Response of Net Primary Productivity of Zambezi teak forests to climate change along a rainfall gradient in Zambia

School of Natural Resources, The Copperbelt University P.O. Box 21692, Kitwe, Zambia. Water Systems and Global Change Group, Wageningen University and Research, P.O. Box 47, 6700AA Wageningen, The Netherlands Department of Earth and Environmental Systems, Indiana State University, Terre Haute, Indiana, 47809 USA VU University Amsterdam, De Boelelaan 1085, 1081 HV Amsterdam, The Netherlands 10 Environmental Systems Analysis Group, Wageningen University and Research, P.O Box 47, 6700AA Wageningen, The Netherlands. IHE Delft Institute for Water Education, PO Box 3015, 2601 DA Delft, The Netherlands


Introduction
Globally, net primary productivity (NPP) plays a pivotal role in the carbon budget.Many authors have demonstrated the existence of a close relationship between shifts in NPP and dynamics in climatic variables (Delire et al., 2008;Pan et al., 2015).
Without a doubt, patterns of terrestrial NPP may respond to changes in climatic variables.Therefore, inferring the NPP patterns and assessing the sensitivity of terrestrial NPP to global change and natural disturbance is essential because it deepens our understanding of ecosystem response to climate change.Future climate trends for most of the Zambezian phyto-region point in the direction of increased aridity.Therefore, the frequency and severity of drought events are expected to intensify under changing climate for most of southern Africa.Further, during the past half a century, available evidence show that the temperature increased by 0.5 °C in Africa and minimum temperatures rose more than maximum temperatures.Specifically, in southern Africa, the prediction is for a temperature rise by more than 3.4 ºC above the 1981-2000 baseline towards the end of the 21 st century (Niang et al., 2014).Both these climatic shifts may lead to increased ecosystem water stress and consequently a reduction in NPP in the Zambezian phytochorion.
However, these climatic variabilities may vary significantly in the way in which they impact forest growth and productivity.
These variations may occur at spatial, ecosystem types, and climate zones (Wu et al., 2011).For example, while increased temperature stimulates plant growth up to its optimal temperature in some plants (Wu et al., 2011), it has also been demonstrated to exponentially stimulate autotrophic plant respiration (Burton et al., 2008;Wu et al., 2011).Therefore, the potential impacts of increasing temperature can either be enhanced or moderated and may depend on whether water availability decreases or increases (Chen et al., 2013).Without a doubt, reduced rainfall coupled with an increase in temperature is known to suppress NPP in most terrestrial ecosystems (Wu et al., 2011).
Although the Miombo woodlands chiefly dominate the Zambezian phytochorion, other important vegetation types therein play an important role in the lives of the local community as well as providing environmental services.The Zambezi teak forests are one such unique ecosystem that have come under intense human pressure in the most recent past (Matakala et al., 2015;Theilade et al., 2001).These forests are a source of valuable commercial timber produced from Baikiaea Plurijuga.The forests also provide employment through wood-based industries (Piearce, 1986a;Piearce, 1986c).Additionally, the Zambezi teak forests play a substantial role in mitigating climate change.
To examine the response of Net Primary Productivity to climate change in the Zambezian phytochorion, we studied the Zambezi teak forests in southern Africa.These forest type occur along a moderately steep rainfall gradient and as expected presents an exciting opportunity to make inference on ecosystem response to the changing climate.In this study, we applied the LPJ-GUESS model (Ahlström et al., 2012;Smith et al., 2001) to quantify the projected future effects of changes in temperature, rainfall, CO₂ concentration, solar radiation, and number of wet days on NPP under RCP 4.5 and RCP 8.5 scenarios.Our objective was to assess the NPP's future response of the Zambezi teak forests to climate change along a rainfall gradient in Zambia.983.46 (1944-2011) 905.20 (1944-2011) 643 (1947-2011) Mean annual temperature (°C) 21.35 (1959-2003) 21.61 (1959-2011) 21.46 (1950-2011) Nitrogen (%)

LPJ-GUESS model description
LPJ-GUESS (Ahlström et al., 2012;Smith et al., 2001) is a dynamic vegetation model (DVM) optimised for local, regional, and global applications.The model uses temperature, precipitation, solar radiation, number of wet days, CO₂ concentrations, and soil texture as input variables to simulate the exchange of water and carbon between soils, plants, and the atmosphere.The ecosystem composition and structure is then determined for each simulated scale.One grid cell has a number of patches of approximately 0.1 ha in size (Smith et al., 2001).Each patch has a mixture of PFTs (Ahlström et al., 2012;Sitch et al., 2003), distinguished by their bioclimatic niche (distribution in climate space), growth form (tree or herb), leaf phenology (evergreen, summer green, or rain green), photosynthetic pathway (C3 or C4), and life history type (shade-tolerant or shade-intolerant).In a patch, each woody plant belongs to one PFT and has a unique set of parameters that control establishment, phenology, carbon allocation, allometry, survival response to low light conditions, scaling of photosynthesis and respiration rates, and the limits in climate space the PFT can occupy.In the model, leaf longevity has a direct relationship with carbon storage, and in LPJ-GUESS the relationship is implemented by relating the specific leaf area (SLA; m² kg C -1 ) to leaf longevity (See Eq. ( 1)) according to the 'leaf economics spectrum' (Reich et al., 1997).
where α is leaf longevity (in years).
Photosynthesis, stomatal conductance, plant water uptake and evapotranspiration are modelled concurrently on a daily time step by a coupled photosynthesis and water module, which was adapted from the BIOME3 model (Haxeltine and Prentice, 1996).Soils have an upper (0.0 m to 0.5 m) and a lower (0.5 m to 1.5 m) layer, identical in texture.Water enters the upper soil layer through precipitation.Transpiration and evapotranspiration deplete the water content of the soil.Additional depletion of soil water may occur through percolation beyond the lower soil layer and out of reach by plant roots.Uptake by plants is partitioned according to the PFT specific fraction of roots situated in each layer (Smith et al., 2001).
Net Primary Productivity (NPP) is determined from Gross Primary Productivity (GPP) after accounting for maintenance and growth respiration.The accrued NPP is allocated on an annual basis to leaves, sapwood and fine roots, enabling tree growth (Sitch et al., 2003).This allocation is adjusted such that the following four allometric equations, or "constraints", controlling the structural development of the average individual, remain satisfied: Leaf area to sapwood cross-sectional area relationship (McDowell et al., 2002) (See Eq. ( 2)), the functional balance constraint (See Eq. ( 3), the stem mechanics equation (Huang et al., 1992) (See Eq. ( 4)), and the crowding constraint (See Eq. ( 5)) (Reineke, 1933).In LPJ-GUESS model, crown area (m 2 per individual) is determined from stem diameter (See Eq. ( 6)) and tree diameter is derived from the sapwood, heartwood, and wood density (See Eq. ( 7)).
Changes in PFT populations occur through the establishment and mortality of individuals.Bioclimatic limits (average climate of the last 30 years) determine which PFTs are able to establish under current climatic conditions, and establishment is implemented at the end of each simulation year for each PFT.Individual plants die due to stress, senescence, disturbance, and fire.Fire depends on litter load, flammability and the available water content.Available water content is determined from the uppermost soil layer as a surrogate for the litter moisture content, which is not modelled explicitly (Thonicke et al., 2001).
Biomass-destroying disturbances are simulated as a stochastic (random) process, affecting individual patches.The generic disturbances with a 100 year expected interval were prescribed.These kill all individuals on an affected patch, converting their biomass to litter (Smith et al., 2001).We used LPJ-GUESS version 3.0 and implemented a 'cohort mode' for our study (Braakhekke et al., 2017;Smith et al., 2001).Though this model version accounts for nitrogen dynamics in soil and vegetation, we did not switch nitrogen on during our simulations.

𝐿𝐴𝐼 = 𝐾 𝑙𝑎𝑠𝑎 × 𝑆𝐴
(2) Where Klasa, Klr , Krp , Kallom1, Kallom2, and Kallom3 are all constants, LAI is the leaf area index (m²), SA is the sapwood cross section area (m²), Cleaf is leaf carbon (kg C m²), Croot is root carbon (kg C m²), ω is the mean annual value of a drought-stress factor which varies between 0 and 1 and higher values represent greater water availability.In our study we used a value of 0.35, which is the water stress threshold for leaf abscission (i.e. the point at which the leaves start shading).H stands for total tree height (m), D is tree diameter (m), N stands for population density (individuals per m²), CA is crown area (m²), CAmax is maximum crown area (m²), WD stands for wood density (kg C m ¯³), Csapwood is sapwood carbon (kg C m²), and Cheartwood is heartwood carbon (kg C m²).
We collected data on crown area, tree diameter, and total tree height from the field survey in our previous studies (Ngoma et al., 2018a, b), while data on leaf longevity was determined from Specific Leaf Area (SLA) (Reich et al., 1997) to parameterize LPJ-GUESS model.We determined SLA from tree leaves we collected from the trees that we felled to develop allometric equations (Ngoma et al., 2018a, b).Data on vegetation carbon and tree ring indices for LPJ-GUESS model validation was taken from the biomass (Ngoma et al., 2018a, b) and dendrochronological (Ngoma et al., 2017) studies respectively.

Description of the modelled climate data
We used the Representative Concentration Pathways 4.5 and 8.5 (RCP 4.5 and RCP 8.5) scenarios with an ensemble of five Global Circulation Models (GCMs): CNRM-CM5, EC-EARTH, HADGEM2-ES, IPSL-CM5A-LR, and MPI-ESM-LR (See We applied the method by Piani et al. (2010) to bias-correct daily rainfall and temperature (minimum and maximum) values from the five GCMs against the WATCH Forcing Data (Weedon et al., 2011).The solar radiation data was bias-corrected following the method by Haddeland et al. ( 2012) using WATCH forcing data series  as a reference.

Climate change
In this study, we defined climate as the average weather pattern over a period of 30 years.Climate change was thus, defined as the difference between the climates of two periods.We used 1960-1989 as the baseline to determine the relative climate change for the end of the 21 st century (2070-2099).

Description of the Zambezi teak forests.
Following the defined PFTs (Ahlström et al., 2012;Sitch et al., 2003), we used the "deciduous tropical broadleaved rain green" PFT in our study.Deciduous tropical trees shed their leaves during the dry season (See Appendix A in Ngoma et al. (2017) for the Zambezi teak forests in different seasons of the year).Trees of the Zambezi teak forests tolerate shade.For example, seedlings of Baikiaea plurijuga need some shade to survive (PROTA4U, 2017).Shade tolerant species are able to dominate a closed-forest and seeds are able to germinate in a closed forest.For Baikiaea plurijuga, regeneration is mainly from seeds, though seedlings are usually destroyed by wild animals within the forests (Piearce, 1986a).
The Zambezi Teak forests are two storeyed forests with either a closed or open canopy (Mulolwa, 1986).They are composed of 80 species (Ngoma et al., 2018a, b) but Baikiaea plurijuga Harms is most abundant (i.e.50 %) (Ngoma et al., 2018a, b;Ngoma et al., 2017).Trees of the Zambezi teak forests grow up to 20 m high and 120 cm in diameter (Piearce, 1986b).The forests have a deciduous shrub layer which is locally known as mutemwa and that grows up to 3 m to 6 m high.During the rainy season the forests have a ground layer of herbs and grasses (Mulolwa, 1986).These herbs and grasses have shallow root systems that develop during the rainy season and die or become dormant during the dry season.

Model set-up
We initiated the model with a 1000 year spin-up at each site to allow the model time to reach equilibrium in all carbon pools.
We  After the spin-up period, and using observed local climate data at the respective sites as forcing, we performed a factorial experiment to determine the effects of various tree parameters (Table 2) and soil textures (Tables 1 and S1) on different model output.We first ran the model with default tree parameters that were provided together with the model code (These are tree parameters from literature, but provided together with the model code.See Table 2).After identifying some limitations (Section 3.2), we tested the effects of local tree parameter values listed in Table 2 that coincided with the locations of our measurement plots (Ngoma et al., 2018a).We assessed effects of changing each parameter separately, and of changing all parameters combined at each site (Table 2).We further assessed the effects of soil by running the model with default soil parameters (provided with the model code on a 0.5 x 0.5 global grid) and with local soil parameters derived from samples at the respective sites (Table S1).Results at each site were averaged for 45 years

3.1
Projected climate conditions: RCP 4.5 and RCP 8.5 Temperature (Fig. 2b) and incoming solar radiation (Fig. Rainfall is projected to decrease by 33 mm and 23 mm at Kabompo and Sesheke respectively, and increase by 28 mm at Namwala under RCP 8.5 by 2099.Under RCP 4.5, rainfall will increase at Namwala and Sesheke and decrease at Kabompo by the end of 21 st century (Fig. 2a).The number of wet days will decrease at all sites under both scenarios by end of the 21 st century (Fig. 2d).Carbon dioxide concentration is projected to almost double under RCP 8.5 by 2099 (Fig. 2e).

LPJ-GUESS model validation
We forced LPJ-GUESS model with observed local climate data and used local tree (Table 2) and soil parameter values (Table S1) to validate the model.We validated the model by comparing standardised tree-ring indices to LPJ-GUESS simulated annual NPP, i.e. for the period 1970-2003 at the Kabompo site, and 1959-2011 at the Namwala and Sesheke sites.The relationships were not significant at all the three sites (Fig. 3).We also validated the model by comparing measured vegetation carbon with simulated vegetation carbon at the respective study sites.We forced the model with local climate data and ran it with default soil and tree parameters to assess its performance and the model over-estimated vegetation carbon stock at all sites by between 44 % and 145 %.However, replacing default with local soil parameters (Table S1), maximum crown area, wood density, leaf longevity, and allometry (Table 2), the error reduced to 5 %, 47 %, and 17 % at the Kabompo, Namwala, and Sesheke sites compared to measured vegetation carbon (Fig. 4).We further assessed the LPJ-GUESS model performance by comparing measured and simulated tree heights and crown size.
Using Eq. ( 4), tree heights estimated using default tree parameter values (Table 2) of  2 and  3 were taller than those estimated using local tree parameters of these same constants for the measured tree diameter at breast height (DBH) at all sites (Fig. 5).Applying the Mean Absolute Percentage Error (Sileshi, 2014) to indicate allometric model performance, tree heights were over-estimated by 111 % at Kabompo, 156 % at Namwala, and 56 % at Sesheke sites when we used default tree parameters values of  2 and  3 in the allometric equation compared to measured tree heights.Using local tree parameter values (Table 2), tree heights were over-estimated by 2 % and 1 % at Kabompo and Namwala and under-estimated by 8 % at Sesheke respectively.Thus, both default and local tree parameters over-estimated tree heights at Kabompo and Namwala compared to measured heights, though the over-estimation was largest with default parameters (Fig. 5).The crown size, estimated with Eq. ( 6), was under-estimated by 61 % at Kabompo and Namwala and by 76 % at Sesheke when we used default tree parameters.However, with local tree parameters, the model under-estimated crown size by 15 %, 11 %, and 23 % at Kabompo, Namwala, and Sesheke, respectively compared to measured crown size (Fig. 6 and Table 2).5

LPJ-GUESS model performance
We generated new soil texture and tree parameter values for maximum crown area, wood density, leaf longevity, and allometry, and results simulated with the LPJ-GUESS model improved when we used these local soil and tree parameter vales compared to using the default parameters.The over-estimation of vegetation carbon that resulted from using default soil parameter values indicates the differences in clay, silt, and sand proportions between default and local soils of the Zambezi teak forests.Our field measurements (Ngoma et al., 2018a, b) showed that trees were between 2 m and 21 m tall.The high default tree heights of between 8 m and 47 m led to over-estimating vegetation carbon by between 33 % and 92 %.
We found no correlation between LPJ-GUESS-simulated NPP and tree-ring indices at all sites.This lack of correlation hint at the limited amount of soil water that is available to trees in the model, given the standard 1.5 m maximum soil depth we adopted in our simulations.In the sites, trees access soil water down to more than 5 m depth according to the trees' rooting depth in the Zambezi teak forests (Childes, 1988;Högberg, 1984;Ngoma et al., 2018a, b).However, we adopted the standard 1.5 m soil depth in our model runs following the maximum soil depth from which we conducted soil assessment (Table S1).
The limited amount of soil water availability in LPJ-GUESS model was clearly indicated by the non-significant correlations between NPP and rainfall at all sites (Fig. S2) as oppose to the significant positive relationship we recorded between tree ring (DGVM's) on trees' productivity was also reported by other researchers (Babst et al., 2013).Thus, apart from the limited water accessibility in the model, the lack of representation of carry-over effects of previous year's rainfall limits NPP's growth in LPJ-GUESS model resulting in lack of correlation between LPJ-GUESS-simulated NPP and tree-ring indices.This opens the novel concept to improve and validate LPJ-GUESS model.

NPP's climate response
We projected an increase in NPP at Kabompo and Namwala caused by increasing CO₂ concentration and temperature.The positive temperature and CO₂ effects were clearly observed from the high positive correlations between NPP and temperature (Fig. S5), and NPP and CO₂ (Fig. S4).However, the positive temperature effects could just be up to an optimal temperature level.For tropical trees, carbon uptake reduces with leaf temperature of above 31 °C (Doughty and Goulden, 2008).Activity of photosynthetic enzymes also reduces (Farquhar et al., 1980) resulting in reduced NPP.
The projected NPP increase at Kabompo and Namwala is in the same direction as the results reported by other researchers (Alo and Wang, 2008;Mohammed et al., 2018;Pan et al., 2015) for some parts of Africa (Table 3).Some modelling studies on tropical forests (Braakhekke et al., 2017;Ciais et al., 2009;Doherty et al., 2010;Melillo et al., 1993;Midgley et al., 2005;Pan et al., 2015;Thuiller et al., 2006) also reported high positive effects of increased CO₂ concentration on forests' productivity.However other researchers report that herbaceous plants and deciduous trees sometimes acclimate to increased CO₂ concentration by reducing photosynthetic capacity and stomatal conductance (Ellsworth, 1999;Mooney et al., 1999).As a results, the required nitrogen and water needed to fix a given amount of carbon is reduced (Chapin et al., 2007), though in some cases, acclimation has no effect on photosynthetic rate and stomata conductance (Curtis and Wang, 1998).It is therefore, not clear to what extent modelling results are realistically since CO₂ enrichment experiments are lacking in the tropics.Little is known about the response of tropical forests to increased CO₂ concentration (Thomas et al., 2008).In our study, the correlations between tree ring indices and CO₂ concentration were not significant at all sites (Fig. S3), contrary to modelling results, indicating the need for more research.
The projected decreased NPP under RCP 8.5 at the Sesheke site results from high negative effects of the projected reduced rainfall coupled with increased temperature.NPP of the drier areas is mainly influenced by water by enhancing the water use efficiency of vegetation (Yu and Chen, 2016).Reduced rainfall decreases soil water availability needed by the plants.High temperature enhances evapotranspiration resulting in reduced soil moisture (Miyashita et al., 2005).When soil water decreases, the stomata close to restrict water loss.The closure of stomata prevents the movement of carbon into the plant, resulting in reduced NPP (McGuire and Joyce, 2005).Decreased soil water also limits nutrient absorption (e.The differences in NPP response to climate change at each of the study sites is especially caused by variability in rainfall and 5 nutrient distribution (Fig. 1 and Table 1).Though the photosynthesis process is dependent on CO₂ concentration, plant's response to increasing CO₂ is limited by the availability of soil water and nutrients.Thus, plants growing in poor nutrient condition respond less to rising CO₂ concentration (Lloyd and Farquhar, 1996).This could be the case with the reduced NPP response at Sesheke where nitrogen content is lower than at Kabompo and Namwala (Table 1) despite the increasing projected CO₂ concentration.However, deciduous trees sometimes acclimate to increased CO₂ concentration by reducing photosynthetic 10 capacity and stomatal conductance (Ellsworth, 1999;Mooney et al., 1999).As a result, the required nitrogen and water needed to fix a given amount of carbon is reduced (Chapin et al., 2007), resulting in decreased NPP.The different NPP response to climate change at the three sites could also be attributed to differences in species composition and the variable responses of these distinct tree species to the environment caused by variation in there physiological properties.
While 9 % of the total tree species are common in all the three sites, 25 % of the total surveyed species are found at Kabompo, 38 % at Namwala and 16 % at Sesheke only (Ngoma et al., 2018b).
Although we projected different NPP patterns at the three study sites, these projections depend on the accuracy of climate data.
In our study, we averaged climate data from five GCMs considering the different climatic values generated by each GCM.The use of ensembles improves the modelling of climate data compared to a single model, thereby reducing the uncertainty that might be caused by sources of climate data.
Generally, there are some similarities in the results we generated in our study with literature (Tables 3) for similar forest types.
The differences in actual values hint on the differences in models applied and the extent of area coverage.For example, while we conducted our study at local level, other researchers conducted similar studies at regional level (Doherty et al., 2010).
Studies conducted at regional level constitute average results of different biomes while our study covered one biome only at all the three sites.Other factors such as species composition and soils also differ between our study sites and study sites of other researchers.We compared our results to few studies due to limited literature on modelling studies reported for African biomes.Also, studies using the same model as our study (LPJ-GUESS) are limited in Africa.We could not find any studies applying LPJ-GUESS model at local level in Africa as most studies are conducted at global level (Cao and Woodward, 1998;Schaphoff et al., 2006).Availability of such studies would give much insight on our results.This therefore presents an opportunity to focus modelling research in Africa so as to determine the potential response of the different biomes to climate change.

Conclusions
We generated new soil texture and tree parameter values for maximum crown area, wood density, leaf longevity, and allometry.
The results simulated with the LPJ-GUESS model improved when we used these newly generated local parameters.NPP is projected to increase at the wetter Kabompo and intermediate Namwala sites under both scenarios especially caused by the increased carbon dioxide concentration by the end of 21 st century, while at the drier Sesheke site, NPP will decrease by the end of the 21 st century under both scenarios.The projected decreased NPP under RCP8.5 at the Sesheke site results from the reduced rainfall.We thus demonstrated that differences in rainfall pattern influence the way in which climate change will affect forests resources.We also showed that using local parameter values is essential to obtaining reasonably reliable simulations.
Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License. the study for the Zambian Zambezi teak forests at the Kabompo (14° 00.551S, 023° 35.106E),Namwala (15° 50.732S, 026° 28.927E), and Sesheke (17° 21.278S, 24° 22.560E) sites.At the Sesheke site, the Masese forest reserve was assessed while at the Namwala site, we assessed the Ila forest reserve.At the Kabompo site, we studied the Kabompo and Zambezi forest reserves.While the Masese forest reserve is found in the drier agro-ecological zone I, the Kabompo and Zambezi forest reserves are located in the wetter ecological zone II.The Ila forest reserve at the Namwala site stretches along ecological zones I and II (Fig. 1 and
Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.spin-up period, the model was forced with a 53-year observed climate and CO₂ values, corresponding to the 1959-2011 period at Namwala and Sesheke sites.We forced the model with a 45-year observed climate and CO₂, corresponding to the 1959-2003 period at Kabompo site.CO₂ had reached 375 ppm and 390 ppm by 2003 and 2011 respectively.Before forcing the model with projected climate data, we first spun-up the model with 30 years modelled climate data from 1960-1989 and a constant CO₂ of 317 ppm, corresponding to 1960.We then forced the model with 46-year contemporaneously modelled climate data for the period 1960-2005.We used CO₂ data for the same period of 1960-2005 and by 2005, CO₂ had reached 379 ppm.
at Kabompo and for 53 years at the Namwala and Sesheke sites.Forcing the model with observed climate data, and using local tree and soil parameters, we compared the LPJ-GUESS simulated carbon stocks and NPP with measured carbon stock(Ngoma et al., 2018a, b) and treering indices(Ngoma et al., 2017) respectively.We performed a factorial experiment for projected effects of temperature, rainfall, CO₂ concentration, incoming solar radiation, and number of wet days per month for the end of the 21 st century (2070-2099) following RCP 4.5 and RCP 8.5 scenarios.To isolate the contemporaneous effects of each of these climatic variables, the model was forced with the 1960-2005 values of the input climate variable of interest while keeping the 1960 values constant for the other input climatic variables.When assessing the projected effects, we forced the model with projected climate values for the period 2006-2099 of the input climate variable of interest, while keeping the 2006 value constant for the other input climatic variables.
2c) are projected to increase by the end of the 21 st century (2070-2099) at all sites under both scenarios relative to 1960-1989.Temperature increases by 3°C at all sites by end of the 21 st century under RCP 4.5 while, under RCP 8.5, temperature is projected to increase by 5°C at the Kabompo and Namwala sites, and by 6°C at the Sesheke site.

Figure 2 .
Figure 2. Projected changes in rainfall (a), temperature (b), incoming solar radiation (c), number of wet days (d), and carbondioxide concentration (e) under RCP 4.5 and RCP 8.5 by the end of 21 st century.End of the 21 st century is the period 2070-2099.Values were determined as means of the five GCMs and changes were determined with reference to 1960-1989 period as baseline.For sources of data, refer to Section 2.3.

Figure 4 .
Figure 4. Measured versus LPJ-GUESS simulated vegetation carbon stock simulated with default soil parameters, default tree parameters, and observed local climate (a), local soil, local tree parameters, and observed local climate (b), and with local soil, local tree parameters, and modelled contemporaraneously climate (c).
Running the LPJ-GUESS model with local soil and tree parameters, and forcing it with local observed climate data for the period 1960-2003, vegetation carbon stocks, and Leaf Area Index (LAI) were highest at Kabompo, and Sesheke had the lowest values.The aggregated three carbon pools (vegetation, litter, and soil carbon) were highest at Kabompo and lowest at Namwala.Vegetation carbon was lower when we forced the LPJ-GUESS model with contemporaneously modelled climate data for the period 1960-2003 at all sites compared to the values simulated with observed local climate data.Vegetation carbon stocks, LAI, and NPP simulated with both local soil and local tree parameters, and forcing the model with local climate data were lower at all sites compared to values generated by default tree and soil parameters apart from at the Sesheke site where NPP was higher by 0.0532 kg C m¯² year¯¹ (Fig. 7).Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 7 . 10 Figure 8 .
Figure 7. Mean annual vegetation carbon stocks (a ), LAI (b), and NPP (c) simulated with local and default soil and tree parameter values, and forcing the model with local and modelled climate data.Simulations were done for the period 1959-2003.5 Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.indices and rainfall (Fig. S2 (a) and (i)).The significant positive relationship between tree ring indices and rainfall of previous two years at Sesheke (Fig. S2 (i)) indicates a carry-over effects of rainfall on trees' productivity, an aspect not addressed in LPJ-GUESS model.This lack of representation of carry-over effects of rainfall in Dynamic Global Vegetation Models g. Nitrogen) by the roots and transportation to the plants.Increased temperature enhances plant respiration thereby reducing photosynthetic activities Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.(Burton et al., 2008;Wu et al., 2011).The projected reduced number of wet days likely have more effects on NPP at Sesheke under RCP 4.5 by the year 2099.The projected NPP decrease at Sesheke is in the same direction as the findings ofDelire et al. (2008) who reported an NPP reduction of 12 % for the savanna forests by 2080.Alo and Wang (2008) also projected NPP decrease in west and southern Africa.

Table 1 .
Climate and soil characteristics at Kabompo, Namwala, and Sesheke.For rainfall and temperature, the period covered for average values presented are given in brackets.
while climate data was collected from local weather stations and the dataset of the Coupled Model Inter-comparison Project phase 5 (CMIP5).Forcing data on observed temperature rainfall, and cloud cover were collected from local weather stations within the respective ecological zones.We collected climate data from 15, 13, and 28 weather stations for Sesheke, Kabompo and Namwala sites respectively.The surveyed Ila forest reserve at the Namwala site stretches in zones I and II, thus climate data was averaged from all local weather stations in both zones.Contemporaneously number of wet days were downloaded from Climatic Research Unit (CRU) website (University of East Anglia Climatic Research Unit et al.

Table 2
for full names).Using more than one scenario and the use of ensemble means reduces the uncertainty in projected climate data compared to a single model.The climate data was re-gridded from the original spatial resolution of the climate model to a resolution of 0.5° x 0.5°.Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.
spun-up the model with observed climate data from local weather stations and contemporaneously modelled climate data during the respective model runs.Observed climate data are temperature, rainfall, and cloud cover data observed from local weather stations in the respective study sites, while contemporaneous data on CO2 concentration were downloaded from the RCP database (RCP Database, 2018).Data on the number of wet days per month were downloaded from Centre for Environmental Data Analysis (University of East Anglia Climatic Research Unit et al., 2015).Contemporaneously modelled climate data are temperature, rainfall, number of wet days per month, and solar radiation averaged from the five GCMs described under section 2.4, and CO2 concentration data downloaded from RCP data base(RCP Database, 2018).Using observed local climate data, we forced LPJ-GUESS during the spin-up with repeated cycle of 30-year climate data for 1959-1988 and a constant CO₂ concentration of 316 ppm, corresponding to the observed value for 1959.After the 1000-year

Table 3 .
Projected changes in NPP: Current study compared to literature Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-421Manuscript under review for journal Biogeosciences Discussion started: 24 October 2018 c Author(s) 2018.CC BY 4.0 License.Lund-Potsdam-Jena General Ecosystem Simulator ORCHIDEE ORganizing Carbon and Hydrology in Dynamic EcosytEms CEVSA Carbon Exchange between Vegetation, Soil, and the Atmosphere DLEM Dynamic Land Ecosystem Model