Impact of Climate Change on Species Distribution and Carbon Storage of Agroforestry Trees on Isolated East African Mountains

Changes in climate will affect conditions for species growth and distribution, particularly along elevation gradients, where environmental conditions change abruptly. Agroforestry tree (AGT) species on the densely inhabited slopes of Mount Kilimanjaro and Taita Hills will change their elevation distribution, and associated carbon storage. This study assesses the potential impacts of climate change by modelling species distribution using maximum entropy. We focus on important agroforestry tree species (Albiziagummifera, Mangiferaindica and Perseaamericana) and projected climate variables under IPCC-AR5 RCP 4.5 and 8.5 for the mid-century (2055) and late century (2085). Results show differential response: downward migration for M. indica on the slopes of Mount Kilimanjaro is contrasted with Avocado that will shift upslope on the Taita Hills under RCP 8.5. Perseaamericana will lose suitable habitat on Kilimanjaro whereas M. indica will expand habitat suitability. Potential increase in suitable areas for agroforestry species in Taita Hills will occur except for Albizia and Mango which Climate change will affect AGT species and the amount of carbon stored dif-ferently between the sites. Such insight can inform AGT species choice, and conservation and support development by improving carbon sequestration on sites and reliable food production.


Introduction
Atmospheric accumulation of anthropogenic greenhouse gases, in particular a doubling of CO 2 , is linked to the rise in global temperature estimated at an average of 1.5˚C [1]. Increase in global temperature has caused shifting of species distribution, population structure and abundance towards the pole [2]. Species on montane ecosystems with steep climate gradient are at risk from global warming due to potential upward shift of species ranges and habitat fragmentation [2]. Climate change is envisaged to change potential species distribution area and survival in the 21st century; though at present, land use by human population remains the main driver of species extinction and habitat loss [3].
Climatic conditions have changed in the recent years in East Africa due to natural climate variability and land use change [4]. The temperatures in the region have risen by approximately 1.3˚C since 1960 [5]; similarly, rainfall distribution and quantity have varied over space in the Horn of Africa [6]. Although an increase in average precipitation over the region is predicted [7], incidences of increased drought and floods have been recorded alongside seasonal variation of precipitation [7] [8]. Climate change projections indicate annual temperatures will increase by 1.8˚C to 4.3˚C by 2080 in East Africa with greatest warming occurring from June to August [7]. Projected climate change is likely to affect East Africa's socio-economic sectors, particularly those reliant on land. Where greater rainfall variability within and between seasons occurs, crop productivity will considerably be affected. For instance, maize and certain bean varieties may decline in yield [7]. Subsistence farming is most vulnerable to climate change due to lack of sufficient resources to adapt to climate change [9].
The agroforestry system is increasingly gaining importance in increasing food production, enhancement of crop productivity, soil enrichment and wider mitigation effects of climate change through enhanced carbon sequestration and other ecosystem services [9] [10] [11]. Agroforestry increases amount of carbon sequestered in the above and belowground compared to a monoculture field of crop plants or pasture [10] [12] [13]. Improvement of cropland management practices such as management of trees increases carbon sequestration [10] [14].
Agroforestry plays an important role in enhancement and maintenance of long-term soil productivity and sustainability by improving amount of organic matter and releasing and recycling of soil nutrients [10] [15]. Agroforestry sys- tems have high potential to conservation due to their structural complexity, high floristic diversity and close resemblance to forest ecosystems [16]. Biodiversity conservation by agroforestry is through provision of habitat for species and connectivity of scattered habitats; preservation of germplasm of sensitive species, and ecosystem services [10]. For instance, the shade coffee agroforestry system has high potential to enhance biodiversity compared to traditional agricultural practices [10] [17] [18].
Less effort is directed in understanding how species of agroforestry system would be affected by climate change despite the attributed importance on carbon sequestration and biodiversity conservation. Insight is required to establish extent to which species would be affected in order to plan for their future conservation and management [19]. Robust prediction models for impact of climate change on agroforestry systems are not available [20]; however prediction of habitat suitability can be made by models that employ species location [21] via species distribution modelling (SDM) which provides an alternative approach to projecting climate change impacts [20]. The potential species distribution areas are considered to possess conditions that are suitable for survival of the species and can be used to estimate species' realized distribution [22]. Thus, agricultural systems can be evaluated with methods typically used for studying organisms.
One challenge with exotic agroforestry trees is that they are they are often grown outside their normal environmental niche so the record of where these are found can be a challenge to fit a modelling framework. For instance, most of agroforestry trees consist of exotic species that are not sufficiently covered by most Natural History Museums/Botanical Garden herbarium collections. Thus, the number of sightings for the species (training examples); will often be small, a hundred or less [22]. When occurrence records for the species are distributed within a small geographic area, modeling affects accuracy of coinciding the species realized niche and fundamental niche [23]. Lastly, existing environmental data does not have suitable resolution that synchronizes with the local climate variables [20] [24]. Bioclimatic conditions often vary widely over short distances, especially in mountainous terrain, which is characterized by steep gradients often not reflected in the available datasets [19] [20]. In order to solve this challenge, a high-resolution bioclimatic variable (downscaled) can be used in order to capture local topographic characteristics that affects climate in the mountainous areas [24]. The issue of small sample size can be solved by using cross-validation which uses all the data for validation that makes better use of small data sets [23]. Accuracy for species fundamental niche along the elevation gradients can be improved by using data from a larger geographical extent and zooming along the slopes of the mountains [23]. for mid-century 2055 (2041-2070) and the late century 2085 (2071-2100) [24].
The IPCC models and scenarios are used for projections of future climatology.

Study Area
The study sites used in this research were the Taita Hills and Mount Kilimanjaro year. The long rains take place from March to May and the short rains from October to December. In Taita Hills over 1400 m a.s.l., the average rainfall is more than 1300 mm annually, while the surrounding plains have between 400 to 700 mm [27]. Due to the orographic rainfall pattern, the southeastern slopes of Taita Hills receive more precipitation than the northwestern slopes [28].

Species Distribution Modelling
Distribution data for Avocado, Mango and Albizia spanning over landscapes in Kenya and Tanzania landscape ( Figure 2) were acquired from the field data and online database specifically from the Global Biodiversity (7), Replicate type (subsample), Replicated run type (crossvalidate), Maximum iteration (5000). Cross-validation has an advantage of using all data for validation, thus making better use of small data sets [23]. Other parameters were set at maxent default values. Fewer numbers of parameters were used for calibration since tuning large numbers of parameters for modelling often poses several challenges for instance, it may be prohibitive and time consuming to tune the method on each species separately. Besides, model accuracy is affected by several settings that determine the type and complexity of dependencies on the environment that maxent tries to fit [23]. Relative influence of the climate variables was assessed for each species in each climate models (baseline, RCPs and periods). This is performed by examining the contribution of each predictor to the final regularized training gain when all variables of the particular model were included in the Maxent run [34]. Maxent model performance was measured by the test omission rate which is considered good when the omission rate is close to the predicted omission. The area under the ROC curve (AUC), was used for measuring the performance of the maxent model.

Species Above-Ground Carbon Calculation
Biometric data for avocado, mango and Albizia were extracted from plot carbon data that were collected from the research transects. The plot carbon data were generated using a standard size plot of 1 ha [35] within which heights and diameter at breast height (dbh) ≥ 10 cm of woody trees were sampled. A number of 12 Plots were established randomly within 20 km long transects in Taita Hills and Kilimanjaro in 2012 ( Figure 1) [25]. The initial step of estimating the aboveground carbon is determining the aboveground biomass using the tree parameters (DBH & height) measured on a plot and the wood density for each species (g/cm 3 ). We retrieved the wood densities for each species, genus or family from the Global Wood Density Database [36] [37]. An allometric model developed by Chave et al., (2014) [38] was used for estimating the above-ground biomass where ρ is the Wood Specific Gravity, D is the dbh, and H is the height of the tree. After deriving the AGB for each tree, data was aggregated to plot level from which 50% of AGB was assumed to be carbon sequestration in a plot [25] [38] [39].

Prediction of Suitable Area and Carbon Storage
Selection of suitable areas for Albizia, Mango and Avocado on the inhabited slopes of Mount Kilimanjaro and Taita Hills was performed on the probability distribution layers at probability ≥ 0.6. The probability value of ≥0.6 was selected as a threshold value based on the predicted baseline probability value of first occurrence point for Albizia on the slope of Mount Kilimanjaro. Once suitable areas for the species are derived with a threshold 0.6 probability distribution, potential shift of minimum elevation range for the species were assessed within transect buffer of 16,970 ha along the elevation. The elevation at the minimum suitable elevation range of a species suitable contiguous or fragmented polygon areas of the species were used to determine the shift of species distribution along the elevation. Shift in minimum suitable elevations range of the species was derived from the difference in predicted minimum suitable elevation range of projected climate change, IPCC-AR5 (RCP 4.5 and RCP 8.5) in the mid-century 2055 and late-century 2085, from the baseline prediction. While, change in area of distribution was indicated by percentage of increase or decrease in area of prediction of projected climate change, IPCC-AR5 (RCP 4.5 and RCP 8.5) in the mid-century 2055 and late-century 2085, and the baseline prediction.
The above-ground carbon (AGC) was predicted on assumption that the aver-  (2)). The total amount of species AGC depends on the average species plot AGC and the total area predicted as suitable on the slope of the mountains. Thus, total amount of AGC is normalized by the total area of the transect buffer to derive predicted mean species AGC on the slopes of Mount Kilimanjaro and Taita Hills (Equation (3)).

Statistical Analysis
The area under the ROC curve (AUC), was used for measuring the performance of the maxent model. The AUC is the probability that a randomly chosen presence site will be ranked above a randomly chosen absence site [23]. A random ranking has an average AUC of 0.5, and a perfect ranking achieves the best possible AUC of 1.0; models with values above 0.75 were considered potentially useful [40]. Suitable areas and AGC for a species were compared between projected baseline species distribution and future projected periods using Chi 2 test.
Tables and map models were used for presentation of analysis of suitable areas, elevation shift of species and carbon storage for the species on the slopes of the mountains.

Species Elevation Shift
The

Change in Suitable Areas and Above-Ground Carbon Storage
An estimated area of 77% of the transect area is potentially suitable for Albizia followed by Avocado (69%) and Mango (67%) under the baseline prediction in Taita Hills. In Kilimanjaro, species that had highest potential suitable area along the slope was Albizia with 39% of the area, followed by Mango (36%) and Avocado (28% and Avocado (65%). In Kilimanjaro slope, 37% of the transect area is predicted potentially suitable for Albizia, Mango apparently has the highest area in the transect (59%) predicted potential for distribution with Avocado having low area of 32% predicted for potential distribution. Comparison of suitable areas between Kilimanjaro and Taita Hills predicted under the baseline climate condition for Albizia, Mango and Avocado shows significant difference (F = 153.17, p = 0.01). While, no significant difference was observed among the three species on suitable areas within site. Under the RCP 4.5, 2055 Climate Projection comparison of the suitable areas for the three species between Kilimanjaro and Taita Hills was significantly different (F = 303.76, p = 0.00); while no significant difference was observed among the three species on suitable areas in each sites. The projection of climate change based on the RCP 8.5, 2055 show suitable areas for each species between Kilimanjaro and Taita Hills differed significantly (F = 216.96, p = 0.00). While no significant difference was observed among the suitable areas for the three species in each sites. Suitable areas for Albizia, Mango and Avocado in Kilimanjaro and Taita Hills significantly differed in size (F = 393.02, p = 0.00) based on the projection of climate change based on the RCP 4.5, 2085. Also, significant difference (F = 28.12, p = 0.03) was observed among the three species on suitable areas in each sites. Based on RCP 8.5, 2085, potentially suitable areas for Albizia, Mango and Avocado in Kilimanjaro and Taita Hills do not differ significantly in size, and; no significant difference was observed among the species in each site.
Change of predicted suitable areas analyzed against predicted baseline area and AGC shows decrease in areas for most of the species on the slopes of Mount Kilimanjaro for the RCPs and periods of climate change projection. The slopes of Taita Hills will experience increase of predicted areas for the agroforestry tree species (Figure 4) but AGC will apparently decrease in size (kg/ha./yr) except for Avocado that will increase in Taita Hills. Avocado will decrease in suitable areas more than other species in Kilimanjaro under RCP 4.5 for periods 2055, 2085 and RCP 8.5 in 2055 (Figure 4). AGC for Avocado will as well decrease in size in the same RCP and periods. However, much decrease in AGC will occur in 2055 and 2085 under RCP 4.5 for Albizia on the slopes of Mount Kilimanjaro.
Considerable suitable areas will increase for the species under RCP 4.5 and 8.5

Discussion
Agroforestry system supplement carbon sequestration in the above and below ground sectors [41]. The system; however has less carbon than indigenous and plantation forest but they certainly have more carbon than in agricultural cropped land [25] [42]. Besides, the system provide an array of products including fruit food, timber for building, fuelwood, and ecosystem services [43]. The  (Table 3). Even though precipitation is expected to increase in East Africa in the late century from the current 5% to about 20% [24], variation will occur across the landscape. These variations will affect how species will respond to  [44] predicted that some species will shift downslope and this will be driven by changes in seasonality and water availability.
This study highlights a relative increase in temperatures on the slopes of Mount Kilimanjaro and Taita Hills. However, there is difference in precipitation between the two sites where Taita Hills will experience relatively higher amount of precipitation than in Kilimanjaro (Table 3). This means there will be more water available in Taita Hills than on the slopes of Kilimanjaro. On the other hand, a potential increase of the mean annual temperature on the slope of Mount Kilimanjaro will be accompanied by a relatively stable precipitation except in 2085 under RCP 8.5 when precipitation will be relatively high ( Table 3).
Plants that will be in areas with an increased warming but relatively stable precipitation will be affected adversely on growth and restricted distribution. Water availability is the primary determinant of vegetation distribution across the landscape due to its importance on the growth of plant species [45]. The increased Avocado and Albizia in Taita Hills ( Table 4).
The spatial patterns of woody covers and other vegetation are strongly controlled by the frequency, extreme events of precipitation and the amount of rainfall [45]. In this study, carbon storage of the selected agroforestry trees is also observed responding to the gains and loss of climate variables. Carbon storage of the selected agroforestry trees species is predicted to decrease where: 1) there will be gains of temperature and loss of precipitation variables, and 2) probably, where some important precipitation variables loss and less important variable to species growth gain. In these scenarios, reduction in species carbon storage will take place whether there is upshift or downshift, or no shift in species distribution. The projected increase in species carbon storage will occur due to gains in precipitation variables and the minimum temperatures. The projected upslope movement of Avocado is associated with the gains in temperature variable which probably causes reduced cold condition in higher elevation of the inhabited slopes. This creates optimal growth temperature condition in the upland but increases the minimum suitable elevation, reduces suitable areas in the lower elevations and causes species carbon storage to decrease. The response of species to climate is also observed by Platts et al., [44] that forest cover will increase in uplands due to increased temperature; however lower limits of cover is not explained. Some of the findings contradicts observation that phenomenal species shift induced by climate change is likely to affect adversely the area for species distribution, the conservation and agricultural crop in tropical montane areas [9]. Increase in temperatures with a relatively stable or increased precipitation in the lowland will cause downshift of Mango on the slopes of the mountains (Table 4) which increase the population and production of the species. A contrasting response of species to climate change is observed on Albizia between Kilimanjaro and Taita where the species will shift upwards and decrease in carbon storage in Kilimanjaro probably due to the gains in temperature variables in the upland. In contrast, Albizia will shifts downwards in the lowlands of Taita Hills with relatively stable carbon storage due to the gains in monthly maximum temperature in Taita Hills (Table 4).
The choice for agroforestry tree species for improvement of carbon storage on the slopes of Mount Kilimanjaro and Taita Hills will depend on the species response to the projected climate change scenario. Species that increases in carbon storage under the two scenario and periods would be a good candidate for farmers. While considering that upshift and downshift of species is inevitable; the main objective would be to choose species that no matter what, carbon storage will improve on the slopes of the mountains. The future potential priority species for increasing carbon storage on the slopes of Taita Hills in both uplands and lowland is Avocado which shows potential relative increase in suitable areas under RCP 4.5 in 2055. Climate change under scenario RCP 8.5 will provide the inhabited slopes of Mount Kilimanjaro with a potential increase in carbon storage for Mango in 2055 and 2085. Species that will increase in suitable areas where precipitation variables will gain but show no shift in elevation can improve carbon sequestration by additional land management techniques ( Table  4). The selected agroforestry tree species for modelling restricted choice of species for carbon sequestration on the inhabited slopes of Mount Kilimanjaro and Taita Hills under different climate change scenario. Consideration of several species with varied contribution of carbon sequestration will provide a holistic strategy for managing carbon storage on the slopes. The success in modelling of the agroforestry tree species therefore provide information that potentially contribute to conservation and development of agroforestry resources for improvement of carbon sequestration. In addition, local farmers would be able to adjust and putting up appropriate measures for ensuring reliable food security and income from agroforestry trees.
The use of species occurrence data and climate model variables in this study potentially present uncertainty in interpretation of the results. Dataset acquired from observations and herbarium often shows strong geographic bias (sampling bias) due to some areas being visited more often than others because of their accessibility [46] [47]. The availability of presence distribution data for species poses challenge as most of herbaria in the region lack sufficient data for modelling [47]. Such challenges has however been overcome by use of cross-validation in maxentmodelling technique which uses few data points. Species occurrence data can be biasedly distributed on the landscape which can contribute to local biasness and over-smoothing affecting reliability of the model [46] [48]. We, however, rely on maxent ability in using jackknifing to achieve a robust new estimator, called jackknife kriging, which retains ordinary kriging simplicity and global unbiasedness while at the same time reducing local bias and over-smoothing tendency [48]. The climate model consist of potentially highly correlated variables; however, maxent can only select one of variables in a pair of highly correlated variables and still model performance is not affected [22] [34]. However, the selection of variables by maxent is associated with the risk of diminishing importance of other predictor in the pairs of variables [22] [34]. It is also important to note that there are confounding factors that may affect the distribution of agroforestry tree species on the slopes of Mount Kilimanjaro and Taita Hills.
These factors include but not limited to soil, market forces and land use plan and management.

Conclusion
Climate change will cause an increase in the mean annual temperature across the landscape of Kilimanjaro and Taita Hills under RCP 4.5 and 8.5 by the period 2055 and 2085. The inhabited slopes of Taita Hills will, however, experience relatively higher precipitation but the slope of Kilimanjaro will be relatively stable in precipitation. These phenomenal changes in climate will cause some species to shift upslope, downslope or no directional shift along the slopes. Variations in gain and loss of particular climate variable at local scale will induce a unique difference in the response of agroforestry tree species. In particular, the upshift of species distribution is associated with the reduction in species carbon storage on the slopes of Mount Kilimanjaro and Taita Hills which apparently is explained by the gains in temperature variables. An exception is observed on Albizia in Kilimanjaro which shift upslope but increases in carbon storage. Downshift of species is mostly associated with the gain in precipitation; however spe-American Journal of Climate Change cies carbon storage would increase or reduce on the slopes of Mount Kilimanjaro, or no change in species carbon storage. Avocado will shift upwards under RCP 8.5 and 4.5 in 2055 and 2085, respectively. The upshift will cause reduction in area of distribution which will adversely affect species carbon storage in Kilimanjaro and Taita Hills. While under RCP 4.5, avocado will shift downslope in Kilimanjaro and Taita Hills by 2055 though species carbon will decrease in the former but increase in the latter. Response of species to other RCPs and period contrast between the two sites. Upshift of agroforestry species will cause decrease in carbon storage and affect adversely livelihood of local population that depend on the agroforestry resources on the slopes of the mountains.