.0 was the new scenario combinations, and the SSP1-2.6,SSP2-4.5 and SSP5-8.5 were the updated version of the RCP scenarios 71 .In this study, we exclude SSP4-6.0 as it has the highest inequality, followed by SSP3-7.0.Both SSP1-2.6 and SSP5-8.5 feature relatively equitable development and a rapid catch-up of the world's poorest countries over the coming century 73 .SSP1-2.6 is the lower end of radiative forcing expected to produce < 2 °C warming, and for brevity here we will refer to it as "low" emission scenario.SSP2-4.5 represents the medium range of future pathways, and for brevity here, we will refer to it as "medium".SSP5-8.5 represents the high end of the range of future pathways 72,73 , and here we refer to it as "high" climate change/emission scenario.SSP2-4.5 and SSP5-8.5 project global temperature anomalies of 2.4 °C and 4.9 °C above pre-industrial levels by 2100 with atmospheric CO 2 equivalents of 650 and 1370 ppm, respectively 74,75 .Finally, the SSP dataset were used for model the Moringa sp. for future climate scenarios.

Species distribution modelling
We applied an ensemble of six models implemented in the 'sdm' package 76 in R (4.1.3version).The models included one regression and five machine learning methods.We selected the multivariate adaptive regression splines (MARS) regression model and the support vector machine (SVM), boosted regression trees (BRT), random forest (RF), classification and regression tree (CART) and maximum entropy (MaxEnt) machine learning methods for their high predictive accuracy 77 .
The model was trained with 10 replicates and evaluated per algorithm through five cross-validations.The occurrence data were split into training (70%) and test (30%) data to explore the implications of different environmental variables.Presence-absence models used equal numbers of presences and pseudo-absences, within the grid cell around presence locations 78 .Model accuracy and validation were judged by the True Skill Statistic (TSS) and the area under the curve (AUC).AUC values typically range from 0 to 1, with values closer to 1 indicating a more potent model 79,80 ; AUC values < 0.7 are considered poor, 0.7-0.9moderate, and > 0.9 good 81,82 .In contrast, TSS is a threshold-dependent measure of model accuracy, with values of + 1 indicating complete agreement between predictions and observations and 0 or below indicating agreement no better than random classification 83,84 .TSS value was classified as poor (< 0.40), fair (0.40-0.55), good (0.55-0.70), very good (0.70-0.85), excellent (0.85-0.99) and perfect (0.99-1.0) 85,86 .The Pearson correlation coefficient (COR) and deviance statistic were also considered to evaluate the model performance.COR estimates the correlation between continuous prediction with observation 87 , and the Deviance statistic calculates the deviation between observed and fitted values 88 .
Based on the potential species distribution habitat suitability index (0-1), the ensemble output maps were divided into four classes: unsuitable (0.00-0.2), least (> 0.2-0.40),moderate (> 0.4-0.6), and high potential (> 0.6-1.00) 89.The reclassified tool in ArcGIS 10.8.2 was used to classify the ensemble maps.A field calculator was used to determine the raster area of each class of map (Count area × Area pixel/1,000,000).The relative variable relevance for each model's present prospective Moringa distribution was established using the averaged variable response curve.The percentage contribution of each variable to the ensemble model was used to determine the variables' relative relevance.It draws attention to significant environmental factors that had a key role in shaping the geographic range of the species.

Figure 1 .
Figure 1.Map of locations of data used for the species distribution modelling (a) and the current distribution of Moringa oleifera and Moringa stenopetala in the tropical regions.All maps were generated by authors of this work using ArcGIS 10.8.2 (https:// www.arcgis.com/ index.html).

Figure 2 .
Figure 2. Changes in habitat suitability area in current and future (2050 and 2070) climate scenarios for M. oleifera and M. stenopetala.

Figure 7 .
Figure 7. Changes in the overlapping area of Moringa oleifera and Moringa stenopetala under SSP1-2.6 and SSP2-4.5 by 2050 and 2070.All maps were generated by authors of this work using ArcGIS 10.8.2 (https:// www.arcgis.com/ index.html).

Figure 8 .
Figure 8. Changes in the overlapping area of Moringa oleifera and Moringa stenopetala under SSP3-7.0 and SSP5-8.5 by 2050 and 2070.All maps were generated by authors of this work using ArcGIS 10.8.2 (https:// www.arcgis.com/ index.html).

Table 1 .
Performance evaluation of SDMs using different statistical parameters for the current distribution of Moringa oleifera and Moringa stenopetala in Tropical countries.

Table 2 .
Ecological niche overlap of suitable habitat between M. oleifera and M. stenopetala.

Table 3 .
Characterization of highly suitable areas for M. oleifera and M. stenopetala.