Biome Change in Southern Africa

Abstract

Biomes are regional to global vegetation formations characterised by their structure and functioning. These formations are thus valuable for both quantifying ecological status at sub-regional spatial scales and defining broad adaptive management strategies. Global changes are altering both the structure and the functioning of biomes globally, and while detecting, monitoring and predicting the outcomes of such changes is challenging in Southern Africa, it provides an opportunity to test biome theory with the goal of guiding management responses and evaluating their effectiveness. Here, we synthesise what is known about recent and expected future biome-level changes from Southern Africa by reviewing progress made using dynamic global vegetation modelling (based on archetypal plant functional types), phytoclime modelling (based on species-defined plant growth forms) and phenome monitoring (based on the seasonal timing of vegetation activity). We furthermore discuss how monitoring of indicator species and indicator plant growth forms could be used to detect and monitor biome-level change in the region. We find that all the analysis methods reviewed here indicate that biome-level change is likely to be underway and to continue, but that the analytical approaches and methods differed substantially in their projections. We conclude that the next phases of research on biome change in the region should focus on reconciling these differences by improving the empirical opportunities for model verification and validation.

1 Introduction

Biomes are conceptual constructs that categorise terrestrial ecosystems into structural and functional units and thereby help us organise our knowledge on how ecosystems work (Moncrieff et al. 2016). Despite their importance, there is surprisingly little consensus on how to define biomes (Moncrieff et al. 2016). Most biome schemes use a combination of plant growth forms, leaf phenology and sometimes climate to recognise formations that include Forest (evergreen, deciduous and mixed), Savannas (mixed tree and grass formations), Shrublands, Grasslands, Deserts and high elevation or high latitude Tundra. Whittaker (1975), refining ideas developed by Schimper (1903) made the case that biomes are strongly dependent on climate, while observing that it was not possible to predict the dominant biome in seasonally dry, subtropical climates, which happen to cover vast areas of the planet. Bond (2005) addressed this conundrum, hypothesising that fire and herbivores override climate forcing in the subtropics by preventing vegetation from attaining its “climatic potential”, while (Walter 1973) drew attention to how soils and orographic processes may also cause vegetation to deviate from climatic potential.

The manifold impacts of anthropogenic climatic change make it extremely likely that biomes are changing in character and distribution, given that there is ample evidence for wide-scale redistribution of global biomes under the changing climates of the Pleistocene and earlier epochs (Huntley et al. 2021). Indeed, several studies have used satellite imagery to detect changes in vegetation cover and ecosystem functioning in recent decades that are large enough to qualify as biome shifts (Seddon et al. 2016; Higgins et al. 2016; Buitenwerf et al. 2015, Song et al. 2018; Zhu et al. 2016). Such shifts can have large impacts on biodiversity, for example, the number of endemic species has been shown to be higher in areas of relative past biome constancy (Huntley et al. 2016, 2021).

One way of forecasting biome shifts is to use Dynamic Global Vegetation Models (DGVMs). DGVMs are ambitious models that seek to simulate how resource assimilation, growth, competition and consumption (fire, herbivory) processes interact over ecological time scales and thereby how biomes might shift in response to changes in the climate system (Prentice et al. 2007). DGVMs also account for the impacts of changes in atmospheric chemistry, in particular the plant physiological impacts of enhanced CO${ }_{2}$ levels (which increases carbon uptake through photosynthesis and reduces water loss through transpiration Walker et al. 2021). The aDGVM is a dynamic global model that has been specifically developed to model the biome boundaries between forest, savanna and grassland and is therefore particularly suited to modelling the Southern African sub-region (Scheiter and Higgins 2009; Scheiter et al. 2012). How DGVMs define biomes however varies considerably. In the aDGVM, the relative cover of C3 grasses, C4 grasses, savanna trees and forest trees is used to define biomes.

An alternative method for forecasting biome shifts is to use a data-driven approach to represent plant growth forms. Conradi et al. (2020), for example, estimated the climatic suitability of geo-locations for plant growth forms typically used to define biomes. This was achieved by parameterising physiological growth models for 23,500 African plant species categorised into the growth forms, projecting the distribution of climatically suitable geo-locations for each species and then calculating the proportion of species of each growth form for which a geo-location was suitable. This proportion was used to characterise the climatic suitability of a geo-location for each growth form. This approach allows the researchers to forecast changes in the ability of a geo-location’s climate to support different plant growth forms. The climate suitability of a geo-location for different plant growth forms describes its capacity to support different types of plants that ecologists use to define biomes. The vector of growth form suitability scores can then be used to classify different geo-locations into groups. The groups can be conceptualised as phytoclimes, where a phytoclime is defined as a geographic region where the climate’s influence on growth form suitability is similar. Phytoclimes align conceptually with Holdridge (1947) and Whittaker (1975) who emphasised links between climate and vegetation formations even if phytoclimes establish these links differently. Using the term phytoclime emphasises that such analyses model the potential of a region’s climate to support different plant growth forms; phytoclimes are not equivalent to biomes. Rather, the phytoclimate describes which plant types could potentially grow at a location, whereas a range of processes ignored by phytoclimes such as competition, facilitation, herbivory, predation, dispersal and historical contingencies determine how climate suitability translates into vegetation formations and biomes.

Southern Africa is a challenging arena for forecasting biome shifts because climate and consumption (fire and herbivory) processes interact to shape the regions’ major vegetation formations. South Africa alone has 9 recognised biomes (Mucina and Rutherford 2006), and the biome concept has for decades shaped and structured both the practice of ecological science in the country and the national environmental policy. This is well illustrated by the biodiversity vulnerability assessment, which was part of the South African country study on climate change (Rutherford et al. 1999). The Rutherford et al. (1999) study developed the case that changes in the climate forces that regulate the distribution of the biomes of South Africa would lead to a large and dramatic re-organisation of the country’s vegetation. A combination of effective science communication and the severity of the impacts predicted in the study served to stimulate an intensification of research on climate change impacts in the region. Perhaps more importantly, the report has had a sustained impact on South African climate change policy. Here, we review recent analyses of how climatic change may impact on the distribution of the vegetation of Southern Africa.

The forecasts of predictive models such as DGVMs or phytoclime models have uncertainties originating from multiple sources. These include uncertainty on the key processes represented in the models, the parameterisation of the processes, the Global Circulation Models (GCMs) used to forecast future climates that ecologists use to force their ecological models and the global emission scenarios that these GCMs assume (Thuiller et al. 2019). It follows that such models need to be critically evaluated using independent information. Indeed, much can be inferred about the trajectories of change that ecosystems are on by observing the recent past using remote sensing and on-the-ground monitoring. For this reason, we would suggest that modelling approaches endeavour to provide output in a form that can be compared to monitoring data. In particular, we highlight the potential of using the MODIS satellite record and field-based monitoring using Biome Shift Monitoring Phytometers (BISMOPs) and indicator species.

2 Phytoclimes

2.1 Phytoclime Methods

A phytoclime analysis of Southern Africa (Higgins et al. 2023) used species distribution data from the National Vegetation Database (Rutherford et al. 2012), ACKDAT (Rutherford et al. 2003) and BIEN version 4.1 (http://bien.nceas.uscsb.edu), thereby combining South Africa’s excellent vegetation data legacy with a leading global data base on plant species distributions. These distribution data and the CHELSA 2.0 climate data (Karger et al. 2017), which scales CMIP6 climate projections to a 1 km resolution, were used to parameterise a process-based physiological plant growth model, the TTR-SDM (Thornley Transport Resistance Species Distribution Model Higgins et al. 2012), for 5006 species. The model considers how spatial and monthly variations in temperature, soil moisture, solar radiation and atmospheric CO${ }_{2}$ concentrations influence the growth of plant species. The model fitting procedure estimates the influence of these environmental factors on growth that are consistent with the observed distribution data. The fitted models were projected in geographical space to identify climatically suitable grid cells for each species. The growth form of each of these species was then used to group species by growth form. The analysis assigned species to the following growth forms, trees, shrubs, C3 grasses, C4 grasses, restioids, geophytes, annual forbs, other forbs, succulents and climbers using several data bases: C4 grass data base—Osborne et al. (2014), succulent plant data base—Eggli and Hartmann (2001), Eggli and Nyffeler (2020), POSA—http://newposa.sanbi.org, BIEN—http://bien.nceas.uscsb.edu and GIFT—Weigelt et al. (2020).

The average of the geographical projections of the climate suitability scores of each species belonging to a growth form was then used to estimate the suitability of the sub-region for each of the 10 growth forms (Fig. 14.1). The growth form suitability scores of grid cells were then classified, using unsupervised classification, to yield phytoclimes—geographical regions where the climate favours plant growth forms in similar ways.

Fig. 14.1

Ambient suitability surfaces for growth forms derived from the TTR-RED-LD model. The preference scores are the averaged preference scores of the species belonging to each growth form, transformed to scale between 0 and 1

Using CHELSA 2.0, climate projections derived from the CMIP6 (Eyring et al. 2016) programme future (225, 2055, 2085) shifts in phytoclimes were projected. These projections considered uncertainty in TTR-SDM model used (different TTR-SDM variants make different assumptions, 4 variants were considered), the GCM used (5 GCMs were considered ) and the SSP assumed (3 SSPs were considered).

2.2 Phytoclime Findings

The suitability surfaces for each growth form (Fig. 14.1) revealed patterns that are broadly consistent with prior knowledge, and we highlight a few of these patterns here. The arid region associated with the Namib was generally unsuitable for all growth forms, as were the higher lying parts of Lesotho. In general, suitability for growth forms gradually increased from west to east (Fig. 14.1). C3 grasses had preferences for the south coast and eastern coast of South Africa, and this preference area extended over the escarpment and into the highveld areas including the Soutpansberg mountains and Inyanga mountains in Zimbabwe. C4 grasses by contrast showed a preference for the north eastern part of the sub-region. Both grass types revealed a low preference for the arid central and western parts of the sub-region, although this trend was stronger for C3 than for C4 grasses. Perennial forbs had a lower preference for an arid area starting at the Cunene River mouth and extending towards the south coast. Climatic suitability for annual forbs was often higher than for perennial forbs, and in particular they revealed less aversion to the aforementioned arid area. Geophytes had a high preference for the south coast and grassland regions, including the Soutpansberg mountains in South Africa and Inyanga mountains in Zimbabwe. Restioids exhibited the same preferences as geophytes, but a more distinct aversion to almost all other parts of the sub-region. Succulents exhibited preferences for the Cape provinces of South Africa, central Namibia and a region centred on the intersection of the border between South Africa, Botswana and Zimbabwe. Shrubs showed preference for the southern coast regions and the central east of the sub-region, and they showed low preferences for the arid regions of Namibia and the Kalahari. Trees showed a preference for the north eastern part of the sub-region and had an aversion to an arid triangle that extends from the Cunene River mouth to the southern coast. Climbing plants had a similar preference surface to trees.

Additional insight can be gathered by examining how the growth form suitability values shown in Fig. 14.1 change over time (from the present to the end of the century). Figure 14.2 shows the rate of change (% change in suitability per 100 years) in growth form suitability averaged over the different TTR-SDM variants, SSPs and GCMs. The rates of change were frequently as high as 35% per 100 years, forecasting a fundamental change in the influence of climate on the region’s vegetation. The most striking trend is that eastern South Africa is forecast to experience an increase in suitability for C4 grasses, trees and climbers and to a lesser extent shrubs, succulents and annual forbs (Fig. 14.2). Furthermore, Lesotho is forecast to be climatically more suitable for all growth forms in the future, suggesting that these high-lying parts of the Drakensberg mountains could serve as a climate refuge for many species and growth forms if they could migrate there. A similar trend can be seen in the Cape Fold mountains. A comparison of Figs. 14.1 and 14.2 suggests that C4 grasses, which under ambient conditions had an intermediate preference for the central plateau, will find this region more suitable in the future. These same areas will become less suitable for C3 grasses. C4 grasses were also forecast to find the fynbos, karoo and Namib more favourable in the future. C3 grasses were predicted to find the south coast, an area for which they have a high ambient preference, less suitable in the future. Perennial forbs, which under ambient conditions had a low preference for an arid triangle from the Cunene River mouth to southern coast, were predicted to find this area slightly more suitable in the future. Geophytes were largely predicted to face decreases in climate suitability except in Lesotho and surrounding areas. Restios were predicted to face a marked decrease in climatic suitability in the areas in which they currently have a high climatic suitability (the south coast). Projected loss and gain patterns in climbers and trees resembled those projected for C4 grasses. Succulents were projected to face losses in their ambient high-preference areas such as the Succulent Karoo and the area around the intersection of the borders between South Africa, Botswana and Zimbabwe). Shrubs were projected to find the Drakensberg and surrounding areas more suitable, but parts of the bushveld of South Africa, Botswana and Zimbabwe as well as the Albany Thicket biome of the Eastern Cape less suitable.

Fig. 14.2

The mean rate of predicted change in growth form suitability (% change in suitability per 100 years) for the 10 growth forms considered in this study. The rate of change is calculated assuming a linear change in preference over time (from the present to the end of the century) and using TTR-variant, GCM and SSP as covariates

When the ambient growth form preferences (Fig. 14.1) are classified into phytoclimes, it is possible to produce a range of phytoclime maps. Figure 14.3 illustrates phytoclimes for the region when assuming 6–32 phytoclimes. There is no a priori reason why the region should have 6 or 32 phytoclimes. That is, this analysis explicitly acknowledges that phytoclimes are abstractions designed to help us understand climate’s influence on a region’s vegetation ecology, but they do not represent an absolute truth. In this analysis, 24 different phytoclime maps were generated (4 TTR-SDM variants $\times$ 6 numbers of phytoclimes), which quickly complicates the interpretation of how phytoclimes may change since there are 3 emission scenarios (SSPs) to consider, and these scenarios have been simulated by different global circulation models (we consider 5 GCMSs). Figure 14.4 provides an illustrative example of phytoclime change that uses a 6-phytoclime map generated using the TTR-RED-LD model variant, the SSP 585 scenario as predicted by the gfdl-esm-4 GCM for a climatology centred on the year 2085. This single scenario illustrates that phytoclime 6 (climate that currently supports grassland and fynbos) will decrease in area, mostly due to losses to phytoclime 5 (climate currently supporting savanna) (Fig. 14.4). Figure 14.5 reveals that the area currently occupied by phytoclime 5 will lose territory to phytoclime 4 (which currently supports more arid savannas). The area currently under phytoclime 5 will become increasingly unsuitable for all growth forms, particularly perennial forbs and C4 grasses. Similarly, the area currently under phytoclimate 6, which currently suits all growth forms, is predicted to lose suitability for restioids, C3 grasses and geophytes and acquire an enhanced suitability for C4 grasses, annual forbs, climbers, succulents, trees and shrubs.

Fig. 14.3

Ambient phytocolime maps derived from the TTR-RED-LD model for 6, 9, 12, 18, 24 and 32 clusters. The clustering algorithm detects multivariate discontinuities in the growth form suitability surfaces displayed in Fig. 14.1 to delimit the phytoclimes

Fig. 14.4

Ambient and future (climatology centred on year 2085, SSP 585, GCM gfdl-esm4) phytoclime maps of Southern Africa derived from the TTR-RED-LD model for a 6-phytoclime classification. The right-hand panel summarises the phytoclime transitions between the two maps

Fig. 14.5

Growth form suitability for the 6-phytoclime map derived from the TTR-RED-LD model under ambient conditions and the change in the suitability scores in the ambient phytoclime regions forecast for the year 2085 under SSP 585 using the gfdl-esm4 GCM (as in Fig. 14.4). The values in the left panel express the normalised average climatic suitability for the growth forms in the phytoclimes and the values in the right panel show their change

2.3 Synthesis of Phytoclime Change Scenarios

To synthesise phytoclime change, Higgins et al. (2023) recorded the time point of phytoclime changes observed between the ambient climatology and climatologies centred at 2025, 2055 and 2085. Using a Kaplan–Meier estimator, the mean time to phytoclime change was estimated for each location. The averaged mean year of phytoclime change, averaged over GCM, SSP, TTR-SDM variant and biome classification scheme (i.e. 6, 9, 12, 18, 24 or 32 phytoclimes) is shown in Fig. 14.6. This average year of phytoclime change is to be interpreted as the relative time point at which the climate forcing which defines the phytoclimes is sufficient to force a change into another phytoclime (cf. Fig. 14.5); this interpretation emphasises that realised change in vegetation will lag behind climate forcing. This analysis revealed a strong continentality trend, with the centre of the region forecast to change earlier than the coastal regions. This suggests an overriding effect of temperature which is forecast to change more severely in the interior than in coastal regions (Engelbrecht and Engelbrecht 2016).

Fig. 14.6

The average time to phytoclime transition averaged across TTR-SDM model variants ($n=4$), SSPs ($n=3$), climate model ($n=5$) and the number of phytoclimes assumed ($n=6$) using a Kaplan–Meir survival estimator

The patterns summarised in Fig. 14.6 average away some important sources of variation. Anchor Environmental, under stakeholder review explored the phytoclime change trends in more detail for South Africa. They found that very few areas are expected to change by 2025 under both the SSP 126 and the SSP 585, with relatively modest changes across phytoclime configurations by 2055 under both pathways. However, by 2085, the changes in phytoclimates are forecast to be widespread. Very few parts of the country experienced no change at all with only parts of the Fynbos and Desert Biomes appearing to have no change by 2100. The likelihood of phytoclime change for SSP 1–2.6 (the Sustainability pathway) and SSP 5–8.5 (the Fossil-Fuelled Development pathway) is remarkably similar up until ca. 2055. By 2025, the likelihood of phytoclime change appears to be mostly limited to a few isolated areas of the Kalahari Duneveld and Eastern Kalahari Bushveld Bioregions in the North West and Northern Cape Provinces of South Africa. The likelihood of phytoclime change increases notably by 2055. By 2055, a high change score region extends from the Kgalagadi to the south and east. Moderate to high change scores are also predicted for large areas of the Grassland Biome (particularly, the Dry and Mesic Grassland Bioregions) by 2055. Moderate to high change scores are also predicted for large areas of the Central Bushveld Bioregion spanning the Savanna Biome in the North West and Limpopo Province. Phytoclime change scores are higher under SSP 5–8.5. Change to South Africa’s coastline occurs later, with projected phytoclime change scores remaining low at the 2055 time slice, which may be due to the cooling effect of the proximity to the ocean. The Succulent Karoo, Nama-Karoo, Albany Thicket and Fynbos biomes overall are likely to be least affected. The northern half of the Kruger National Park in Limpopo also appears to remain relatively unchanged in the 2055 time slice.

By the 2085 time slice, the phytoclime change scores are high throughout South Africa, only a few areas, such as patches of mountainous Fynbos in the Western Cape, and isolated areas of Namaqualand and the Desert Biome remain unchanged. Lower change scores were derived for many areas of the Eastern Cape and Southern KwaZulu-Natal (Sub-Escarpment and Drakensberg Grassland) and the Bushmanland Bioregion of the Nama-Karoo. The central interior of the country had very high phytoclime change scores. The Lowveld and Sub-escarpment Savanna areas of Mpumalanga, Limpopo and KwaZulu-Natal exhibited large increases in change scores between 2055 and 2085.

Overall, the Savanna and Grassland Biomes have the highest phytoclime change scores. This is perhaps partly due to their size, occupying 59% of South Africa’s terrestrial extent. This matches findings by Midgley et al. (2011) who identified the Orange River Basin and Highveld plateau as being climate change hotspots.

For planning, it is useful to combine spatial information on both the likelihood of ecological change due to climate change and the level of protection against other anthropogenic impacts. For this purpose, Anchor (2022) calculated a vulnerability index that considered the level of ecosystem health (DEA, 2019) and the level of conservation protection (DFFE, 2021). The resulting vulnerability indices were summarised by the recognised biomes of South Africa and contrasted for SSP1-2.6 (Fig. 14.7) and SSP 5–8.5 (Fig. 14.8). Across both SSPs, the extent of high vulnerability areas increases substantially over time, particularly between the period centred on 2055 and 2085. There is almost no difference between the SSPs in the 2011 to 2040 period, since protected areas ensure vulnerability remains very low in this time window. The most noticeable changes in the following period centred on 2055 are the increase in the extent of areas with high vulnerability > 0.5 in the Kalahari, Bushveld and southern Kruger National Park (all in the Savanna Biome). There is also an overall increase in vulnerability in the Bushmanland Bioregion of the Nama-Karoo. In the period centred on 2085, the extent of very high vulnerability ($>$0.75) increases dramatically, particularly under SSP 5–8.5 (under SSP 1–2.6, various areas in the Grassland and Savanna regions in Gauteng, Free State, Mpumalanga, North West and Limpopo remain slightly more stable). There are also substantial increases in vulnerability across KwaZulu-Natal and areas of the Northern Cape. The Desert Biome area, eastern parts of the Succulent Karoo, the mountainous areas of the Fynbos and eastern Fynbos Biome areas remain the most stable.

Fig. 14.7

The vulnerability index for change of vegetation under SSP 1–2.6 for the periods 2011–2040, 2041–2070 and 2071–2100 (left) and the biome vulnerability index (mean vulnerability index per biome) for each time period’s median year. The vulnerability index considers phytoclimatic change, land use change and protected area status

Fig. 14.8

The vulnerability index for change of vegetation under SSP 5–8.5 for the periods 2011–2040, 2041–2070 and 2071–2100 (left) and the biome vulnerability index (mean vulnerability index per biome) for each time period’s median year. The vulnerability index considers phytoclimatic change, land use change and protected area status

The increase in mean biome vulnerability index can also be seen between the SSPs with slight increases between 1–2.6 and 5–8.5. For SSP 5–8.5, in the earliest time period (2025), the Indian Ocean Coastal Belt (IOCB) has the highest biome vulnerability index with 0.52, followed by Grassland with 0.46 and Fynbos 0.36. The Desert and Albany Thicket have the lowest overall. These values largely reflect the existing extents of modified land in each of the biomes with the IOCB and Grassland having the highest values of all nine terrestrial biomes (Skowno et al. 2021). By 2055, Grassland, Savanna and IOCB have the highest biome vulnerability indices with values of approaching 0.55. The Desert and Albany Thicket biomes continue to have the lowest. By 2085, the biome vulnerability index of Savanna increases to 0.92, followed by Grassland (0.83) and IOCB (0.81). Desert remains the lowest but increases to 0.31 (still the only biome below 0.5) followed by Fynbos with 0.52. Similar between-biome differences are forecast under SSP 1–2.6 albeit with slightly lower scores.

3 Insights from DGVM Modelling of Biomes

3.1 DGVM Methods

Here, we analysed results from four different DGVMs, three global ones that were not adapted for the study region and the aDGVM originally developed for Africa (Scheiter and Higgins 2009; Scheiter et al. 2012). As the global models do not represent all major biome types in the study region, we first describe the methodology and results for aDGVM and provide more detail than for the global models. The aDGVM uses concepts and processes commonly used in other DGVMs such as leaf-level ecophysiology or the representation of the carbon cycle (Prentice et al. 2007). In addition, processes such as carbon allocation to different plant compartments or leaf phenology are adjusted based on environmental conditions. The aDGVM is individual-based and simulates growth, biomass, allometry, reproduction and mortality of individual trees. This approach enables simulations of the impacts of disturbances such as fire, herbivory or fuelwood collection on individual plants while relating these impacts to plant traits such as tree height or stem diameter. In contrast to trees, grasses are only simulated by super-individuals representing grasses between and under tree crowns.

The aDGVM simulates fire impacts on vegetation (Scheiter and Higgins 2009). Fire spreads if an ignition event occurs and if the fire intensity is high enough to carry fire. Days with ignition events are randomly distributed during a year, but the number of ignition events decreases with tree cover. This approach ensures that the likelihood of fire is high in open grassland or savanna vegetation but low in dense forests. Fire intensity is defined by fuel biomass, fuel moisture and wind speed. The aDGVM only simulates surface fire such that grass biomass is the main component of fuel biomass. Fire removes the entire aboveground grass biomass. Fire impacts on trees are related to tree height. Fire removes aboveground biomass of small trees in the flame zone, whereas tall trees survive fire. Both grasses and trees can resprout after fire and recover. By default, aDGVM simulates natural fire but anthropogenic management fire with fixed fire return interval and burning season can be simulated as well.

The aDGVM simulates four different plant functional types (PFTs): fire-tolerant savanna trees, fire-sensitive forest trees, C${ }_{4}$ grasses and C${ }_{3}$ grasses. Savanna and forest trees differ in their shade tolerance and fire tolerance. Shade tolerance is simulated by linking growth rates to light availability which is in turn influenced by light competition between neighboring trees and by shading. At low light availability, forest trees have greater growth rates than savanna trees. Fire tolerance is implemented by different topkill probabilities and resprouting rates, where savanna trees have lower topkill probabilities and higher resprouting rates than forest trees. Based on these assumptions, forest trees outcompete savanna trees in closed forest stands without fire, whereas savanna trees can survive in open and fire-driven environments. The difference between C${ }_{4}$ and C${ }_{3}$ grasses is simulated based on the physiological differences between C${ }_{4}$ and C${ }_{3}$ photosynthesis and shade tolerance. To simulate differences in shade tolerance, we used different light competition parameters for C${ }_{4}$ and C${ }_{3}$ grasses, which describe how shading effects by neighboring plants influence the light availability and photosynthetic rate of a target plant. The relative abundances of these four PFTs are influenced by the prevailing environmental conditions, competition between individual plants, and disturbance regimes.

Similar to suitability surfaces of different growth forms derived from the TTR-RED-LD model, relative abundances of the four PFTs derived from aDGVM simulations can be used to create suitability surfaces. In addition, vegetation simulated by aDGVM can be classified into different biome types using simulated model state variables. Following the classification scheme developed by Martens et al. (2021), vegetation is classified as desert if tree cover is below 10% and grass biomass is below 1.5 t/ha and as grassland if tree cover is below 10% and grass biomass is above 1.5 t/ha. Grassland is further separated into C${ }_{3}$ or C${ }_{4}$ grassland based on the relative abundance of C${ }_{3}$ and C${ }_{4}$ grass PFTs. If tree cover is between 10 and 80%, vegetation is classified as either woodland, if forest tree cover exceeds savanna tree cover, or as savanna, if savanna tree cover exceeds forest tree cover. This separation between woodland and savanna reflects the absence or presence of regular fire, which favours savanna trees. Savannas are further split into C${ }_{4}$ or C${ }_{3}$ savanna, depending on the abundance of grass PFTs. If the total tree cover (i.e. forest and savanna tree cover) exceeds 80%, vegetation is classified as forest. Hence, biome shifts, simulated for example in response to climate change or changes in land use, occur if simulated biomass of different grass PFTs or cover of different tree PFTs exceeds or falls below the respective thresholds used for biome classification. While this scheme represents important biomes of Southern Africa, it ignores biomes including Fynbos or Karoo. In the current version, aDGVM lacks PFTs required to simulate these biome types, including CAM plants, succulent shrubs and flammable shrubs (Moncrieff et al. 2015).

Martens et al. (2021) used aDGVM to study climate change impacts on future vegetation in Africa for an ensemble of climate change scenarios. Specifically, simulations were conducted for the period between 1971 and 2099 for RCP4.5 and RCP8.5. For each scenario, down-scaled climate forcing from six different GCMs was available. Down-scaling to a spatial resolution of 0.5° was conducted with the variable-resolution conformal-cubic atmospheric model (CCAM, McGregor 2005; Engelbrecht et al. 2015; Engelbrecht and Engelbrecht 2016). All details of the modeling protocol are provided by Martens et al. (2021). Here, we evaluate the results provided by Martens et al. (2021) focusing on the Southern African study region.

Several DGVMs were run globally with harmonised past to future environmental forcing and land use data within the Intersectoral Impact Model Intercomparison Project (ISIMIP; https://www.isimip.org). The modelling protocol is described in Frieler et al. (2017). Only three models produced outputs that enabled us to analyse results per PFT and to transform the results with land use into simulations of the potential natural vegetation, which we considered more relevant here than grid cell averages with land use. Therefore, we analysed results for these three models: LPJ-GUESS (Smith et al. 2014), ORCHIDEE (Krinner et al. 2005) and CARAIB (Dury et al. 2011). As the global models have not been adapted for Africa and do not represent all major biomes in the study region well, we only analysed mean results across all three DGVMs and available climate scenarios (four for LPJ-GUESS and CARAIB and two for ORCHIDEE, whereby results for each DGVM were weighted equally), and we used a rather coarse biome classification. More details of the analyses are described by Wilhelm (2021).

3.2 DGVM Findings

aDGVM simulations showed that most of the study region is suitable to support both C${ }_{3}$ and C${ }_{4}$ grasses, except the Namib region and Lesotho (Fig. 14.9) with low precipitation and low temperature, respectively. Suitability is generally higher for C${ }_{4}$ grasses than for C${ }_{3}$ grasses. Hence, in our simulations C${ }_{4}$, grasses dominate the grass layer except a small region in the Karoo (Fig. 14.10). The model simulated increasing suitability for tree PFTs from west to east with low suitability in the Namib region and the Karoo, intermediate suitability in the savanna regions of Namibia, Botswana, South Africa and Zimbabwe and high suitability along the south and east coast of South Africa and Mozambique. Suitable regions for savanna trees and forest trees are almost disjoint with little overlap in their distributions. While savanna trees dominate fire-driven savanna regions with intermediate tree cover, forest trees dominate forests in the East of the study region. This result is not surprising given that savanna trees in aDGVM are fire-tolerant and able to outcompete fire-intolerant forest trees in fire-driven regions, whereas forest trees are shade-tolerant and able to outcompete shade-intolerant savanna trees in dense forests.

Fig. 14.9

Suitabilities/Niches in 2000–2019 and their changes until 2080–2099 under RCP4.5 and RCP8.5 for plant functional types (PFTs) as simulated by aDGVM. PFTs simulated by aDGVM are C${ }_{3}$ grasses (c3g), C${ }_{4}$ grasses (c4g), fire-tolerant savanna trees (svt) and shade-intolerant forest trees (frt). Suitability surfaces (left) are based on maximum values in 2000–2019. For grasses, 0.9 times the maximum value of both grass PFTs was used to scale grid cells. For savanna trees, 0.9 times the maximum savanna tree canopy cover was used because inherently the savanna tree PFT rarely has a closed canopy. Forest tree canopy cover was used as suitability surface for forest trees. Changes are derived from the differences between 2080–2099 and 2000–2019. The simulation setup is described in Martens et al. (2021)

Fig. 14.10

Biomes in 2000–2019 and biome changes until 2080–2099 under RCP4.5 and RCP8.5 simulated by aDGVM. Grid cells were classified into biomes based on grass biomass, dominance of C${ }_{3}$ or C${ }_{4}$ grasses, tree cover, dominance of savanna or forest tree types. In the biome change subfigures, biomes that were simulated for 2080–2099 for grid cells where biome transitions were simulated are shown. This figure is based on Martens et al. (2021)

Under climate change, aDGVM simulates changes in the suitability of different PFTs. Suitability of C${ }_{3}$ grasses was predicted to increase in almost the entire study region. Suitability of C${ }_{4}$ grasses was predicted to increase in Namibia, Botswana and Mozambique, whereas it was predicted to decrease in Zimbabwe and most of South Africa. These results can be explained by CO${ }_{2}$ fertilisation effects that, in aDGVM, enhance C${ }_{3}$ photosynthesis but not C${ }_{4}$ photosynthesis. Suitability of savanna trees was predicted to decrease primarily in Namibia, Botswana and the North of South Africa, whereas it was predicted to increase in Zimbabwe and most of South Africa. Suitability of forest trees was predicted to increase in South Africa, but decreases were simulated in the forest regions of Mozambique. Taken together, these changes cause woody encroachment and an increase of woody biomass in the entire study region until the end of the century (not shown, Martens et al. 2021). Patterns of change of suitability were similar for RCP4.5 and RCP8.5 for all PFTs, but on average, changes were higher for RCP8.5.

Predicted changes in suitability of PFTs imply changes in simulated biome patterns (Fig. 14.10). Increases in the suitability of tree PFTs implied transitions from desert to woodland (along the coast of Namibia), from grassland to savanna or woodland (e.g. in the Karoo) and from savanna to woodland or forest (e.g. East of South Africa, Botswana). In Mozambique, both forest dieback and transitions to woodland as well as transitions from woodland to forest were predicted.

The ensemble mean of the three global DGVMs roughly reproduced the main distribution of biomes across the study region as reconstructed in a global PNV (potential natural vegetation) map. Major shrub biomes, such as the Nama Karoo, however, were not distinguished (Fig. 14.11). The models also confirmed the increasing suitability for tree PFTs from west to east predicted by aDGVM. For the future, the DGVMs predicted woody encroachment and increasing tree cover in particular in eastern parts of the study area, and much more pronounced under RCP6.0, as a result of higher CO${ }_{2}$ fertilisation effects on woody plants under this scenario. Biome shifts were also much more pronounced under RCP6.0 and concentrated in the eastern part of the study region and, under RCP 6.0, the northwestern fringe (Fig. 14.12). These results were roughly in line with those by the aDGVM, but they clearly differ for large parts of Zimbabwe, where the aDGVM predicted increasing climatic suitability for fire-tolerant savanna trees (Fig. 14.9), while the global DGVM ensemble predicted decreasing woody leaf area index (Fig. 14.13). These differences might have been caused by differences in vegetation process representations or by using different climate models as forcing. The general strong woody encroachment into savannas is consistent with results from other DGVM-based studies (Scholze et al. 2006; Gonzalez et al. 2010).

Fig. 14.11

Biomes according to expert reconstruction (reference data) based on Haxeltine and Prentice (1996); multi-model ensemble mean (3 DGVMs and 2–4 GCMs) for recent past (mean 1986–2005) and future (mean 2080–2099) for RCP2.6 and RCP6.0

Fig. 14.12

Sum of simulations with biome change between historical (mean 1986–2005) and future (mean 2080–2099) under RCP2.6 (left) and RCP6.0 (right) according to a multi-model ensemble

Fig. 14.13

Woody LAI change between recent past (mean 1986–2005) and future (mean 2080–2099) under RCP2.6 (left) and RCP6.0 (right), according to a multi-model ensemble mean (3 DGVMs and 2–4 GCMs)

Regarding biome stability, simulations with LPJ-GUESS spanning the last 140.000 years suggest a remarkable biome constancy along the southern and eastern coast and in the southern Kalahari, which appears to be correlated with high species diversity (Huntley et al. 2016, 2021). Simulations for the future with climate scenario data input, however, suggest substantial biome shifts in these areas of previous biome constancy (Huntley et al. 2021).

4 Monitoring Biome Change

Process-based forecasting models such as the phytoclime models and aDGVM are influenced by process and parameter uncertainty, which can make them difficult to interpret. A complementary alternative is to use statistical monitoring of change that is already ongoing. Global change impacts are already manifesting, meaning that statistical detection and description of the recent trajectories of change should also be a priority for adaptation and mitigation science. That is, process-based models are essential for exploring future scenarios of change, but statistical analysis of recent trajectories of change can help us understand change that is already ongoing. In this section, we describe three promising change monitoring and detection activities.

4.1 Birds as Indicators of Biome Change

Indicator species are organisms which are easily monitored and whose status reflects the state of the environment in which they are found (Siddig et al., 2016). The use of indicator species as early warning systems in climate change science is still in its infancy. Anchor (2022) has recently suggested a range of bird species that could be used as indicators of biome change. Birds have the advantage that they are readily observed by a large population of hobby birdwatchers and citizen scientists and are also mobile enough to respond immediately to changes in climate and vegetation structure. The premise is that species that are strongly associated with a particular biome can serve as sensitive indicators for changes in biome structure and functioning that are not necessarily detectable using satellites or standard vegetation monitoring. A comparable approach has been used by BirdLife International for global analyses, and however the BirdLife protocol needs to be adapted to the sub-region. Anchor (2022) identified 16 bird species with range boundaries that correspond with biome boundaries and where the range boundaries have relatively hard edges. Birds as indicators have the potential to run as a citizen science project which would reduce field costs and enhance public engagement. However, the lure of volunteer field work should not distract from the fact that a substantial investment is nonetheless required to ensure that an efficient computational back-end exists that allows data to be uploaded, processed and visualised in real time. Another issue is that detailed research will be required to understand and anticipate the different ways in which bird species may respond to changes in climate versus vegetation structure (e.g. Seymour et al. 2015) or direct human influence on habitat structure and resource availability (e.g. Weideman et al. 2020).

4.2 Phytometers as Indicators of Biome Change

A phytometer is a special kind of indicator species: a phytometer is a transplanted plant whose growth, survival or reproduction are used to provide information on environmental conditions. Phytometers have the advantage that their performance can be interpreted as an integrative measure of the environment. Here, we provide an illustration of how phytometers could be used to monitor climatically induced biome changes. The system we describe is a biome shift monitoring phytometer (BISMOP), which we have implemented in South Africa (Fig. 14.14).

Fig. 14.14

Biome shift monitoring phytometers (BISMOPs) have been installed in the major biomes of South Africa (left panel). A BISMOP is a mini-common garden experiment that monitors the performance of the plant growth forms characteristic of the biomes of South Africa. Each BISMOP monitors 3 individuals of 11 plant species and records their NDVI over a diurnal cycle as well as associated climate system variables (middle panel). The Nieuwoudtville Hantam Botanical Garden BISMOP (right panel)

A BISMOP is a miniature common garden experiment where plants representative of the biomes of South Africa are grown in common garden settings in the different biomes. BISMOP plants are selected based on two criteria. First, their distributions should closely match the distribution of the biomes. Second, they should represent a plant growth form typical of the biome in question. The existing BISMOPS used 11 vascular plant species. Individuals were planted in 9 litre pots filled with perlite substrate. Perlite was used because it has good hydraulic properties (reasonable drainage and water retention) and is commercially available in various quality classes ensuring that the same substrate can be sourced in the future. Three replicate plants of each species are grown in each BISMOP. The planted pots and one control (34 in total) are arranged in a 250 $\times$ 200 cm area. An aluminum frame supports a camera system 250 cm above the pots. The cameras are based on the Raspberry Pi Camera Module 2, which uses a Sony IMX219 8-megapixel sensor and a wide angle and low distortion lens (MX219 Camera Module 160°FoV). The camera modules have no infra-red filter and use a blue filter (Roscolux 2007 Storaro Blue, Rosco, UK). The blue filter blocks most red and green light, allowing near-infrared (NIR) to be captured in the camera’s red and green channels, whereas the camera’s blue channel records mostly blue light allowing a vegetation index with similar properties to NDVI to be calculated. Each BISMOP station further records air temperature and air humidity. The control pot, filled with perlite only, is instrumented with a soil moisture and soil temperature probe. This bare ground pot is used to calculate a simple water balance that allows the current level of moisture stress at the BISMOP site to be estimated. Images from the cameras are captured every daylight hour and environmental variables are recorded every 10 minutes.

The data from the BISMOP systems produce time series (at daily or hourly resolution) that allow the analysis of how the physiological performance of individual plants are influenced by climate system variables. Since the data represent key South African plant growth forms, growing in environmental situations that represent the diversity of climate systems in the region, the data will additionally be valuable for testing the predictions of DGVM and phytoclime models.

4.3 Remote Sensing of Phenome Change

There is a long history of monitoring vegetation activity using satellite-based remote sensing. Buitenwerf et al. (2015), for example, developed a method for analysing change in a satellite-based vegetation activity using NDVI data (NDVI, normalised difference vegetation index) from the AVHRR satellite program. The primary advantage of the (Buitenwerf et al. 2015) method is that it stratifies the phenological metrics into groups, or phenomes, thereby allowing change in the phenological metrics to be analysed by group. This allows the change analysis to be stratified by phenological groups, which in turn aids change detection since phenologically divergent signals are not diluted by averaging.

Higgins et al. (2023) applied this method to the Southern African region; this analysis used MODIS EVI (enhanced vegetation index), which has a higher resolution (the 1 $\times$ 1 km${ }^2$ MOD13 product spanning 2000–2019 was used) than the AVHRR NDVI used by Buitenwerf et al. (2015). EVI is additionally less prone to saturation at high vegetation density and less sensitive to atmospheric sources of error than NDVI. Higgins et al. (2023) further focused exclusively on protected areas, which removes the potential for confounding the effects of land use and climate on vegetation activity. The world database on protected areas was used (UNEP-WCMC and IUCN 2021) to select areas that do not contain land use effects (nature reserves, game parks, national parks and forest reserves, Fig. 14.15).

Fig. 14.15

To exclude land use effects, phenological change was analysed in the protected areas of Southern Africa. Protected areas were derived from the world database on protected areas (UNEP-WCMC and IUCN 2021)

Following Buitenwerf et al. (2015), Higgins et al. (2023) estimated 21 phenological metrics for each phenological year for the time series in each 1 km${ }^2$ pixel (Fig. 14.16). The metrics describe the annual phenological curve and include metrics with time units (e.g. day of peak EVI) and metrics in EVI units (e.g. maximum EVI). To quantify change, the mean of each metric was calculated for the first and second half of the record. Figure 14.17 summarises by how many standard deviations these metrics have changed revealing widespread change hotspots in Namibia, Botswana and Zimbabwe (Fig. 14.17). These analyses revealed shifts in both the timing magnitude of vegetation activity.

Fig. 14.16

A schematic representation of the EVI activity of a remotely sensed grid cell. At each labelled point, metrics in both EVI units and day of year units are recorded. The integral of the EVI between two subsequent trough days, although not labelled in this diagram, is also calculated. gsl: growing season length

Fig. 14.17

Changes in vegetation phenological activity between 2000 and 2019 reported in standard deviations. (a) Aggregated change of all metrics. (b) Aggregated changes in metrics representing vegetation activity (measured in EVI units). (c) Aggregated changes in metrics that represent the timing of the vegetation activity (measured in days or day of the year)

The ecological interpretation of the pattern of change is aided by classifying pixels into groups (Buitenwerf et al. 2015 phenomes) with similar phenological behaviour. Higgins et al. (2023) used the phenological metrics (Fig. 14.16) to classify pixels into 7 phenomes (Fig. 14.18). The phenome classification identified regions superficially consistent with desert, arid savanna, bushveld, grassland, woodland savanna and fynbos.

Fig. 14.18

Phenomes of the protected areas of Southern Africa. The phenomes were produced by performing an unsupervised clustering of the phenological metrics as defined in Fig. 14.16. The 7 phenomes represent zones with similar phenological behaviour

For each phenome (Fig. 14.18), the mean phenological signature in the first (2000–2009) and second (2011–2019) parts of the time series is shown in Fig. 14.19. The vectors in Fig. 14.19 show the average change in phenological metrics in both the EVI (vegetation activity) and time (timing of activity) dimensions between the first and second parts of the time series. Although the phenological signature plots suggest minor absolute changes in each phenome’s phenological signature, the vector plots reveal large relative changes.

Fig. 14.19

Phenological change per phenome. The phenological signature of each phenome is plotted for the first (2000–2009, blue lines) and second (2011–2019, green lines) parts of the time series. The vector plots show change in all metrics (cf. Fig. 14.16) in both EVI and time dimensions. For example, the circle plotted phenome 5 shows change in the metric “Trough.” In this example, the trough day has increased (it is delayed) and its EVI value has increased. Vectors parallel to the axes represent metrics that have only EVI or time dimensions

Phenome 1 covered a diverse range of Cowling’s phytogeographical regions (Cowling et al. 2004), from the Zambezian region, the Kalahari-Highveld Transition Zone, the Tongaland-Pondoland Region and Afromontane Region to the Cape region. It spread from the Moremi National Park in northwest Botswana through national parks in South Africa such as the Songimvelo in the northeast to the Maloti-Drakensberg in the southeast. The growing season length decreased in phenome 1, caused by both a delayed start and an earlier end of the growing season.

Phenome 2 was primarily distributed in the xeric Karoo-Namib and the Cape regions, spreading southwards from the Skeleton Coast in northwestern Namibia towards the Anysberg in South Africa. This phenome was characterised by a low vegetation activity and the amplitude and integral of the annual NDVI decreased over the study period. A decrease in the length of the growing season was also detected, caused by both a delayed start and an earlier end of the growing season.

Phenome 3 was mainly distributed in the Zambezian phytogeographical region, spreading from the Bwabwata in northeast Namibia through the Chobe in northern Botswana to the Hwange in northwest Zimbabwe. Further in Zimbabwe, it was distributed in the Charara in the north as well as the Gonarezhou in the south. This phenome revealed an increase in the amplitude and integral of the annual EVI signal. The length of the GSL did not change substantially although the timing metrics were delayed, with the exception of the onset of the peak and peak of the growing season which were earlier.

Phenome 4 was mainly distributed in the Zambezian and the Kalahari-Highveld Transition Zone phytogeographical regions. This phenome stretched from the Moremi in northern Botswana through the Chewore in northern Zimbabwe to the northern parts of Kruger in northeastern South Africa. The growing season length extended over the observation period, caused by both an earlier onset and a delayed end. As a consequence, the integral of the EVI increased and as did the peak EVI which was also delayed.

Phenome 5 occurred mainly in the Zambezian and the Kalahari-Highveld Transition Zone phytogeographical regions, spreading from eastern Etosha in northern Namibia towards the Kalahari in central Botswana to the eastern parts of Kruger in northeast South Africa and the Tsehlanyane in eastern Lesotho. In this phenome, the growing season length decreased mostly due to delayed start of the growing season. In addition, the peak, integral and amplitude of the EVI also decreased, combining to produce an overall decrease in vegetation activity.

Phenome 6 was primarily distributed in the Zambezian and the Kalahari-Highveld Transition Zone phytogeographical regions, spreading from the Khaudum in northeast Namibia through the Kalahari in central Botswana to the eastern parts of Kruger in South Africa and the western parts of Tsehlanyane in Lesotho. The growing season length in this phenome decreased primarily due to an earlier end of the growing season. Moreover, the peak, integral and amplitude of EVI decreased.

Phenome 7 was primarily distributed in the Kalahari-Highveld Transition Zone phytogeographical region, spreading from western Etosha in northern Namibia through the Kgalagadi in southwestern Botswana to Camdeboo in southern South Africa. In this phenome, the growing season length increased due to both an earlier start and a delayed end to the growing season, although the delayed end effect dominated. The peak, integral and amplitude of EVI all increased in this phenome.

5 Discussion

This review summarises recent analyses of vegetation change in Southern Africa. The focus is on predictive simulations of vegetation structure and function, as derived from both established mechanistically based dynamic global vegetation models (DGVMs) and novel species-based modelling of the climatic preferences of major plant growth forms (phytoclimatic modelling). The review additionally suggests options for monitoring vegetation and biome change that can be used to detect and understand change as well as to test the ability of the prediction models to provide information robust enough to support policy and management actions. Over the past two decades, the capacity to develop such models has evolved rapidly from the rudimentary yet insightful correlative modelling work of Rutherford et al. (1999) at both biome and species levels, to more sophisticated correlative modelling (Thuiller et al. 2006) and earlier dynamic vegetation modelling efforts (Scheiter and Higgins 2009; Moncrieff et al. 2014). The results of these early efforts raised awareness and concerns regarding the potential severity of climate change impacts, despite the level of uncertainty that was clearly communicated. However, reported observations of historically observed changes tempered these projections even further, showing trends that ran counter to projections in some cases (Masubelele, Stevens), or revealing changes consistent with projections (Stevens et al.) that may have occurred via unanticipated mechanisms. Two main causes of these divergences have been noted, namely the importance of disturbance via fire in over-riding climatic drivers of change, and the role of rising atmospheric CO${ }_2$ via increased water use efficiency and productivity that were not accounted for in correlative approaches. The mechanistically based approaches described here provide new insights that permit a deeper understanding that may be of more value for policy and management guidance.

We first discuss what light the results of the modelling approaches used here sheds on our understanding of the current representation of biomes and vegetation types in the region. The (Rutherford and Westfall 1986) treatment of biomes of Southern Africa, updated by Mucina and Rutherford (2006), remains one of the clearest in terms of its simple plant growth form definitions (that is, simple combinations of Raunkiaer plant life forms trees, shrubs, grasses and annual plants). Their mapping of combinations of dominant growth forms at the 100 km grid scale provides a useful template for comparison, and it remains the basis of fundamental regional ecological distinction for a wide range of applications, including the planning of adaptation responses. Despite being based purely on growth form dominance, their approach uncovered floristic distinctiveness between biomes (e.g. Gibbs Russel et al.), which makes a comparison with the phytoclime approach extremely interesting.

It is most insightful to compare the phytoclimes identified by successive increases in the number of species clusters with the biomes as mapped by Mucina and Rutherford (2006). It must be noted that the phytoclime maps represent the potential of the climate and substrate to support different growth forms (Fig. 14.1), whereas ecological processes not considered by the phytoclimatic analysis, such as fire, herbivory, competition, fertility and dispersal, influence which combination of growth forms will actually grow at a given location.

At its most parsimonious 6-cluster level (Fig. 14.3), the arid and semi-arid phytoclime unit 1 closely matches the Desert, Nama-Karoo and Succulent Karoo biome borders with Fynbos Biome in the south and Savanna, Grassland and Albany thicket biomes in the east. A hyper-arid subdivision of desert indicates a level of sensitivity lacking in the Mucina and Rutherford (2006) definition, while three phytoclimes are discerned in what is traditionally seen as Savanna across Namibia, Botswana, Zimbabwe and Mocambique. Thus, a predominance of tropical assemblages overwhelms the purely plant life form approach of Mucina and Rutherford (2006). Even at the 9-cluster level (an equivalent number of biomes recognised by Mucina and Rutherford 2006), hyper-arid and subtropical assemblages dominate, with latitudinal divisions emerging to differentiate the 6-phytoclime classification’s units 1, 2 and 6, more clearly separating grassland and savanna-type biomes in central and northern South Africa and differentiating further North/South divisions in arid and semi-arid shrubland and savanna types in the west. Phytoclime units comparable with Succulent karoo, Grassland, Thicket, Coastal Belt and Fynbos biomes emerge only when clustering 18 phytoclimes levels and higher, while the forest biome does not emerge at all.

Overall, unit 6 in the 6-phytoclime classification retained coherence at successive clustering levels while also showing low concordance with purely plant life form-defined biomes for this region. This is a novel revelation of interesting climatic control of vegetation, mediated by plant form and function, in this region. This could be further refined by expanding the phytoclime analysis to consider how soil fertility modulates the influence of climate. However, low confidence in currently available soil fertility products precludes consideration of soil fertility.

The representation of Mucina and Rutherford (2006) biomes by DGVM simulations is also interesting in how the arid- and semi-arid Succulent Karoo and Desert biomes are somewhat recognisable, but C4-dominated Grassland, C4-co-dominated Nama-Karoo and Savanna biomes occupy the central portion of the region, with increasing dominance of trees from the arid west to the more mesic eastern reaches, resulting in a less refined differentiation relative to both (Mucina and Rutherford 2006) and the phytoclime approach. The DGVMs only consider a limited set of plant functional types, and it is likely that consideration of additional functional types such as succulent plants and different types of shrubs (Gaillard et al. 2018) may enhance the ability of the DGVMs to represent the Mucina biome map more closely.

The phytoclime and DGVM analyses do make predictions of where we should expect vegetation change in the future, but these are quite different. The aDGVM, for example, forecasts that the core of the large C4 savanna biome in the central interior does not change with climate change (Fig. 14.10), whereas this is an area of high change in the phytoclime analysis (e.g. Fig. 14.4). This is confirmed when examining how the suitability of geographic space for aDGVM plant types changes across Southern Africa with changing climate. The aDGVM predicts, for example, that central Botswana will become increasingly suitable for C3 grasses and to a lesser extent for C4 grasses and that C4 grasses will decrease in central South Africa (Fig. 14.9). The phytoclimatic analysis by contrast suggests that C3 grass suitability will decrease almost everywhere except Lesotho and that C4 grass suitability will increase over the Namibian coast, the South African coast and the higher altitude areas of the region but will decrease over most of the central and northern parts of the region (Fig. 14.2). In effect, the phytoclimatic analysis is emphasising temperature effects which will favour C4 grasses, whereas the aDGVM is emphasising CO${ }_2$ effects which will favour C3 grasses.

The strong positive effects of increasing atmospheric CO${ }_2$ on the photosynthesis and the growth of C3 plants are a general feature of the DGVMs. However, DGVMs that do not simulate nutrient limitation most likely overestimate these effects (Hickler et al. 2015; Martens et al. 2021). Nonetheless, the woody encroachment observed in the eastern part of the study region was simulated by the LPJ-GUESS model (Wilhelm 2021), which does include a nitrogen cycle and nitrogen limitation of plant growth (Smith et al. 2014).

A further reason for these differences might be that the aDGVM model predicts the realised niche of the functional types, that is, the biomass that can accumulate under a given climate under the influence of fire and competition, whereas the phytoclimatic suitability surfaces represent estimates of the climate’s potential to support different plant growth forms. In effect, the phytoclimatic suitabilities (Fig. 14.1) represent the climatic limits hard coded in earlier developed DGVM model versions. These global DGVMs include hard-coded limits that relate to cold temperature tolerance and are thus unlikely to be important in this study region. The aDGVM by contrast does not include hard-coded climatic limits for its different plant types. Thus, in aDGVM, biome borders in the study region are mainly determined by the competition between grasses and trees as mediated by the simulated frequency and intensity of fire.

One of the apparently important similarities between projections made by the models is that tree dominance will increase in parts of the savanna regions (Figs. 14.4 and 14.9). For example, the region currently occupied the phytoclime suitable for grasses (phytoclime 6, Fig. 14.4) will be increasingly occupied by a phytoclime suitable for trees (phytoclime 5, see also Fig. 14.5). It is now well established that these regions are centres of widespread thickening of woody plants and that this process may drive substantive changes in the functioning and diversity of these landscapes over vast areas (e.g. Stevens et al review).

The phenome analysis (Higgins et al. 2023) provides convincing evidence that the vegetation of all parts of the sub-region is responding to climate change. This analysis is unique in that it focuses on change in protected areas and thereby excludes the potential confounding effects of land use change. The phenome analysis reveals that plants in some regions are experiencing an increase in growing season length (phenomes 1, phenome 4, phenome 7), whereas decreases were observed in other regions (phenome 2, phenome 6). In addition, in some regions there was an overall increase in vegetation activity (phenome 3, phenome 4, phenome 7), whereas for others there was an overall decrease (phenome 5) in vegetation activity. These findings are in agreement with previous work. For example, in northeast South Africa, Masia et al. (2018) reported early green-up dates and early leaf drop dates among species typical of phenomes 1 and 6. Furthermore, in the Cape Floristic region of South Africa, herbarium specimen of species has been used to show that between 1901 and 2009, leaf flowering has advanced by 12 days (Williams et al. 2021). Phenomes 1 and 2 represent this region, and they both show an early start to the growing season. Higgins et al. (2023)’s findings further agree with the early greening patterns reported by Whitecross et al. (2017) who used MODIS NDVI data to detect early green-up dates along a latitudinal gradient from Zambia to South Africa. Phenomes 1 and 4 cover portions of their study region, and they both show an early start to the growing season. Overall the findings of the phenome study emphasise that different parts of the study region are responding in qualitatively different ways to changes in the climate system. Further work is needed to ground verify the changes detected by the Earth observation satellites used in the phenome analysis. The analysis reviewed in this study provides the data for guiding such field work. In parallel, additional work is needed to attribute these changes to particular climate system drivers.

This study illustrates that modelling the biomes of Southern Africa with DGVMs reveals priorities for model development. For example, simulating field experiments with a DGVM identified that improving the representation of demographic events associated with drought is a priority (Baudena et al. 2015). In a more general sense, DGVMs can be improved by calibrating and or testing them using the Earth observation system (EOS) data; the potential of this avenue is under-exploited, particularly in Southern Africa. Using multiple lines of EOS evidence (e.g. biome maps, NPP and GPP, fire activity, surface temperature) would improve the identifiability of DGVM parameterisations and the robustness of their predictions. Furthermore, more regionalised approaches might be necessary in order to capture, for example, the peculiarities of the Karoo which are poorly captured by the PFT’s in existing DGVMs. Strategically, a balance between parameterisation and calibration needs to be found: short- and medium-term predictions can be enhanced by calibration, whereas understanding is best enhanced by adjusting parameterisations and assumptions.

Our premise when initiating the work reported on here was that using different methods to analyse biome-level change in vegetation in the Southern Africa would allow us to make robust inferences about the nature of change. However, we have learned that different inference methods represented by statistical inference of the capacity of the regions climate to support different plant growth forms (the phytoclimatic analysis), forward simulation of ecosystem dynamics (DGVM simulations) and monitoring of change using remote sensing time series of vegetation activity (the phenome analysis) have yielded different insights. The only unifying insight is that change is widespread. This unfortunately leaves us in the position where we have to advocate increased effort to refine the forecasting models. Satellite-based monitoring of vegetation change using the phenome method as described here and phytometer monitoring as represented by the BISMOP system are promising avenues for generating the data needed to test the models’ predictions and the assumption upon which they are based. However, it is also clear that experimental work aimed at exploring the interactions between CO${ }_2$ fertilisation, temperature increases and moisture deficits is necessary to resolve the disparate predictions of the phytoclimatic and dynamic vegetation models. This means that atmospheric CO${ }_2$ experiments designed to quantify the interactive effects of CO${ }_2$ enrichment, temperature and moisture should be a regional research infrastructure priority.