The influence of trees, shrubs, and grasses on microclimate, soil carbon, nitrogen, and CO2 efflux: Potential implications of shrub encroachment for Kalahari rangelands

Shrub encroachment is a well‐documented phenomenon affecting many of the world's drylands. The alteration of vegetation structure and species composition can lead to changes in local microclimate and soil properties which in turn affect carbon cycling. The objectives of this paper were to quantify differences in air temperatures, soil carbon, nitrogen, and CO2 efflux under trees (Vachellia erioloba), shrubs (Grewia flava), and annual and perennial grasses (Schmidtia kalahariensis and Eragrostis lehmanniana) collected over three seasons at a site in Kgalagadi District, Botswana, in order to determine the vegetation‐soil feedback mechanism affecting the carbon cycle. Air temperatures were logged continuously, and soil CO2 efflux was determined throughout the day and evening using closed respiration chambers and an infrared gas analyser. There were significant differences in soil carbon, total nitrogen, CO2 efflux, light, and temperatures beneath the canopies of trees, shrubs, and grasses. Daytime air temperatures beneath shrubs and trees were cooler compared with grass sites, particularly in summer months. Night‐time air temperatures under shrubs and trees were, however, warmer than at the grass sites. There was also significantly more soil carbon, nitrogen, and CO2 efflux under shrubs and trees compared with grasses. Although the differences observed in soils and microclimate may reinforce the competitive dominance of shrubs and present challenges to strategies designed to manage encroachment, they should not be viewed as entirely negative. Our findings highlight some of the dichotomies and challenges to be addressed before interventions aiming to bring about more sustainable land management can be implemented.

Over longer time scales, unsustainable grazing can affect competition between plant species, initially reducing perennial grass cover in favour of less palatable annual grasses and ultimately replacing grasses with shrubs (Eldridge et al., 2011;Van Auken, 2000). Shrub encroachment has long been a feature of the Kalahari, with reports that >1 million ha of Kalahari rangeland was affected by Senegalia mellifera expansion as long ago as the 1950s (Ebersohn, Roberts, & Vorster, 1960, cited in Hagos & Smit, 2005. In southern Africa, numerous studies have attributed shrub encroachment to long-term grazing (e.g., Joubert, Smit, & Hoffman, 2013;O'Connor, Puttick, & Hoffman, 2014). The universal occurrence of shrub encroachment across the world's drylands is, however, suggestive of global drivers of change (Stevens, Lehmann, Murphy, & Durigan, 2017;Van Auken, 2000). The increase in atmospheric CO 2 favours C 3 photosynthetic shrubs because it leads to improved water and N use efficiencies over lower CO 2 adapted C 4 grasses (Leakey et al., 2009). Warming could also enhance the survival and growth rates of cold intolerant woody shrubs (D'Odorico et al., 2013). In southwestern Ethiopia, pollen records suggest that the savannah has been affected by multiple phases of shrub encroachment over the last two millennia, the primary drivers of which were changes in rainfall and fire occurrence (Gil-Romera, Lamb, Turton, Sevilla-Callejo, & Umer, 2010). Nevertheless, although global and regional factors may be a driver of vegetation change in drylands, they cannot alone explain local differences (Bond & Midgley, 2012;D'Odorico, Okin, & Bestelmeyer, 2012). Ultimately, the vulnerability of an area to shrub encroachment will depend on land use and the functional traits of plants that govern their responsiveness to all drivers, human, and climatic (Bond, 2008;Stevens et al., 2017;Van Auken, 2000).
The decline in grazing value associated with shrub encroachment is the primary reason why it is considered both a cause and symptom of land degradation in drylands. Changing vegetation composition will, however, have a range of impacts, including the development of islands of fertility resulting from the unequal distribution of resources (Schlesinger et al., 1990). Localised nutrient enrichment occurs as plants (primarily shrubs and trees) obstruct sediment and water movement (Tongway & Ludwig, 1994) and through dung enrichment from animals attracted by the shade (Dean, Milton, & Jeltsch, 1999). Shading also reduces direct losses of C from surface litter via photo-degradation (Austin & Vivanco, 2006), altering soil microbial, temperature, and moisture conditions, all of which will affect organic matter mineralisation rates and soil C stores. This partly explains why Eldridge and Soliveres (2014) found a positive correlation between shrub density and a range of ecosystem characteristics such as biodiversity, C sequestration, soil fertility, and rainfall capture. The positive impact of shrubs on a range of ecosystem characteristics is, however, density dependent. Where canopy cover exceeds 40-60%, they found that impacts across a range of parameters became negative (Eldridge & Soliveres, 2014) as the ability of individual shrubs to capture runoff and sediment declines (Breshears, 2006).
Despite widespread vegetation changes, the species specific impact of trees, shrubs, and grasses on microclimate and C cycling is poorly documented in the sandy soils of the Kalahari. Quantifying these changes for different plant species and understanding their implications are vital if we are to improve our understanding of terrestrial C stores and cycles and a prerequisite to inform sustainable land management in drylands. This paper presents soil and climatic data collected during three contrasting seasons in the Kgalagadi District, Botswana, relevant to the C cycle and soil-vegetation feedback processes in shrub encroached rangelands. The objectives were to determine how tree, shrub, and grass cover affect (a) microclimate (light and air temperature) and (b) soil C, N, CO 2 efflux, and the δ 13 C of respired gases. The results are discussed in the context of terrestrial C cycles, land degradation, and management options that could be employed in order to facilitate more sustainable land management in the Kalahari rangelands.  (Wang, Okin, Caylor, & Macko, 2009). Mean annual precipitation is 334 mm. At the time of sampling, a 3-4-mm deep biocrust, composed of bacteria, cyanobacteria, and fungi (Elliott et al., 2014), covered approximately 30% of the soil surface.

| Study site and data collection
Data were collected within a fenced area enclosing several hectares typical of the rangelands in the region, although for the duration of the study, livestock was excluded to allow installation of monitoring equipment. Three replicate microsites characterised by trees (Vachellia erioloba E. Mayer), shrubs (Grewia flava DC), and 2.3 | Soil CO 2 efflux and δ 13 C Soil CO 2 efflux was determined using closed respiration chambers and an infrared gas analyser (PP Systems, Amesbury, USA) which facilitated multiple replication and near simultaneous determination of efflux.
Details of the chambers, field methods, and quality control procedures are supplied in the Supporting Information. CO 2 efflux measurements were undertaken in the early morning, at noon, in the mid-afternoon, and evening on each day to encompass a range of temperature, humidity, and light conditions representative of diurnal cycles. For the shrub and tree microsites, measurements were taken on both east and west sides of the canopies and the mean used as a single value for the analyses. In total, there were 144 CO 2 flux measurements from soil under each grass type and 288 measurements from soils beneath both the shrub and tree canopies. The short-duration but high intensity measurement protocol repeated over contrasting seasons allows for more reliable quantification of efflux than periodic daily or weekly measurement for identifying differences between soils under different vegetation types and under different weather conditions. This protocol was adopted on the basis of experience in soil efflux measurement in this environment (e.g., Thomas, 2012). Biological respiratory systems discriminate to a greater or lesser extent between the two stable C isotopes, 12 C and 13 C. This fractionation can be used to identify biological sources and pathways in soils and their interaction with the atmosphere (Amundson, Stern, Raisden, & Wang, 1998). On March 8, 2012, the 6 p.m. cycle was used to collect gas samples for determination of their 12 C and 13 C isotopic compositions. Three gas samples were collected from the chambers and injected into pre-evacuated 12 mL borosilicate Exetainer® vials. To prevent contamination, the vials were over-pressurized, and lids were dipped in hot wax before transportation to a stable isotope facility (CEH, Lancaster, UK). Samples of leaf litter from beneath G. flava and E. lehmanniana were collected, and their 12 C and 13 C isotopic compositions determined.
Stable 13 C/ 14 C isotopic ratios for each sample, R sample , = δ 13 C are reported in parts per thousand (per mil, ‰) variation from the ratio in the Pee Dee Belemnite (PDB) international standard (Equation 1).

| Soil total C, N, and bulk density
Soil samples were collected from five depths (0-1, which included biocrust where present, 1-2, 2-5, 5-10, and 10-20 cm). Three replicate pits were sampled from soils under each of the microsites (tree, shrub, and annual and perennial grass) during each field season. Soils were sieved to remove particles >2 mm, air-dried, bagged, and stored prior to analysis. Total C and N were determined using a Leco TruSpec CN element analyser (Matejovic, 1997). Soil bulk density was determined on samples collected from 5-cm depth using a metal cylinder with internal dimensions of 10 cm × 4.8 cm. Sample mass was determined after sieving and oven drying at 105°C.

| Photosynthetically active radiation
Incoming solar radiation in the photosynthetically active wavelengths (400-700 nm) was measured continuously and simultaneously 2 m above the ground surface in an open area and beneath a G. flava shrub canopy using photosynthetically active radiation quantum sensors (Skye Instruments Ltd., UK) connected to data loggers.

| Statistical analyses
All statistical analyses were performed using SPSS (IBM v. 24 Cohen's d values of <0.2 were attributed to small effect sizes; up to 0.5 medium effect sizes and >0.8 were considered large effect sizes. To test the effects of temperature and moisture on CO 2 efflux, linear and multiple linear regression tests were performed. The temperature sensitivity of CO 2 efflux at each site within each season was described with a Q 10 exponential model. We used a fitting algorithm to maximise the correlation coefficient r 2 (Q 10 ) with the initial conditions R 0 = R (0) and T 0 = 0, where is R 0 is efflux at reference temperature T 0 (Thomas & Hoon, 2010).

| Air temperature and solar radiation
There were significant differences in mean daytime and night-time air temperatures between vegetation microsites in all months (Figure 3a (Table 1c). The effect size of these differences was large in all cases. Differences in soil C and N concentrations beneath G. flava and beneath grasses were not significant (Table 1c, Figure 3d).
Total C and N concentrations declined with depth at all sites ( Figure 5) but remained higher at depth beneath shrubs and trees compared with grasses. Incorporation of leaf litter into the surface soils reduced bulk densities from 1.51 ± 0.05 g cm −3 under shrubs to 1.28 ± 0.05 g cm −3 in soils under grass, but soils still contained significantly greater C stores than the surrounding grass covered areas. C: N ratios at the surface of soils under grasses were 8.4-8.6 declining with soil depth to 6.4-6.9. C:N ratios in surface soils under shrubs and trees were higher (10.6 and 12.1, respectively) declining to 9.5 and 6.2, respectively, at 10-20 cm.

| Soil CO 2 efflux
There were significant differences in mean soil CO 2 efflux between vegetation microsites in all months (Figure 3c) Table 1d).
The temperature sensitivity of soil CO 2 efflux is strongly affected by soil moisture conditions (Figures 2 and 6). In November 2011, when soils were dry, CO 2 efflux from soil under all vegetation types was uniformly low and unresponsive to soil temperature, with Q 10 ranging from 1.20 to 1.25. However, with elevated soil moisture in both cool winter and warm autumn conditions, soil CO 2 efflux increased with temperature at all sites, and the Q 10 was between 1.35 and 1.40.
The δ 13 C of efflux gases in March 2012 shows different signatures between soil CO 2 efflux originating from soil beneath C 4 grasses and C 3 shrubs and trees (Figure 7). There was, however, no significant dif-  This is important for several reasons. First, changes in dryland net primary productivity, soil C storage, and fluxes will have global impacts (Ahlström et al., 2015;Le Quéré et al., 2013). Second, vegetation change will alter the fire regime, and an increase in shrubs at the expense of grasses will reduce fuel loads and likely reduce the frequency and severity of fires (Bond, 2008;Mouillot & Field, 2005    Shrubs and trees typically have a lower albedo than grasses, particularly in the dry season when grass foliage is dead (Hayden, 1998), and this is likely to have an impact on regional heat balances. (1971) hypothesis of grass and tree root niche separation has been used to explain coexistence and competition in mixed tree, shrub, and grass savannahs, although this has been the source of some contention in southern Africa. For example, Scholes and Archer (1997) and Hipondoka, Aranibar, Chirara, Lihavha, and Macko (2003) considered it unlikely because of interspecies root interaction at all soil depths, likely mediated by arbuscular mycorrhizal networks. More recently, however, Ward, Wiegand, and Getzin (2013) provide a convincing case for support in dry savannahs such as the Kalahari. Warmer air temperatures will generate higher evaporation pressures in the soil profile, and percolating rain water will have to penetrate deeper before it is below evaporation depths. Currently, the most frequently occurring rainfall event at the study site is <5 mm, and this infiltrates to relatively shallow depths of approximately 0.1 m. Deep soil moisture recharge was only observed after successive, large (> 20 mm) rainfall events, and this is a much less frequent occurrence (Figure 2). A shift to more intense rainfall events would lead to greater soil moisture recharge accompanied by increased microbial activity, and potentially, the ability of plant roots currently only extending into dry subsoil zones to access new sources of moisture. Conversely, more rain days, with fewer large events, could favour grasses (and biocrusts), delivering water to only shallow depths.

| Vegetation types and C stores and fluxes
Shrubs and trees are hotspots of biological activity, C and N cycling (see, e.g., Tews et al., 2004). Stores and concentrations of soil C and N (Figures 3d and 5), microbial activity, and CO 2 efflux (Figure 3c) were significantly higher under shrubs and trees compared with grasses. areas. Their work concluded that over regional scales, the availability of moisture controls nutrient cycling rates, whereas at a local scale, vegetation patchiness is the key control.
Findings from a parallel study at the same microsites demonstrated that vegetation cover was also related to soil microbial community structure (Elliott et al., 2014). Of particular significance were cyanobacteria, capable of sequestering CO 2 and fixing N 2 (Flores, López-Lozano, & Herrero, 2015). Cyanobacteria were abundant in soil surfaces at grass sites but were present in only very low levels in soils beneath trees and shrubs. This is likely explained by competition from plants, including significantly reduced light levels and changes to the temperature and moisture regime at the soil surface (Figures 2 and   3). Litter inputs from plants and animals will provide resources for heterotrophic microbial competitors and further constrain cyanobacterial populations. Recovery of biocrusts may be affected by a loss of microbial inoculum, particularly cyanobacteria, associated with an increase in shrubs and trees. The protection of small refuge areas could, however, provide natural inoculum to surrounding degraded areas, facilitating crust regeneration when grazing pressure eases.
Higher CO 2 efflux associated with soils under trees and shrubs was most likely due to a combination of factors including greater heterotrophic microbial and plant root respiration, higher concentrations of C, and more favourable conditions for microbial respiration (Tang & Baldocchi, 2005). Soil microbial populations, respired CO 2 , and gas diffusion will all be affected by changes to soil properties associated with vegetation change. The soil surface under shrubs and trees has a lower bulk density than beneath grasses, and this will facilitate water and gas movement through the soil profile. Soil CO 2 efflux of biotic origin is moisture limited for most of the year in the Kalahari, and in dry soils, the sensitivity of CO 2 emissions to changes in temperature is almost zero (Figure 6). In dry soils, heterotrophic respiration will be mainly due to fungal and, to a lesser extent, fungal associated bacteria, due to their ability to translocate water directly or via mycorrhizal associations from host plant sources. The approximately 10% increase in Q 10 during moist soil conditions is likely due to enhanced bacterial activity. The soil moisture regime is thus vital to soil microbial activity, decomposition processes, respiration, and soil CO 2 efflux (see also Wang, D'Odorico, et al., 2009). There are complex temporal variations in soil moisture profiles, associated with rainfall, infiltration, and evaporative pressure (Figure 2), all of which will be affected by vegetation.
Across the continent, land use changes, particularly conversion of natural land to agriculture and agricultural intensification, have been identified as accelerating CO 2 efflux from African soils (Kim et al., 2016). Our results demonstrate that shrub encroachment is another driver that will lead to significantly greater CO 2 efflux from dryland sand soils.
Whether or not the increased CO 2 efflux from soil beneath trees and shrubs represents a long-term decline in the soil, C store depends on the provenance of the C. The majority is likely to be associated with autotrophic (root) respiration and does not necessarily mean a net loss of soil C to the atmosphere. CO 2 efflux will also depend on the nature of the soil organic C which will reflect the form of the litter inputs from grasses, trees, and shrubs, all of which have unique C:N and 12 C: 13 C ratios. The distinctive δ 13 C signatures of CO 2 efflux from soils under grasses, shrubs, and trees demonstrated the different C fractionation processes and C origins (Thompson, Zaady, Huancheng, Wilson, & Martens, 2006; Figure 7). The high standard deviation associated with the mean δ 13 C of gas from soil underneath G. flava suggests that there are two distinct contributory sources with unique isotopic signatures.
The y intercept of the grass Keeling plot is clear and suggests a typical C source that has been fractionated by a C 4 photosynthetic pathway. Vegetation changes will alter the type of organic compounds stored in the soil and the ease with which they are respired and ultimately their residence times. Further work is needed to determine the implications of these differences for the soil C store.  (Eldridge & Soliveres, 2014) because it also forms new habitats for a variety of species (Smit & Swart, 1994), enriches the C and N of soils, provides a more favourable microclimate, and encourages a greater soil microbial diversity across the landscape. The challenge for sustainable grazing in the Kalahari is to adopt management strategies that avoid driving significant longer term shifts in vegetation structure but that also accommodate a degree of vegetation change. Nevertheless, an ever-increasing number of livestock on a decreasing amount of communal grazing land resulting in widespread, dense thickets of shrubs is not advisable, particularly when it could compromise other rangeland uses (e.g., collection of medicinal herbs or thatching grasses [Sallu, Twyman, & Stringer, 2010]).

| Challenges to sustainable pastoral management in the Kalahari
Labour intensive intervention programmes involving removal of shrubs have been recommended (Reed et al., 2015). However, clearance of encroaching shrubs would result in a significant loss of soil nutrients from an already nutrient poor system and more extreme air temperatures, and selective thinning of dense thickets might be a more appropriate intervention (Hagos & Smit, 2005). Strategies that prevent land falling into the later stages of grazing-induced degradation will have numerous benefits for ecosystem services as natural recovery from shrub-encroached, or bare dune states are unlikely without significant intervention. Consequently, management practices that prevent shrub encroachment from occurring in the first place, such as through destocking in times of drought and manual bush removal or stem burning, will be more cost-effective over the long term.

| CONCLUSION
This study clearly demonstrates that there were significant differences in soil C, N, CO 2 efflux, and microclimate beneath the canopies of trees, shrubs, and grasses at a rangeland site in the Kalahari. The soil surface beneath shrubs and trees was cooler during the summer daytime, warmer during winter nights, and experienced less intense solar radiation than grass sites. Soils beneath trees and shrubs contain greater total C and N and contribute to greater soil microbial diversity in the landscape. Consequently, microbial activity was also higher resulting in more rapid nutrient cycling. Although associated with a loss of palatable grasses, an increase in shrubs will also be associated with higher soil C stores and less extreme ground air temperatures. This has important implications for grazing land management which our findings show should not seek to remove low density shrub thickets but rather seek to exploit the wider benefits from a mosaic of dryland vegetation types. We are indebted to Jill Thomas who provided access to her land and logistical support for which we are very grateful. This paper is dedicated to her memory. The authors declare no conflicts of interest.