Aboveground Carbon Storage and Its Links to Stand Structure, Tree Diversity and Floristic Composition in South-Eastern Tanzania

Aboveground Carbon Storage and Its Links to Stand Structure, Tree Diversity and Floristic... Ecosystems (2018) 21: 740–754 DOI: 10.1007/s10021-017-0180-6 2017 The Author(s). This article is an open access publication Aboveground Carbon Storage and Its Links to Stand Structure, Tree Diversity and Floristic Composition in South-Eastern Tanzania 1 1 1 2,3 Iain M. McNicol, * Casey M. Ryan, Kyle G. Dexter, Stephen M. J. Ball, 1,4 and Mathew Williams School of Geosciences, University of Edinburgh, Crew Building, Alexander Crum Brown Road, Edinburgh EH9 3FF, Scotland, UK; 2 3 Mpingo Conservation and Development Initiative, Kilwa Masoko, United Republic of Tanzania; Present address: Farm Africa, Dar Es Salaam, United Republic of Tanzania; The National Centre for Earth Observation, Natural Environment Research Council, Swindon, UK ABSTRACT African savannas and dry forests represent a large, disproportionately contribute to AGC, with the lar- but poorly quantified store of biomass carbon and gest 3.7% of individuals containing half the carbon. biodiversity. Improving this information is hindered Tree species diversity and carbon stocks were posi- by a lack of recent forest inventories, which are tively related, implying a potential functional rela- necessary for calibrating earth observation data and tionship between the two, and a ‘win–win’ scenario for evaluating the relationship between carbon for conservation; however, lower biomass areas also stocks and tree diversity in the context of forest contain diverse species assemblages meaning that conservation (for example, REDD+). Here, we pre- carbon-oriented conservation may miss important sent new inventory data from south-eastern Tanza- areas of biodiversity. Despite these variations, we find nia, comprising more than 15,000 trees at 25 that total tree abundance and biomass is skewed to- locations located across a gradient of aboveground wards a few locally dominant species, with eight and woody carbon (AGC) stocks. We find that larger trees nine species (5.7% of the total) accounting for over half the total measured trees and carbon, respec- tively. This finding implies that carbon production in these areas is channelled through a small number of Received 17 February 2017; accepted 15 August 2017; relatively abundant species. Our results provide key published online 6 September 2017 insights into the structure and functioning of these heterogeneous ecosystems and indicate the need for Electronic supplementary material: The online version of this article (doi:10.1007/s10021-017-0180-6) contains supplementary material, novel strategies for future measurement and moni- which is available to authorized users. toring of carbon stocks and biodiversity, including Author contributions: CMR and MW developed the experimental design for plot establishment and produced the vegetation classification the use for larger plots to capture spatial variations in upon which plot location was based. Inventory data were collected by large tree density and AGC stocks, and to allow the IMM, CMR and partners at Mpingo Conservation and Development calibration of earth observation data. Initiative (MCDI) led by SMJB, with financial support from the Royal Norwegian Embassy in Tanzania. Richard Lamprey from Flora and Fauna International, a project partner of MCDI, provided estimates of canopy Key words: aboveground carbon storage; tree cover based on aerial photographs taken over the permanent sample diversity; Africa; miombo; large trees; biomass– plots. IMM, MW and CMR conceived the research questions. IMM col- lated and analysed the data and wrote the manuscript with input from biodiversity relationship; tree species composition; MW, CMR, SMJB and KGD. permanent plot; monitoring. *Corresponding author; e-mail: i.mcnicol@ed.ac.uk 740 Carbon and Tree Diversity in an African Landscape 741 Increasing human pressure linked to resource INTRODUCTION extraction is currently driving widespread, but Seasonally dry tropical forests and woodlands are uncertain losses of AGC, as well the localised the dominant vegetation cover in southern Africa, extinction of important tree species (Ahrends and extending over 4 million km across 10 countries others 2010; Ryan and others 2012; Jew and others (Mayaux and others 2004). Across their range, 2016). It is therefore important to quantify and variations in climate, soils and disturbance main- reduce uncertainty in our estimates of AGC storage, tain a structurally and floristically diverse mosaic of to better understand future losses, and to underpin habitats, covering a spectrum from open savanna carbon sequestration initiatives aimed at mitigating with a dominant grass layer and scattered trees, this loss. Plot-level estimates of AGC storage are through open canopy savanna woodland with an fundamental for calibrating and interpreting earth understory of grasses and shrubs, to denser wood- observation data, which can then be used to map lands and dry forest (White 1983). The most regional patterns in AGC (Avitabile and others extensive of these formations are the miombo 2016) and its changes over time (Ryan and others woodlands, distinguishable from surrounding veg- 2012). etation types by the dominance of the genera Measuring and managing ecosystems based on Brachystegia and Julbernardia (Fabaceae, Caesalpin- their carbon stocks, particularly under the umbrella ioideae) (Chidumayo 1997). The region as a whole of Reducing Emissions from Deforestation and is highly biodiverse and a priority for conservation Degradation (REDD+), may also benefit biodiver- (Mittermeier and others 2003; Brooks and others sity research and conservation (Scharlemann and 2006), with the miombo woodlands alone thought others 2010; Hinsley and others 2014; Ahrends and to harbour an estimated 8500 species of higher others 2011). It is therefore useful to quantify how plants, including more than 300 tree species (Frost tree diversity and floristic composition co-vary with 1996), many of which are endemic to the region. AGC storage (Hinsley and others 2014) to highlight The range of species supported by the ecosystem any important trade-offs and thus inform mutually helps to underpin the livelihoods of an estimated beneficial conservation schemes (Miles and Kapos 150 million rural and urban dwellers who rely 2008;Dı´az and others 2009; Venter and others heavily on the timber, food, medicine and con- 2009). Such information may also be useful in struction materials that the woodlands and forests elucidating a potential functional relationship be- provide (Ryan and others 2016). tween AGC storage and tree diversity, which could Yet despite their scale and importance for local have additional benefits for conservation if higher livelihoods, the ecology and functioning of these tree species diversity also results in higher AGC seasonally dry ecosystems remain poorly studied in storage. The majority of the current evidence base comparison with the more carbon dense moist for or against a biomass–biodiversity relationship tropical forests in South America (Fauset and others comes from the moist tropical forest biome (Sulli- 2015; Poorter and others 2015), and to a lesser ex- van and others 2016; Chisholm and others 2013), tent, those in Central Africa (Lewis and others and it is still unclear whether these patterns (or lack 2013). As a result, the miombo eco-region still rep- thereof) hold true in drier, mixed tree-grass sys- resents a potentially large, but poorly quantified tems. store of biomass carbon, biodiversity and species Despite the comparatively high diversity of the endemism (Platts and others 2010; Halperin and tropical forest biome, recent studies have found others 2016; Ryan and others 2016; Shirima and that a small number of relatively large trees and others 2011; Jew and others 2016). Forest inventory species contribute disproportionately to tree abun- plots with which to quantify these variables are few dance and AGC stocks in a variety of moist tropical in number and spatially uneven, typically favouring forest ecosystems (ter Steege and others 2013; higher biomass stands and protected areas (Chidu- Fauset and others 2015; Marshall and others 2012; mayo 2013; Ribeiro and others 2008; Marshall and Bastin and others 2015). The evidence base for others 2012; Willcock and others 2014; Ryan and similar patterns in the miombo eco-region is lim- others 2011; Chidumayo 2002). Thus, many ited by a paucity of detailed forest inventories important ecological questions remain poorly re- across a range of representative vegetation types solved, for example, around the magnitude and and ecosystems (Marshall and others 2012; Frost distribution of aboveground woody carbon stocks 1996; Shirima and others 2011). From a measure- (AGC) across these heterogeneous landscapes, and ment perspective, knowing which tree size classes how this relates to patterns in vegetation structure, contain most of the carbon and species diversity tree species diversity and composition. 742 I. M. McNicol and others may also help improve knowledge of how best to the coastal plains to the east up to 740 m m.a.s.l design effective data collection protocols which can along the steep escarpment running north to south be used to expand the current plot network (Mar- dissecting the centre of the district. Approximately shall and others 2012;Re´jou-Me´chain and others 85% of the local population is rural and dependent 2014; Bastin and others 2015). on natural resources for their livelihoods (Khatun In this paper, we aim improve the knowledge of and others 2016). From October 2010–October ecosystem structure and function across these 2011, permanent sample plots were established at heterogeneous landscapes using data collected from 25 locations, originally stratified by three major a new network of 25 forest inventory plots in vegetation types delineated via a supervised land south-eastern Tanzania, which spans a gradient of cover classification, based on Landsat 5 data and woody biomass and different vegetation types. 300 in situ visual assessments of land cover, to Specifically, we explore (1) how patterns in AGC ensure that potential variations in AGC stocks had stocks are related to differences in tree size and been suitably captured (Figure 1). The vegetation number, (2) to tree species diversity within plots types for the original stratification included grass (a-diversity) and (3) to tree species composition. dominated ‘savannas’ with sparse tree cover, sa- vanna woodland (tree-grass mix) and dense woodland and forest (closed tree canopy with no METHODS grass cover), with the number of plots measured Study Area and Sampling Strategy proportional to the areal extent of each vegetation type. Tree canopy cover was estimated by outlining The study area is located in Kilwa District in the the crowns of individual trees identified using Lindi Region of south-eastern Tanzania (Figure 1). aerial photographs collected over the plots in The estimated mean annual precipitation is October 2010 (Figure 1). Pragmatism played a role 821 ± 350 mm (±SD), with a gradient between in site location, with plots located randomly along the east (wetter) and west (drier) (Tropical Rainfall the road and track network (Figure 1); however, a Measurement Mission, 3B43 product; Huffman and 1-km buffer from tracks was enforced to reduce the others 2007). Altitude varies from sea level along Figure 1. Location of our field plots and associated aboveground woody carbon stock (AGC) and canopy cover estimates. Sub-panel A shows the location of Tanzania, and the extent of the miombo woodlands—the dominant vegetation type in our study region, with sub-panel B showing the location of our study region. C Location of our field plots, and the initial land cover classification used for plot location. D The distribution of plot (1 ha) AGC stocks and canopy cover estimates. Carbon and Tree Diversity in an African Landscape 743 Figure 2. A Cumulative percentage of AGC stocks contributed by different tree size classes within plots of similar AGC and canopy cover; B the average number of trees within each size class. Each data point represents the average contribution of plots within each group. likelihood of intense human disturbance. For Kolmogorov–Smirnov tests were used to test sampling, we utilised a 1-ha (100 9 100 m) sized whether the distribution of plot-level AGC in each permanent sample plot in which all trees with a size class was statistically different between plots of diameter of at least 5 cm were recorded, tagged and broadly similar AGC and structure (tree density and spatially located. These 1-ha plots, upon which canopy cover), under the null hypothesis that the most of the analyses in this study are based, were distributions are similar and that variations in AGC nested centrally within a larger 9-ha storage reflect differences in tree density. (300 9 300 m) plot in which only trees larger than To assess species composition and diversity, we 40 cm were recorded. Tree diameter was measured used the species names or genus where known. at 1.3 m height above the ground, and if the tree Where this was not possible, the local name was forked below 1.3 m, each stem was measured and used instead. In some cases, the use of local names counted as one individual. We recorded the local may result in tree species diversity being overesti- name of each measured tree, and where possible, mated if multiple names are used for a single spe- identified each by their scientific name using col- cies; however, the more likely scenario is that lected voucher specimens and published reference diversity will be underestimated as the same local guides (Coates-Palgrave and Moll 2002). Where name is often used for several species (for example, this was not possible, species were identified using based on local usage), with some species also likely a range of local and national species lists (NA- to be indistinguishable without fertile material FORMA 2011). leading to some species being conflated (Ahrends and others 2011). To minimise errors due to the Data Analysis former, we used the same botanists for all plots to ensure species identification was consistent across Aboveground carbon stocks (AGC) were calculated plots. Controlling for the latter is more difficult. using an allometric model developed in the same However, on average, trees identified only by local administrative region (Lindi model: Mugasha and name contributed no more than five of the species others 2013), with biomass assumed to be 47% measured in each plot and thus we consider the carbon. To address our first question about how likelihood that our diversity measures are subject to variations in AGC stocks are related to differences meaningful bias to be small. A small numbers of in stand structure, specifically size and number, individuals that were not identified to any taxo- trees were binned into 5-cm size classes and the nomic level (0.07% of total inventory) were ex- proportional contribution of each size class to the cluded from the analysis. total measured AGC in each plot was calculated. 744 I. M. McNicol and others Tree species diversity was calculated using three To examine how our results (that is, tree diver- measures: species richness, Fisher’s alpha and rar- sity and AGC estimates) would have differed had efied richness. For rarefied richness, we used Mao- we sampled progressively smaller plots instead of Tao individual-based rarefaction analysis. When the 1-ha plots, we simulated single sub-plots of comparing tree diversity and AGC, diversity is re- varying size (0.1, 0.25 and 0.5 ha) at random garded as the independent variable under the locations within each of the 25 9 1 ha plots, with assumption that tree diversity has a deterministic the sub-sampling analysis repeated 1000 times to effect on AGC at the plot level (due to niche ensure the full range of possible subsets was complementarity and selection effects), as opposed achieved. For each subplot, we calculated the tree -1 to if the axis were reversed, which would assume species richness and AGC density (tC ha ) and environmental/disturbance controls on diversity, compared these as a percentage of the corre- which we believe are more likely to occur at larger sponding estimates from the 1-ha plot. For each scales than our field plots (Chisholm and others iteration, we totalled the number of species across 2013; Woollen and others 2012). Multiple models the network to show how sampling smaller plots were fitted to each data set using a variety of across the entire network would have impacted our functional forms based on ecological theory, estimates of landscape diversity. including a linear relationship ðy ¼ ax þ b), satu- All data analyses were performed using the R ration ðy ¼ ax=ðÞ b þ t , quadraticðÞ y ¼ ax þ bx þ c statistical software version 3.0.2 (R Core Team bx 2014, http://cran.r-project.org) and the ‘vegan’ and a parabolic ricker curve y ¼ axe . Model package (version 2.0-10; Oksanen 2013). selection was based on minimising the Akaike information criterion (AICc), corrected for small sample sizes, and the residual sum of squared dif- RESULTS ferences. Patterns in Aboveground Woody Carbon Diversity measures were taken for all trees Stocks and Stand Structure (>5 cm) in each 1-ha plot, then again for small trees (5–15 cm), medium sized trees (15–40 cm) In total, we surveyed 13,098 trees (>5 cm) across and large canopy dominants (>40 cm) separately, the 25 one-ha plots, including 10,694 small trees with the aim of understanding where most of the (5–15 cm), 2139 medium sized trees (15–40 cm) tree diversity occurs in these systems. For the and 265 large trees (>40 cm). The surrounding 9- analysis of large tree diversity (>40 cm), data from ha plots contained an additional 2069 large trees, the 9-ha plots were included to allow a suit- highlighting the importance of larger plots for able number of trees for analysis. Differences in adequate statistical analyses of large trees. AGC species composition between plots (b-diversity) -1 stocks in the 1-ha plots ranged from 2 tC ha in were calculated using the Bray–Curtis Index of -1 an area of open grassland savanna to 54 tC ha in Species Dissimilarity. Overall compositional pat- an area of dense forest (Figure 1), with an overall terns were visualised using non-metric multidi- -1 landscape average of 24 ± 16 tC ha (±indicates mensional scaling, which was performed using the standard deviation throughout). ‘metaMDS’ function. Permutational multivariate This gradient in AGC stocks is associated with analysis of variance (PerMANOVA) was used to test clear changes in both tree density (72–1511 tree- -1 whether there were significant differences in tree sha ; Spearman’s rho, R = 0.95, P < 0.001) and species composition between groups of plots (An- tree canopy cover, with areas of <10% cover— derson 2001). The analysis was repeated separately broadly consistent with the FAO definition of for small, medium and large trees to test whether ‘other wooded lands’ (FAO 2001)—storing -1 composition differed among size classes. Prior to <10 tC ha (n = 7), with plots in more open ca- analysis, the raw species abundance data were nopy savanna ‘woodlands’ (10–45%) storing 15– -1 square root transformed and site standardised to 35 tC ha (n = 12), and plots in more closed ca- -1 account for the number of trees sampled at each nopy ‘forests’ (>50%) containing >40 tC ha site and to reduce the influence of the most com- (n = 6). Large trees contributed around one-third mon species (Barlow and others 2007). We used (32 ± 18%) of plot AGC, despite comprising only ANOVA and Tukey’s HSD tests to look for signifi- 2.6 ± 2.2% of the trees in each plot. Overall, half cant differences in tree structure and diversity be- of the total measured biomass (across the 1-ha tween groups of plots after testing the data for plots) was stored in the 484 largest trees, which normality using Shapiro–Wilk tests. comprised 3.7% of the total trees measured. Carbon and Tree Diversity in an African Landscape 745 Table 1. Top 5 Dominant Species Within Plots of Broadly Similar AGC Stocks Ranked by Their Contribution to the Total Carbon Stock and Total Tree Abundance -1) -1 -1 -1 Rank Low AGC (0–10 tC ha Low to moderate AGC (15–25 tC ha ) Moderate to high AGC (25–40 tC ha ) High AGC (>45 tC ha ) -1 AGC (tC ha ) 1 Diospyros quiloensis Dalbergia melanoxylon Julbernardia globiflora Hymenocardia ulmoides 2 Sclerocarya birrea Pseudolachnostylis maprouneifolia Brachystegia spiciformis Hymenaea verrucosa 3 Combretum apiculatum Julbernardia globiflora Combretum apiculatum Rytigynia sp. 4 Dalbergia melanoxylon Combretum apiculatum Burkea africana Pteleopsis myrtifolia 5 Burkea africana Brachystegia spiciformis Diplorhynchus condylocarpon Euphorbia nyikae % of total 53.6 44.3 60.9 49.7 -1 Stocking density (trees ha ) 1 Combretum apiculatum Diplorhynchus condylocarpon Diplorhynchus condylocarpon Hymenocardia ulmoides 2 Spirostachys africana Combretum apiculatum Combretum apiculatum Suregada zanzibariensis 3 Acacia nilotica Dalbergia melanoxylon Pseudolachnostylis maprouneifolia Euphorbia nyikae 4 Burkea africana Pseudolachnostylis maprouneifolia Hymenocardia ulmoides Uvaria lucida 5 Bauhinia petersiana Bridelia scleroneura Julbernardia globiflora Strychos spinosa % of total 62.8 55.4 64.7 52.0 n plots 7 7 8 3 Species richness 15 (6) 26 (8) 32 (7) 42 (4) Fisher’s a 4.2 (2.3) 6.4 (2.3) 7.5 (1.7) 8.2 (0.7) Total species richness 56 74 95 87 a b b c Bray–Curtis 0.77 (0.11) 0.69 (0.14) 0.52 (0.10) 0.61 (0.10) Number of unique species 9 10 26 32 Plots with a moderate and high AGC density are further separated to better highlight changes in tree species dominance over the gradient, particularly in our three highest AGC plots (> 60% tree canopy cover) which are marked out as floristically distinct from other high AGC plots (Figure 4). Additional information includes the mean tree species richness and Fisher’s a in each plot (±SD), the total number of species recorded in each group, as well as the number that are unique to each group, and the average species dissimilarity between plots [Bray–Curtis Index (±SD)]. The letters in superscript next to the Bray–Curtis index indicate the results of the PerMANOVA which tested whether trees species composition significantly differed between groups of plots. 746 I. M. McNicol and others Table 2. Diversity Indices for Group of Plots Separated by Broad Size Class Size class Small trees (5–15 cm DBH) Medium trees (15–40 cm DBH) Large trees (40 cm + DBH) Low AGC Species richness 14 (7) 6 (3) 7 (4)* Fisher’s a 3.2 (2.0) 3.8 (3.7) 3.4 (2.0) a a a Bray–Curtis Index 0.77 (0.11) 0.89 (0.12) 0.77 (0.12) Moderate AGC Species richness 22 (6) 15 (5) 15 (5)* Fisher’s a 5.5 (1.7) 6.0 (2.3) 4.9 (1.6) b b b Bray–Curtis Index 0.66 (0.14) 0.67 (0.13) 0.64 (0.17) High AGC Species richness 28 (9) 19 (4) 17 (4)* Fisher’s a 6.2 (1.1) 5.6 (1.7) 5.0 (1.1) c b b Bray–Curtis Index 0.73 (0.14) 0.74 (0.15) 0.74 (0.16) As in Table 1 information includes the average species richness and Fisher’s a (±SD) for different size classes within each plot. The Bray–Curtis Index is used to highlight difference in floristic composition within plots. The letters in superscript indicate the results of the PerMANOVA which tested whether the composition of small, medium and large trees significantly varied between groups of plots. Includes the measured trees from the 9-ha plot meaning that comparisons of large tree species richness are only valid between groups, and not between size classes due the larger sample area for large trees compared to medium and smaller trees. Figure 3. Relationships between tree species richness and aboveground woody carbon stocks. Ordinary least squares (OLS) regression models are fitted to the data; A tree species richness (y = 1.15–6.67, r = 0.63, P = <0.001) and B rarefied richness (y = 1.95–5.12, r = 0.22, P = 0.01). The distribution of carbon stocks among tree size between our moderate and high AGC density plots classes differed significantly between our low AGC (P = 0.51), despite a clear trend towards greater -1 density plots (<10 tC ha ) and those with a tree size (that is, >80 cm DBH) at the upper end of moderate and high AGC density (Kolmogorov– the gradient, where these very large trees had a Smirnov; P = <0.001 in both cases). In the low disproportionate contribution to plot AGC (10%) -1 AGC density, typically grassland savanna plots, the relative to their abundance (1 ± 1ha ) (Fig- majority of AGC (42%) was contributed by the ure 2). smallest diameter classes (5–15 cm) (Figure 2), whereas in moderate density savanna ‘woodlands’ Patterns in Tree Species Composition and higher AGC density ‘forest’ plots, the propor- and Diversity tion of AGC stored in small trees was relatively low We identified 158 morphospecies across the (15%), despite the greater number of trees in 25 9 1 ha plots by their local species name, of these areas. There were no significant differences in which 91 were fully identified to species level the distribution of AGC among different size classes Carbon and Tree Diversity in an African Landscape 747 Figure 4. A Plot-pair differences in tree species composition with differences in plot-level AGC stocks; B NMDS ordi- nation based on the Bray–Curtis Index which is used to uncover the main compositional patterns across the gradient in AGC storage. (57%) and a further 16 to genus (10%), with 32 biomass grassland savannas and savanna wood- taxonomic families present. In the surrounding 9- lands (39; 21%), compared to the ‘forests’ (25; 6%) ha plots (>40 cm DBH trees only), 79 morphos- where they were few in number, but large. This pecies were identified, including 26 not found in pattern was also true for potentially nodulating the 1-ha plots, with 54 (68%) of these identified to legumes (Caroline Lehmann and others unpubl. species level, and 3 (4%) to genus, with a further data.) which were almost absent in high AGC three families represented. In both 1- and 9-ha areas, yet gradually more common as AGC stocks plots, the identified taxa contributed 96% of the decreased, comprising 40% of trees in low density total measured trees and AGC across all sites. The plots. data presented in the following sections are from A small number of species were both abundant the 1-ha plots unless otherwise stated. and widespread, with 8 species collectively con- Tree species richness ranged from 9 to 45 per plot tributing over 50% of the trees measured, includ- with both richness and Fisher’s a significantly ing Diplorhynchus condylocarpon (15.9% of all trees; higher in the moderate and high AGC density plots n plots = 17), Combretum apiculatum (10.6%; compared to the lowest density plots (ANOVA + n = 21), and to a lesser extent, Hymenocardia ul- Tukey HSD, P < 0.01) (Table 1). The results were moides (9.9%; n = 8) and Pseudolachnostylis the same when comparing small, medium and maprouneifolia (3.6%; n =16). A similar level of large trees separately (Table 2). Tree species rich- dominance was observed when assessing species ness exhibited a positive linear relationship with contributions to the total carbon stock, with just 9 AGC storage (r = 0.63, P < 0.001) (Figure 3). The species, including the four aforementioned species, significant trend was maintained when controlling containing over half (52.5%) of the total AGC. The for tree density (rarefied richness), though the remaining biomass dominant species were Jul- relationship was markedly weaker (r = 0.22, bernardia globiflora (15.4% of total measured bio- P = 0.01) (Figure 3), indicating that differences in mass), Brachystegia spiciformis (7%), Burkea Africana tree density partly drive this relationship. (4.5%), Pteleopsis mytifolia and the priority conser- Euphorbiaceae was the dominant family across vation and timber species, Dalbergia melanoxylon, the plot network, comprising 39% of the total with the remainder either commonly used for measured AGC and 17% of trees, followed equally charcoal (P. myrtifolia), or occasionally harvested by Combretaceae and Fabaceae (each 21% of for timber. A similar level of species dominance was AGC and 11% of trees), and Apocynaceae (12; observed within each of the broad vegetation types, 17%). Familial dominance differed among vegeta- with approximately 5 species contributing over half tion types with trees in the family Euphorbiaceae of the AGC stocks and trees (Table 1). more common in areas with an AGC density The large majority of species were considerably -1 greater than 40 tC ha (39; 24%), with those in less abundant, with 49 species (31% of total) con- Fabaceae proportionally more dominant in lower tributing fewer than 50 individuals. Many of the 748 I. M. McNicol and others recorded species were restricted to particular habi- with more disturbed systems. The results highlight tats, with nine restricted to the low AGC plots, 36 the obvious importance of maintaining a low DBH to plots with a moderate AGC density, with 32 threshold (that is, 5 cm) in lower biomass stands in species only found in the three highest AGC ‘forest’ order to capture and quantify the majority of AGC plots (Figure 4). Species turnover (b-diversity) stocks. among plots was therefore relatively high, with In the more carbon dense savanna woodlands some areas of similar AGC found to contain entirely and dry forest plots, a greater proportion of AGC different species assemblages (Figure 4). The lowest was contained in larger trees, with the relative AGC plots were the most heterogeneous (Table 1), proportion contained in different size classes sta- as shown by the NMDS ordination plot tistically similar between plots in moderate (10– -1 -1 (stress = 0.12, n dimensions = 3) and were floris- 35 tC ha ) and high AGC (>40 tC ha ) stands. tically distinct to both the moderate and high AGC We therefore conclude that the variations in AGC plots, both when considering all tree together stocks between these areas are due to differences in (>5 cm) (PerMANOVA, P < 0.001; Figure 4; Ta- tree abundance in each size class, although there is ble 1) and small, medium and large trees separately some evidence to suggest that these differences (Table 2). may also reflect the greater density of very large Despite the wider range of AGC storage, we ob- trees (‡80 cm) in forests, which typically numbered served a greater compositional similarity among the only one per hectare in the most carbon dense -1 moderate density ‘woodland’ plots (15–40 tC ha ), ‘forest’ plots (>50% canopy cover), yet con- which tend to be dominated terms of AGC contri- tributed on average 8% of the measured AGC. bution by two of the defining miombo woodland These very large trees were comparatively rare in species—J. globiflora and B. spiciformis—and in the low density, typically grassland savanna plots; number by D. condylocarpon and C. apiculatum (Ta- however, where a very large tree was present on a ble 1). At the upper end of the gradient, species plot (>94.9 cm, Diospyros quiloensis), its contribu- characteristic of wet miombo woodland and coastal tion to the total measured AGC was considerable forest was common, including Suregada zanz- (50%). ibariensis and Hymenaea verrucosa. This shift in tree The concentration of biomass in a small number composition is reflected in the NMDS plot with the of trees has been previously observed in other three highest AGC plots—two of which were located moist forest ecosystems (Bastin and others 2015; at relatively high elevations along an escarpment Fauset and others 2015; Slik and others 2013)and (Figure 1)—exhibiting clear differences in compo- has clear implications for the development of ra- sition (Figure 4), both when considering all trees pid, low-cost forest monitoring protocols. In more -1 together, and when comparing trees in different size wooded areas (that is, >10 tC ha /% canopy classes (PerMANOVA; P < 0.001; Tables 1, 2). cover), large trees—that is, those larger than 40 cm—comprised approximately 40% of the biomass measured in each plot, with half the plot DISCUSSION AGC contained in the top 4.9% of trees (range Links Between Vegetation Structure and 2.7–9%; n trees = 9–64; minimum DBH = 24– Aboveground Carbon Storage 46 cm). These results are consistent with the re- sults of Bastin and others (2015)who detected a Our landscape-level estimates of aboveground similar concentration (that is, 50%) of plot bio- -1 carbon (AGC) stocks (24 ± 16 tC ha ) are similar mass in a similar proportion of trees (5% of total) to those recorded using similar approaches in across Central African moist forests. Similar results Mozambique by Ryan et al. (2011) were also found across an identical plot network -1 (21 ± 11 tC ha ) and Woollen and others (2012) in the miombo woodlands of Mozambique (Ryan -1 (21 ± 10 tC ha ), but lower than the regional 2009; Ryan and others 2011), where approxi- -1 average (28.7 ± 19.1 tC ha ) (Ryan and others mately 50% of plot AGC was contained in trees 2016) which includes many plots from protected larger than 40 cm DBH, suggesting this is a com- areas which are unlikely to be representative of the mon feature of miombo-dominated woodlands. wider miombo eco-region. Our lowest AGC plots, Our results contrast with those of Marshall and defined as areas with a tree canopy cover (%) and others (2012) who found that in the moist forests -1 AGC stock (tC ha ) of less than 10, were charac- of the Eastern Arc Mountains, trees larger than terised by a lower tree density, with the majority of 40 cm stored a much higher proportion (75–80%) trees (80%), and thus AGC (42%) contained in of plot AGC. smallest size classes (5–15 cm DBH), as is common Carbon and Tree Diversity in an African Landscape 749 The tendency towards greater tree size in plots at their maximum tree height (Nzunda and others the upper end of the gradient may be due to their 2014) and shade tolerance. In contrast, the noted location at moderate to high elevations (Marshall compositional similarities among the moderate and others 2012), suggesting a possible topo- density plots mean it is unlikely that differences in graphic, and/or edaphic influence on AGC storage composition are driving the within-vegetation type (Woollen and others 2012). These plots were also heterogeneity in AGC storage. Our results therefore more remote from human populations (Figure 1), suggest that compositional/functional differences meaning that historically lower levels of distur- may be more important in explaining the variation bance (human and ‘natural’) in these areas may between, rather than within vegetation types. have allowed larger trees to persist and AGC to Despite this diversity in tree species composition, accrue over longer periods. In the moderate AGC we find that total tree abundance and biomass is -1 density plots (10–35 tC ha ), we found no trees skewed strongly towards a relatively few locally larger than 75 cm DBH, yet in the surrounding 9- dominant species (Shirima and others 2011), with ha plots, several trees (n = 12) surpassed this limit 8 species (5.7% of the total) accounting for over (max. 112 cm), suggesting that in some cases, even half the measured trees and 9 species for greater 1-ha plots are unable to fully capture the stem size than 50% of biomass. A larger degree of biomass- distribution of woodlands (Anderson and others and stem-‘hyperdominance’ is found in the more 2009). This in turn may lead to high sampling er- diverse rainforests of both Amazonia (Fauset and rors when scaling AGC estimates across the land- others 2015; ter Steege and others 2013), and to a scape (Fisher and others 2008;Re´jou-Me´chain and lesser extent, Central Africa (Bastin and others others 2014), or remote sensing data of coarser 2015), although these results are derived from resolutions than the plots, such as the European much larger regional plot networks. In our study Space Agency’s Biomass mission, which will oper- area, the relatively large proportion of biomass lo- ate at a resolution of 4 ha (Scipal and others 2010). cated in such a small number of trees (90% is This mismatch again highlights the importance of contained in 38 species) suggests that most biomass sampling on a sufficiently large scale, either productivity in these seasonally dry ecosystems is through sampling many smaller plots, or a few also channelled through a relatively small number larger plots, to account for the inherent patchiness of tree species. The additional finding that greater of these ecosystems and presence of rare large trees. than 50% of the biomass is contained in moderate to high value timber suitable trees also highlights Relationship Between AGC Storage, Tree the future sensitivity of woody carbon stocks, and potentially productivity, in this area to logging and/ Species Diversity and Composition or charcoal production (Ahrends and others 2010). The inclusion of biodiversity as a co-benefit in From a conservation standpoint, our finding that carbon sequestration projects necessitates an more carbon dense areas also harbour the greatest assessment on how the two co-vary to assess tree species diversity suggests a ‘win–win’ scenario potential trade-offs, or co-benefits of conservation for forest conservation projects operating under the initiatives. From an ecological perspective, exam- umbrella of REDD+. Among the recorded species ining these linkages along with the extent to which were a number that are endemic to the remaining certain species contribute to carbon storage in these fragments coastal forest in the region, including H. systems, will help with efforts to reveal a more verrucosa and Uvaria kirkii, which is recorded as deterministic relationship between these two vari- ‘Near Threatened’ on the IUCN red list. Lower ables, and likely resilience of these ecosystems to biomass stands, particularly the miombo (Jul- future changes in land use (Hinsley and others bernardia—Brachystegia)-dominated ‘woodlands’, 2014). also contained a relatively diverse assemblage of We find clear differences in tree species compo- trees, including a number of high value timber sition along our AGC gradient, with the lowest species, such as Pterocarpus angolensis which is AGC stands and our three highest biomass plots commercially extinct in many parts of Tanzania marked out as being floristically distinct from the (Jew and others 2016) and classified as ‘Near spatially extensive, and moderate AGC density Threatened’, and the priority conservation species miombo-dominated ‘woodlands’. The composi- Dalbergia melanoxylon. A large number of species tional patterns suggest that the associated varia- were also found to be constrained to either mod- tions in AGC storage along the gradient may be erate or high density stands resulting in localised partially explained by differing functional traits patterns of species endemism. As such, the ‘win– between the dominant species in each area, such as win’ scenario indicated by our results does not 750 I. M. McNicol and others mean that comparatively low biomass areas should tions over whether tree diversity does indeed have be excluded from conservation efforts, as these a mechanistic effect on AGC storage and produc- areas may retain many locally and biologically tivity in these systems, which is important for important species, particularly in the understory understanding how changes in biodiversity will (that is, woody plats < 5 cm), and herbaceous affect these important ecosystem functions (Liang layers, as well as in faunal communities (Murphy and others 2016). It is also unclear whether more and others 2016), none of which were sampled in diverse tree communities help to create greater this study. diversity across multiple trophic levels, and whe- The preservation of biodiversity may have ther these communities also increase the ecosystem additional benefits if higher tree species diversity services provided to humans such as timber re- also results in higher AGC storage. Our finding of a sources and medicinal products (Maestre and oth- positive relationship between diversity and AGC ers 2012), both of which are important areas of storage is consistent with other observational future research. studies from both the miombo eco-region (Shir- ima and others 2015) and other forests globally Potential Implications for Future Tree (Ruiz-Jaen and Potvin 2010; Ruiz-Benito and Measurement and Monitoring others 2014;Vila and others 2007; Maestre and The need to acquire data on AGC stocks has taken others 2012; Liang and others 2016; Poorter and on added significance due to the rise in carbon others 2015). This positive relationship is consis- sequestration initiatives such as REDD+. The col- tent with theories of (1) niche complementarity, lection of species data also needs to be included in where a higher tree species richness leads to a any future measurement campaign to allow co- more functionally diverse community and thus variation between AGC and biodiversity to be ex- greater resource capture and biomass production; plored in the context of forest conservation (Venter and (2) selection effects, which posit that in al- and 2009; Liang and others 2016; Ahrends and ready dense stands there is a greater chance that 2011). Expanding the current network of perma- oneorafewhighlyproductivespecies arepresent nent inventory plots is a necessity, and a stan- (Fridley 2001). The absence of any clear saturation dardised methodology based on existing data sets is in the relationship at higher biomass levels, which crucial to rapidly facilitate the establishment of new would be suggestive of species redundancy or plots in the region and aid cross-plot comparisons. competitive exclusion, indicates that relatively To date, no studies have presented a clear view on dense patches of vegetation are still capable of the most appropriate and efficient strategy (that is, efficiently utilising available resources to allow sample size, plot size, appropriate DBH threshold) many species and high AGC stocks to coexist, for accurately measuring carbon stocks and/or suggesting that some form of complementarity or biodiversity in savanna woodlands (that is, Bar- facilitation is operating in these areas. Yet despite aloto and others 2013), a fact which is evidenced by the statistical significance of the relationships, the wide variety of sampling methodologies used to there was considerable variability in tree diversity for tree measurement (Ribeiro and others 2008; between plots, particularly after accounting for NAFORMA 2010; Chidumayo 2013; Ryan and differences in tree density. Recent studies from others 2011; Willcock and others 2014). The moist tropical forests indicate that diversity con- RAINFOR manual has provided some consistency trols on AGC storage operate at much smaller based on data collected in Amazonian forests scales than the ones observed here (0.1 ha) (Phillips and others 2009; Phillips and others 2003); (Chisholm and others 2013; Poorter and others however, there is no equivalent methodology for 2015; Sullivan and others 2016), which may ex- the dry tropics which are very different in terms of plain the lack of explanatory power. An alterna- their tree structure, diversity and composition tive explanation is that the greater diversity of tree (Fauset and others 2015; ter Steege and others species at higher AGC densities is the result of 2013). The results here provide some insights in more heterogeneous environmental conditions how sampling could be tailored in future to suit the within these areas, leading to greater species aims of a given project and its available resources. turnover related to habitat specialisation in certain For example, we show that in more wooded patches. High AGC may also occur in areas that -1 areas (>10 tC ha , >10% canopy cover), where have fewer major disturbances, allowing species stem size distribution is broadly consistent across less adapted to disturbance to persist. sites, measuring only those trees larger than 10 cm A full assessment of the biomass–diversity rela- DBH would have captured on average 93% of the tionship over larger scales will help answer ques- Carbon and Tree Diversity in an African Landscape 751 total AGC in each plot, yet would have required aloto and others 2013). Based on our data set, it is measuring 40% of the trees, or skipping on average unclear which of these sampling strategies (‘‘few -1 approximately 600 trees ha in denser woodlands large’’ vs. ‘‘many small’’ plots) is more appropriate -1 and dry forests (>40 tC ha ) and approximately for accurately and cost effectively capturing tree -1 275 trees ha in more open canopy savanna species diversity and composition in these areas. -1 woodlands (10–35 tC ha ). Raising the threshold Such information will be important for facilitating to 15 cm would still have captured 86% of the total conservation planning and implementation and AGC stocks in only 20% of the trees. We suggest will likely require the intensive (sub)-sampling of that such an approach would be ideal for con- very large plots to properly address this question ducting rapid inventories of AGC, such as for the (Baraloto and others 2013). calibration of earth observation data. The issue of plot size has additional importance Measuring for biodiversity and species composi- for measuring biomass, with smaller plots more tion would have very different requirements with likely to either overestimate, or completely miss the 50% of the species sampled here likely to be missed presence of rare, large trees, thus creating signifi- when measuring trees larger than 10 cm. These cant small scale variations in AGC stocks (Re´jou- species are likely to be among the rarest; therefore, Me´chain and others 2014; Fisher and others 2008; sampling at a higher DBH threshold will have little Chave and others 2004). Indeed, we find that even value when assessing the biodiversity or conser- the 0.5-ha plots produce highly variable AGC -1 vation value of these areas. Our results also suggest densities (tC ha ) relative to the corresponding that for a given site, the use of smaller inventory 1 ha values (5–95th percentile; 40–120%), tending plots (that is, <0.5 ha) (Willcock and others 2014; towards underestimation (median = 90%) (Chave NAFORMA 2010; Shirima and others 2015), which and others 2003). These sampling errors were are ideally suited for rapid sampling and often used exacerbated when using progressively smaller sub- for species measurement across the tropics plots, with 0.25 ha (25–150%) and 0.1 ha (14– (Stohlgren and others 1995; Baraloto and others 200%) plots generating an ever-larger range of 2013; Phillips and others 2003), are potentially possible AGC values relative to the 1-ha estimates. more sensitive to species clustering and/or likely to The 0.1-ha plots also produced anomalously high -1 exclude rare tree species (Baraloto and others values above 100 tC ha where a large tree(s) is 2013). For example, in the 9-ha plots, we find 26 present. For this reason, we would caution against species not in the 1-ha plots, despite measuring the use of very small plots (that is, <0.25 ha) for only those trees larger than 40 cm in these areas, measuring biomass as they can create large uncer- suggesting that even 1-ha plots fail to fully capture tainties on AGC stocks for a given site. However, if the species diversity at certain sites. We explored replicated in sufficient number, smaller plots may this potential issue further by sub-sampling the 1- still be suitable for estimating the average AGC ha plots which showed that the use of smaller plots density across the landscape, although such esti- would have captured on average 36 ± 13% mates may be less precise (Chave and others 2004). (0.1 ha), 53 ± 14% (0.25 ha) and 71 ± 14% This issue of plot size has clear relevance when (0.5 ha) of the plot-level tree species richness. considering the suitability of the plots for the cali- Hence, smaller plots clearly sample a smaller pro- bration of remotely sensed data; particularly radar portion of tree species for a given site than the 1-ha (for example, ALOS PALSAR) and LiDAR sensors, plots (Phillips and others 2003). However, sampling which in future will be the primary method for 0.5-ha plots instead of the 1-ha plots at each site upscaling ground based AGC estimates to the would still have captured a large majority landscape scale. Smaller plots (for example, (80 ± 2%) of the tree species found across the <0.25 ha) tend to be unsuitable for this purpose entire 1-ha network in only half the sample area, due to the aforementioned scaling issues, but also highlighting that the use of smaller plots may be their larger relative geo-location errors which may more efficient for gathering large-scale floristic be of similar size to the field plot (Ryan and 2012). data. The issue of many potentially rare tree species As a result, AGC stocks measured in larger plots are being missed in the smaller plots could be avoided if often found to exhibit a much stronger relationship sampling a larger number of these across the wider with the remotely sensed observation (Carreiras landscape; however, the physical and financial and others 2013;Re´jou-Me´chain and others 2014; challenges associated with repeat plot establish- McNicol 2014; Robinson and others 2013; Mauya ment and accessing typically remote areas may and others 2015). The mismatch in spatial scale outweigh the costs associated with establishing a between many of the current field inventory plots smaller number of well stratified larger plots (Bar- (Shirima and others 2011; Willcock and others 752 I. M. McNicol and others 2014; Ryan and others 2011) and the larger pixels License (http://creativecommons.org/licenses/by/ of future sensors such as the European Space 4.0/), which permits unrestricted use, distribution, Agency’s Biomass mission (4 ha) (Scipal and others and reproduction in any medium, provided you 2010) also has the potential to introduce consid- give appropriate credit to the original author(s) and erable errors when scaling plot even our 1 ha AGC the source, provide a link to the Creative Commons values to the size of the radar pixel (Re´jou-Me´chain license, and indicate if changes were made. and others 2014). The use of higher DBH thresh- olds would allow for larger areas (that is, >1ha) to REFERENCES be sampled in a more time and cost-efficient Ahrends A et al. 2010. Predictable waves of sequential forest manner, as was achieved in this study with the 9- degradation and biodiversity loss spreading from an African ha plots which were typically sampled in two-third city. Proc Natl Acad Sci USA 107(33):14556–61. of the time taken to sample the 1-ha plots; how- Ahrends A et al. 2011. Conservation and the botanist effect. Biol ever, this would clearly be at the detriment of Conserv 144(1):131–40. doi:10.1016/j.biocon.2010.08.008. biodiversity assessment. As shown here, the sam- Anderson M. 2001. A new method for non-parametric multi- variate analysis of variance. Aust Ecol 26:32–46. [online] URL: pling of large plots (that is, >1 ha) also has the http://onlinelibrary.wiley.com/doi/10.1111/j.1442-9993.2001. additional benefit of capturing of suitable of num- 01070.pp.x/full. Accessed December 13, 2013. ber larger trees, which will be useful for the anal- Anderson L, Malhi Y, Ladle R. 2009. Influence of landscape ysis of large tree mortality. heterogeneity on spatial patterns of wood productivity, wood The development of a standardised field protocol specific density and above ground biomass in Amazonia. that appropriately incorporates measurements of Biogeosciences 6:1883–1902. [online] URL: http://www. biogeosciences.net/6/1883/2009/bg-6-1883-2009.html. Ac- tree species diversity and aboveground carbon cessed August 5, 2013. stocks, but is also suitable for the calibration of Avitabile V et al. 2016. An integrated pan-tropical biomass map earth observation data, is urgently needed in order using multiple reference datasets. Glob Change Biol to ensure the best use of time and resources. For 22(4):1406–20. doi:10.1111/gcb.13139. this reason, we would suggest that larger sample Baraloto C et al. 2013. Rapid simultaneous estimation of plots (that is, ‡1 ha) should be favoured where aboveground biomass and tree diversity across Neotropical possible to capture potentially important variations forests: a comparison of field inventory methods. Biotropica 45(3):288–98. doi:10.1111/btp.12006. in large tree densities, and thus AGC stocks, Barlow J. et al. 2007. Quantifying the biodiversity value of whereas at the same time, allowing the plots to be tropical primary, secondary, and plantation forests. Proc Natl used as a calibration points for earth observation Acad Sci USA 104(47):18555–60. [online] URL: http://www. data, and facilitating cross-project comparisons pubmedcentral.nih.gov/articlerender.fcgi?artid=2141815&tool= (that is, RAINFOR). These plots may form part of pmcentrez&rendertype=abstract. nested sampling strategy to account for the differ- Bastin J.-F. et al. 2015. Seeing Central African forests through their ent data requirements, including the use of smaller largest trees. Sci Rep 5:1–8. [online] URL: http://www.nature. com/srep/2015/150817/srep13156/full/srep13156.html. plots (for example, 0.5 ha) for the sampling of tree Brooks TM et al. 2006. Global biodiversity conservation priori- species diversity, and potentially even smaller plots ties. Sciences 313(August):58–61. for sampling the understory and herbaceous layer, Carreiras J, Melo J, Vasconcelos M. 2013. Estimating the above- which was not sampled at all in this study, yet is a ground biomass in Miombo Savanna Woodlands (Mozam- major store of diversity in these ecosystems (Mur- bique, East Africa) using L-band synthetic aperture radar data. phy et al. 2016). Remote Sens 5(4):1524–48. [online] URL: http://www.mdpi. com/2072-4292/5/4/1524/. [Accessed July 22, 2014]. ACKNOWLEDGEMENTS Chave J et al. 2003. Spatial and temporal variation of biomass in a tropical forest: results from a large census plot in Panama. J We thank Nicholas Berry for his work on the pro- Ecol 91(2):240–52. ject. We also thank Deogratias Ndossi and Juvenal Chave J et al. 2004. Error propagation and scaling for tropical Pantaleo for co-managing the field data collection. forest biomass estimates. Philos Trans R Soc Lond Ser B Biol Sci 359(1443):409–20. IMM was supported as part of his PhD by the UK Chidumayo E. 2002. Changes in miombo woodland structure Natural Environment Research Council (NERC) under different land tenure and use systems in central Zam- and MCDI under their REDD Pilot Project funded bia. J Biogeogr 29:1619–26. [online] URL: http://online by the Royal Norwegian Embassy in Tanzania. library.wiley.com/doi/10.1046/j.1365-2699.2002.00794.x/full. [Accessed December 12, 2012]. OPEN ACCESS Chidumayo EN. 1997. Miombo ecology and management: an introduction. London: IT Publications in Association with the This article is distributed under the terms of the Stockholm Environment Institute. Creative Commons Attribution 4.0 International Carbon and Tree Diversity in an African Landscape 753 Chidumayo EN. 2013. Forest degradation and recovery in a Maestre FT. et al. 2012. Plant species richness and ecosystem Miombo woodland landscape in Zambia: 22 years of obser- multifunctionality in global drylands. Science 335(6065):214– vations on permanent sample plots. For Ecol Manag 291:154– 8.http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid 61. [online] URL: http://linkinghub.elsevier.com/retrieve/pii/ =3558739&tool=pmcentrez&rendertype=abstract. [Accessed S0378112712007074. [Accessed August 7, 2013]. March 19, 2014]. Chisholm RA. et al. 2013. Scale-dependent relationships be- Marshall AR. et al. 2012. Measuring and modelling above- tween tree species richness and ecosystem function in forests. ground carbon and tree allometry along a tropical elevation J Ecol 101(5):1214–24. [online] URL: http://doi.wiley.com/ gradient. Biol Conserv 154:20–33. http://linkinghub.elsevier. 10.1111/1365-2745.12132. [Accessed November 6, 2013]. com/retrieve/pii/S0006320712001607. [Accessed November 5, 2012]. Coates-Palgrave M. 2002. In: Moll E, Ed. Keith Coates-Palgrave trees of Southern Africa 3rd ed. Cape Town: Random House Mauya E. et al. 2015. Effects of field plot size on prediction Struik Publishers. accuracy of aboveground biomass in airborne laser scanning- assisted inventories in tropical rain forests of Tanzania. Carbon Dı´az S, Hector A, Wardle DA. 2009. Biodiversity in forest carbon Balance Manag 10(1):10. [online] URL: http://www.scopus. sequestration initiatives: not just a side benefit. Curr Opin com/inward/record.url?eid=2-s2.0-84928886552&partnerID= Environ Sustain 1(1):55–60. [online] URL: http://linkinghub. tZOtx3y1%5Cnhttp://www.scopus.com/inward/record.url?eid elsevier.com/retrieve/pii/S1877343509000177. [Accessed =2-s2.0-84928886552&partnerID=40&md5=83db6c72f8a246e November 1, 2012]. 4454e8e8e5c48a9d5. FAO. 2001. Global forest resources assessment 2000. [online] Mayaux P. et al. 2004. A new land cover map of Africa for the URL: https://www.soils.org/publications/sssaj/abstracts/73/6/ year 2000. J Biogeogr 31:861–77. [online] URL: http:// onlinelibrary.wiley.com/doi/10.1111/j.1365-2699.2004.0107 Fauset S. et al. 2015. Hyperdominance in Amazonian forest car- 3.x/full. [Accessed March 20, 2013]. bon cycling. Nat Commun 6:6857. [online] URL: http://www. McNicol I. 2014. The biomass and biodiversity of African sa- nature.com/doifinder/10.1038/ncomms7857%5Cnhttp://www. vanna woodlands: spatial patterns, environmental correlates pubmedcentral.nih.gov/articlerender.fcgi?artid=4423203&tool= and responses to land-use change. PhD thesis, University of pmcentrez&rendertype=abstract. Edinburgh. Fisher JI. et al. 2008. Clustered disturbances lead to bias in large- Miles L, Kapos V. 2008. Reducing greenhouse gas emissions scale estimates based on forest sample plots. Ecol Lett from deforestation and forest degradation: global land-use 11(6):554–63. [online] URL: http://www.ncbi.nlm.nih.gov/ implications. Science (New York, NY) 320(5882):1454–5. pubmed/18373680. [Accessed October 8, 2012]. Mittermeier RA. et al. 2003. Wilderness and biodiversity con- Fridley J. 2001. The influence of species diversity on ecosystem servation. Proc Natl Acad Sci USA 100(18):10309–13. [online] productivity: How, where, and why? Oikos 93:514–26. [on- URL: http://www.pubmedcentral.nih.gov/articlerender. line] URL: http://onlinelibrary.wiley.com/doi/10.1034/j. fcgi?artid=193557&tool=pmcentrez&rendertype=abstract. 1600-0706.2001.930318.x/full. [Accessed April 2, 2014]. Mugasha WA, Eid T, Bollandsas OM, Malimbwi RE, Chamshama Frost P. 1996. The ecology of Miombo woodlands. In: Campbell SAO, Zahabu E, Katani JZ. 2013. Allometric models for predic- B, Ed. The Miombo in transition: woodlands and welfare in tion of above- and belowground biomass of trees in the miombo Africa. Bogor: CIFOR. p 11–55. woodlands of Tanzania. Forest Ecol Manag 310:87–101. Halperin J et al. 2016. Model-based estimation of above-ground Murphy BP, Andersen AN, Parr CL. 2016. The underestimated biomass in the Miombo ecoregion of Zambia. For Ecosyst biodiversity of tropical grassy biomes. Philos Trans R Soc B 3(1):14. Biol Sci 371(1703):20150319. doi:10.1098/rstb.2015.0319. Hinsley A, Entwistle A, Pio DV. 2014. Does the long-term success NAFORMA. 2010. National forestry resources monitoring and of REDD+ also depend on biodiversity? Oryx 49(2):1–6. [on- assessment of Tanzania (NAFORMA). Dar es Salaam: Field line] URL: http://www.journals.cambridge.org/abstract_ Manual—Biophysical Survey. S0030605314000507. NAFORMA. 2011. NAFORMA species list. Dar es Salaam: Na- Huffman GJ. et al. 2007. The TRMM multisatellite precipitation tional Forestry Resources Assessment Monitoring and (NA- analysis (TMPA): quasi-global, multiyear, combined-sensor FORMA) of Tanzania. precipitation estimates at fine scales. J Hydrometeorol 8(1):38–55. [online] URL: http://journals.ametsoc.org/doi/ Nzunda EF, Griffiths ME, Lawes MJ. 2014. Resource allocation abs/10.1175/JHM560.1. [Accessed January 27, 2014]. and storage relative to resprouting ability in wind disturbed coastal forest trees. Evolut Ecol 28(4):735–49. [online] URL: Jew EKK et al. 2016. Miombo woodland under threat: conse- http://link.springer.com/10.1007/s10682-014-9698-7 [Ac- quences for tree diversity and carbon storage. For Ecol Manag cessed September 10, 2014]. 361:144–53. Oksanen J. 2013. Multivariate analysis of ecological communi- Khatun K, Corbera E, Ball S. 2016. Fire is REDD+: offsetting ties in R: Vegan tutorial. carbon through early burning activities in south-eastern Tanzania. Oryx 51(1):1–10. [online] URL: http://www. Phillips OL. et al. 2003. Efficient plot-based floristic assessment journals.cambridge.org/abstract_S0030605316000090. of tropical forests. J Trop Ecol 19(6):629–45. [online] URL: http://eprints.whiterose.ac.uk/237/. Lewis SL. et al. 2013. Above-ground biomass and structure of 260 African tropical forests. Philos Trans R Soc Lond Ser B Biol Phillips O. et al. 2009. RAINFOR: field measurement for plot Sci 368(1625):20120295. [online] URL: http://www.ncbi.nlm. establishment and remeasurement. nih.gov/pubmed/23878327. Platts PJ. et al. 2010. Can distribution models help refine Liang J. 2016. Positive biodiversity–productivity relationship inventory-based estimates of conservation priority? A case predominant in global forests. Science 354(6309):aaf8957. study in the Eastern Arc forests of Tanzania and Kenya. Divers 754 I. M. McNicol and others Distrib 16(4):628–42. [online] URL: http://doi.wiley.com/10. Scipal K. et al. 2010. The BIOMASS Mission—an ESA earth 1111/j.1472-4642.2010.00668.x [Accessed February 26, explorer candidate to measure the BIOMASS of the Earth’s 2014]. forests. In: International geoscience and remote sensing symposium (IGARSS). pp 52–5. Poorter L et al. 2015. Diversity enhances carbon storage in tropical forests. Glob Ecol Biogeogr 24(11):1314–28. Shirima DD et al. 2011. Carbon storage, structure and compo- sition of Miombo woodlands in Tanzania’s Eastern Arc R Core Team. 2014. R: a language and environment for statistical Mountains. Afr J Ecol 49(3):332–42. computing. [online] URL: http://www.r-project.org. Shirima DD et al. 2015. Relationships between tree species Re´jou-Me´chain M. et al. 2014. Local spatial structure of forest richness, evenness and aboveground carbon storage in mon- biomass and its consequences for remote sensing of carbon tane forests and Miombo woodlands of Tanzania. Basic Appl stocks. Biogeosci Discuss 11(4):5711–42. [online] URL: http:// Ecol 16(3):239–49. www.biogeosciences-discuss.net/11/5711/2014/ [Accessed April 28, 2014]. Slik JWF. et al. 2013. Large trees drive forest aboveground bio- mass variation in moist lowland forests across the tropics. Glob Ribeiro NS. et al. 2008. Aboveground biomass and leaf area in- Ecol Biogeogr 22(12):1261–71. [online] URL: http://doi.wiley. dex (LAI) mapping for Niassa Reserve, northern Mozambique. com/10.1111/geb.12092 [Accessed August 7, 2013]. J Geophys Res 113(G3):G02S02. [online] URL: http://www. agu.org/pubs/crossref/2008/2007JG000550.shtml [Accessed ter Steege H et al. 2013. Hyperdominance in the Amazonian tree March 14, 2013]. flora. Science 342(6156):1243092. Robinson C. et al. 2013. Impacts of spatial variability on Stohlgren TJ, Falkner MB, Schell LD. 1995. A modified-Whit- aboveground biomass estimation from L-band radar in a taker nested vegetation sampling method. Vegetatio temperate forest. Remote Sens 5(3):1001–23. [online] URL: 117(2):113–21. http://www.mdpi.com/2072-4292/5/3/1001/ [Accessed Au- Sullivan MJ et al. 2016. Diversity and Carbon storage across the gust 26, 2014]. tropical forest biome. Sci Rep 6(39102):1–12. doi:10.1038/ Ruiz-Benito P et al. 2014. Diversity increases carbon storage and srep39102. tree productivity in Spanish forests. Glob Ecol Biogeogr Venter O et al. 2009. Harnessing carbon payments to protect 23(3):311–22. doi:10.1111/geb.12126. biodiversity. Science (New York, NY) 326(5958):1368. Ruiz-Jaen MC, Potvin C. 2010. Tree diversity explains variation Vila` M. et al. 2007. Species richness and wood production: a in ecosystem function in a Neotropical forest in Panama. positive association in Mediterranean forests. Ecol Lett Biotropica 42(6):638–46. 10(3):241–50. [online] URL: http://www.ncbi.nlm.nih.gov/ Ryan CM. 2009. Carbon cycling, fire and phenology in a tropical pubmed/17305807 [Accessed April 2, 2014]. Savanna woodland in Nhambita, Mozambique. PhD Thesis, White F. 1983. The vegetation of Africa: a descriptive memoir to University of Edinburgh. University of Edinburgh. accompany the Unesco/AETFAT/UNSO vegetation map of Ryan CM, Williams M, Grace J. 2011. Above- and belowground Africa, Paris. carbon stocks in a Miombo woodland landscape of Mozam- Willcock S. et al. 2014. Quantifying and understanding carbon bique. Biotropica 43(4):423–32. storage and sequestration within the Eastern Arc Mountains Ryan CM. et al. 2012. Quantifying small-scale deforestation and of Tanzania, a tropical biodiversity hotspot. Carbon Balance forest degradation in African woodlands using radar imagery. Manag 9:2. [online] URL: http://www.pubmedcentral.nih. Global Change Biol 18(1):243–57. [online] URL: http://doi. gov/articlerender.fcgi?artid=4041645&tool=pmcentrez&rende wiley.com/10.1111/j.1365-2486.2011.02551.x [Accessed July rtype=abstract. 13, 2012]. Woollen E, Ryan CM, Williams M. 2012. Carbon stocks in an Ryan CM et al. 2016. Ecosystem services from Southern African African woodland landscape: spatial distributions and scales of woodlands and their future under global change. Philos Trans variation. Ecosystems 15(5):804–18. R Soc Lond Ser B Biol Sci 371:20150312. Scharlemann JPW et al. 2010. Securing tropical forest carbon: the contribution of protected areas to REDD. Oryx 44(March):352–7. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecosystems Springer Journals

Aboveground Carbon Storage and Its Links to Stand Structure, Tree Diversity and Floristic Composition in South-Eastern Tanzania

Free
15 pages
Loading next page...
 
/lp/springer_journal/aboveground-carbon-storage-and-its-links-to-stand-structure-tree-1D6r6gyBx2
Publisher
Springer US
Copyright
Copyright © 2017 by The Author(s)
Subject
Life Sciences; Ecology; Plant Sciences; Zoology; Environmental Management; Geoecology/Natural Processes; Hydrology/Water Resources
ISSN
1432-9840
eISSN
1435-0629
D.O.I.
10.1007/s10021-017-0180-6
Publisher site
See Article on Publisher Site

Abstract

Ecosystems (2018) 21: 740–754 DOI: 10.1007/s10021-017-0180-6 2017 The Author(s). This article is an open access publication Aboveground Carbon Storage and Its Links to Stand Structure, Tree Diversity and Floristic Composition in South-Eastern Tanzania 1 1 1 2,3 Iain M. McNicol, * Casey M. Ryan, Kyle G. Dexter, Stephen M. J. Ball, 1,4 and Mathew Williams School of Geosciences, University of Edinburgh, Crew Building, Alexander Crum Brown Road, Edinburgh EH9 3FF, Scotland, UK; 2 3 Mpingo Conservation and Development Initiative, Kilwa Masoko, United Republic of Tanzania; Present address: Farm Africa, Dar Es Salaam, United Republic of Tanzania; The National Centre for Earth Observation, Natural Environment Research Council, Swindon, UK ABSTRACT African savannas and dry forests represent a large, disproportionately contribute to AGC, with the lar- but poorly quantified store of biomass carbon and gest 3.7% of individuals containing half the carbon. biodiversity. Improving this information is hindered Tree species diversity and carbon stocks were posi- by a lack of recent forest inventories, which are tively related, implying a potential functional rela- necessary for calibrating earth observation data and tionship between the two, and a ‘win–win’ scenario for evaluating the relationship between carbon for conservation; however, lower biomass areas also stocks and tree diversity in the context of forest contain diverse species assemblages meaning that conservation (for example, REDD+). Here, we pre- carbon-oriented conservation may miss important sent new inventory data from south-eastern Tanza- areas of biodiversity. Despite these variations, we find nia, comprising more than 15,000 trees at 25 that total tree abundance and biomass is skewed to- locations located across a gradient of aboveground wards a few locally dominant species, with eight and woody carbon (AGC) stocks. We find that larger trees nine species (5.7% of the total) accounting for over half the total measured trees and carbon, respec- tively. This finding implies that carbon production in these areas is channelled through a small number of Received 17 February 2017; accepted 15 August 2017; relatively abundant species. Our results provide key published online 6 September 2017 insights into the structure and functioning of these heterogeneous ecosystems and indicate the need for Electronic supplementary material: The online version of this article (doi:10.1007/s10021-017-0180-6) contains supplementary material, novel strategies for future measurement and moni- which is available to authorized users. toring of carbon stocks and biodiversity, including Author contributions: CMR and MW developed the experimental design for plot establishment and produced the vegetation classification the use for larger plots to capture spatial variations in upon which plot location was based. Inventory data were collected by large tree density and AGC stocks, and to allow the IMM, CMR and partners at Mpingo Conservation and Development calibration of earth observation data. Initiative (MCDI) led by SMJB, with financial support from the Royal Norwegian Embassy in Tanzania. Richard Lamprey from Flora and Fauna International, a project partner of MCDI, provided estimates of canopy Key words: aboveground carbon storage; tree cover based on aerial photographs taken over the permanent sample diversity; Africa; miombo; large trees; biomass– plots. IMM, MW and CMR conceived the research questions. IMM col- lated and analysed the data and wrote the manuscript with input from biodiversity relationship; tree species composition; MW, CMR, SMJB and KGD. permanent plot; monitoring. *Corresponding author; e-mail: i.mcnicol@ed.ac.uk 740 Carbon and Tree Diversity in an African Landscape 741 Increasing human pressure linked to resource INTRODUCTION extraction is currently driving widespread, but Seasonally dry tropical forests and woodlands are uncertain losses of AGC, as well the localised the dominant vegetation cover in southern Africa, extinction of important tree species (Ahrends and extending over 4 million km across 10 countries others 2010; Ryan and others 2012; Jew and others (Mayaux and others 2004). Across their range, 2016). It is therefore important to quantify and variations in climate, soils and disturbance main- reduce uncertainty in our estimates of AGC storage, tain a structurally and floristically diverse mosaic of to better understand future losses, and to underpin habitats, covering a spectrum from open savanna carbon sequestration initiatives aimed at mitigating with a dominant grass layer and scattered trees, this loss. Plot-level estimates of AGC storage are through open canopy savanna woodland with an fundamental for calibrating and interpreting earth understory of grasses and shrubs, to denser wood- observation data, which can then be used to map lands and dry forest (White 1983). The most regional patterns in AGC (Avitabile and others extensive of these formations are the miombo 2016) and its changes over time (Ryan and others woodlands, distinguishable from surrounding veg- 2012). etation types by the dominance of the genera Measuring and managing ecosystems based on Brachystegia and Julbernardia (Fabaceae, Caesalpin- their carbon stocks, particularly under the umbrella ioideae) (Chidumayo 1997). The region as a whole of Reducing Emissions from Deforestation and is highly biodiverse and a priority for conservation Degradation (REDD+), may also benefit biodiver- (Mittermeier and others 2003; Brooks and others sity research and conservation (Scharlemann and 2006), with the miombo woodlands alone thought others 2010; Hinsley and others 2014; Ahrends and to harbour an estimated 8500 species of higher others 2011). It is therefore useful to quantify how plants, including more than 300 tree species (Frost tree diversity and floristic composition co-vary with 1996), many of which are endemic to the region. AGC storage (Hinsley and others 2014) to highlight The range of species supported by the ecosystem any important trade-offs and thus inform mutually helps to underpin the livelihoods of an estimated beneficial conservation schemes (Miles and Kapos 150 million rural and urban dwellers who rely 2008;Dı´az and others 2009; Venter and others heavily on the timber, food, medicine and con- 2009). Such information may also be useful in struction materials that the woodlands and forests elucidating a potential functional relationship be- provide (Ryan and others 2016). tween AGC storage and tree diversity, which could Yet despite their scale and importance for local have additional benefits for conservation if higher livelihoods, the ecology and functioning of these tree species diversity also results in higher AGC seasonally dry ecosystems remain poorly studied in storage. The majority of the current evidence base comparison with the more carbon dense moist for or against a biomass–biodiversity relationship tropical forests in South America (Fauset and others comes from the moist tropical forest biome (Sulli- 2015; Poorter and others 2015), and to a lesser ex- van and others 2016; Chisholm and others 2013), tent, those in Central Africa (Lewis and others and it is still unclear whether these patterns (or lack 2013). As a result, the miombo eco-region still rep- thereof) hold true in drier, mixed tree-grass sys- resents a potentially large, but poorly quantified tems. store of biomass carbon, biodiversity and species Despite the comparatively high diversity of the endemism (Platts and others 2010; Halperin and tropical forest biome, recent studies have found others 2016; Ryan and others 2016; Shirima and that a small number of relatively large trees and others 2011; Jew and others 2016). Forest inventory species contribute disproportionately to tree abun- plots with which to quantify these variables are few dance and AGC stocks in a variety of moist tropical in number and spatially uneven, typically favouring forest ecosystems (ter Steege and others 2013; higher biomass stands and protected areas (Chidu- Fauset and others 2015; Marshall and others 2012; mayo 2013; Ribeiro and others 2008; Marshall and Bastin and others 2015). The evidence base for others 2012; Willcock and others 2014; Ryan and similar patterns in the miombo eco-region is lim- others 2011; Chidumayo 2002). Thus, many ited by a paucity of detailed forest inventories important ecological questions remain poorly re- across a range of representative vegetation types solved, for example, around the magnitude and and ecosystems (Marshall and others 2012; Frost distribution of aboveground woody carbon stocks 1996; Shirima and others 2011). From a measure- (AGC) across these heterogeneous landscapes, and ment perspective, knowing which tree size classes how this relates to patterns in vegetation structure, contain most of the carbon and species diversity tree species diversity and composition. 742 I. M. McNicol and others may also help improve knowledge of how best to the coastal plains to the east up to 740 m m.a.s.l design effective data collection protocols which can along the steep escarpment running north to south be used to expand the current plot network (Mar- dissecting the centre of the district. Approximately shall and others 2012;Re´jou-Me´chain and others 85% of the local population is rural and dependent 2014; Bastin and others 2015). on natural resources for their livelihoods (Khatun In this paper, we aim improve the knowledge of and others 2016). From October 2010–October ecosystem structure and function across these 2011, permanent sample plots were established at heterogeneous landscapes using data collected from 25 locations, originally stratified by three major a new network of 25 forest inventory plots in vegetation types delineated via a supervised land south-eastern Tanzania, which spans a gradient of cover classification, based on Landsat 5 data and woody biomass and different vegetation types. 300 in situ visual assessments of land cover, to Specifically, we explore (1) how patterns in AGC ensure that potential variations in AGC stocks had stocks are related to differences in tree size and been suitably captured (Figure 1). The vegetation number, (2) to tree species diversity within plots types for the original stratification included grass (a-diversity) and (3) to tree species composition. dominated ‘savannas’ with sparse tree cover, sa- vanna woodland (tree-grass mix) and dense woodland and forest (closed tree canopy with no METHODS grass cover), with the number of plots measured Study Area and Sampling Strategy proportional to the areal extent of each vegetation type. Tree canopy cover was estimated by outlining The study area is located in Kilwa District in the the crowns of individual trees identified using Lindi Region of south-eastern Tanzania (Figure 1). aerial photographs collected over the plots in The estimated mean annual precipitation is October 2010 (Figure 1). Pragmatism played a role 821 ± 350 mm (±SD), with a gradient between in site location, with plots located randomly along the east (wetter) and west (drier) (Tropical Rainfall the road and track network (Figure 1); however, a Measurement Mission, 3B43 product; Huffman and 1-km buffer from tracks was enforced to reduce the others 2007). Altitude varies from sea level along Figure 1. Location of our field plots and associated aboveground woody carbon stock (AGC) and canopy cover estimates. Sub-panel A shows the location of Tanzania, and the extent of the miombo woodlands—the dominant vegetation type in our study region, with sub-panel B showing the location of our study region. C Location of our field plots, and the initial land cover classification used for plot location. D The distribution of plot (1 ha) AGC stocks and canopy cover estimates. Carbon and Tree Diversity in an African Landscape 743 Figure 2. A Cumulative percentage of AGC stocks contributed by different tree size classes within plots of similar AGC and canopy cover; B the average number of trees within each size class. Each data point represents the average contribution of plots within each group. likelihood of intense human disturbance. For Kolmogorov–Smirnov tests were used to test sampling, we utilised a 1-ha (100 9 100 m) sized whether the distribution of plot-level AGC in each permanent sample plot in which all trees with a size class was statistically different between plots of diameter of at least 5 cm were recorded, tagged and broadly similar AGC and structure (tree density and spatially located. These 1-ha plots, upon which canopy cover), under the null hypothesis that the most of the analyses in this study are based, were distributions are similar and that variations in AGC nested centrally within a larger 9-ha storage reflect differences in tree density. (300 9 300 m) plot in which only trees larger than To assess species composition and diversity, we 40 cm were recorded. Tree diameter was measured used the species names or genus where known. at 1.3 m height above the ground, and if the tree Where this was not possible, the local name was forked below 1.3 m, each stem was measured and used instead. In some cases, the use of local names counted as one individual. We recorded the local may result in tree species diversity being overesti- name of each measured tree, and where possible, mated if multiple names are used for a single spe- identified each by their scientific name using col- cies; however, the more likely scenario is that lected voucher specimens and published reference diversity will be underestimated as the same local guides (Coates-Palgrave and Moll 2002). Where name is often used for several species (for example, this was not possible, species were identified using based on local usage), with some species also likely a range of local and national species lists (NA- to be indistinguishable without fertile material FORMA 2011). leading to some species being conflated (Ahrends and others 2011). To minimise errors due to the Data Analysis former, we used the same botanists for all plots to ensure species identification was consistent across Aboveground carbon stocks (AGC) were calculated plots. Controlling for the latter is more difficult. using an allometric model developed in the same However, on average, trees identified only by local administrative region (Lindi model: Mugasha and name contributed no more than five of the species others 2013), with biomass assumed to be 47% measured in each plot and thus we consider the carbon. To address our first question about how likelihood that our diversity measures are subject to variations in AGC stocks are related to differences meaningful bias to be small. A small numbers of in stand structure, specifically size and number, individuals that were not identified to any taxo- trees were binned into 5-cm size classes and the nomic level (0.07% of total inventory) were ex- proportional contribution of each size class to the cluded from the analysis. total measured AGC in each plot was calculated. 744 I. M. McNicol and others Tree species diversity was calculated using three To examine how our results (that is, tree diver- measures: species richness, Fisher’s alpha and rar- sity and AGC estimates) would have differed had efied richness. For rarefied richness, we used Mao- we sampled progressively smaller plots instead of Tao individual-based rarefaction analysis. When the 1-ha plots, we simulated single sub-plots of comparing tree diversity and AGC, diversity is re- varying size (0.1, 0.25 and 0.5 ha) at random garded as the independent variable under the locations within each of the 25 9 1 ha plots, with assumption that tree diversity has a deterministic the sub-sampling analysis repeated 1000 times to effect on AGC at the plot level (due to niche ensure the full range of possible subsets was complementarity and selection effects), as opposed achieved. For each subplot, we calculated the tree -1 to if the axis were reversed, which would assume species richness and AGC density (tC ha ) and environmental/disturbance controls on diversity, compared these as a percentage of the corre- which we believe are more likely to occur at larger sponding estimates from the 1-ha plot. For each scales than our field plots (Chisholm and others iteration, we totalled the number of species across 2013; Woollen and others 2012). Multiple models the network to show how sampling smaller plots were fitted to each data set using a variety of across the entire network would have impacted our functional forms based on ecological theory, estimates of landscape diversity. including a linear relationship ðy ¼ ax þ b), satu- All data analyses were performed using the R ration ðy ¼ ax=ðÞ b þ t , quadraticðÞ y ¼ ax þ bx þ c statistical software version 3.0.2 (R Core Team bx 2014, http://cran.r-project.org) and the ‘vegan’ and a parabolic ricker curve y ¼ axe . Model package (version 2.0-10; Oksanen 2013). selection was based on minimising the Akaike information criterion (AICc), corrected for small sample sizes, and the residual sum of squared dif- RESULTS ferences. Patterns in Aboveground Woody Carbon Diversity measures were taken for all trees Stocks and Stand Structure (>5 cm) in each 1-ha plot, then again for small trees (5–15 cm), medium sized trees (15–40 cm) In total, we surveyed 13,098 trees (>5 cm) across and large canopy dominants (>40 cm) separately, the 25 one-ha plots, including 10,694 small trees with the aim of understanding where most of the (5–15 cm), 2139 medium sized trees (15–40 cm) tree diversity occurs in these systems. For the and 265 large trees (>40 cm). The surrounding 9- analysis of large tree diversity (>40 cm), data from ha plots contained an additional 2069 large trees, the 9-ha plots were included to allow a suit- highlighting the importance of larger plots for able number of trees for analysis. Differences in adequate statistical analyses of large trees. AGC species composition between plots (b-diversity) -1 stocks in the 1-ha plots ranged from 2 tC ha in were calculated using the Bray–Curtis Index of -1 an area of open grassland savanna to 54 tC ha in Species Dissimilarity. Overall compositional pat- an area of dense forest (Figure 1), with an overall terns were visualised using non-metric multidi- -1 landscape average of 24 ± 16 tC ha (±indicates mensional scaling, which was performed using the standard deviation throughout). ‘metaMDS’ function. Permutational multivariate This gradient in AGC stocks is associated with analysis of variance (PerMANOVA) was used to test clear changes in both tree density (72–1511 tree- -1 whether there were significant differences in tree sha ; Spearman’s rho, R = 0.95, P < 0.001) and species composition between groups of plots (An- tree canopy cover, with areas of <10% cover— derson 2001). The analysis was repeated separately broadly consistent with the FAO definition of for small, medium and large trees to test whether ‘other wooded lands’ (FAO 2001)—storing -1 composition differed among size classes. Prior to <10 tC ha (n = 7), with plots in more open ca- analysis, the raw species abundance data were nopy savanna ‘woodlands’ (10–45%) storing 15– -1 square root transformed and site standardised to 35 tC ha (n = 12), and plots in more closed ca- -1 account for the number of trees sampled at each nopy ‘forests’ (>50%) containing >40 tC ha site and to reduce the influence of the most com- (n = 6). Large trees contributed around one-third mon species (Barlow and others 2007). We used (32 ± 18%) of plot AGC, despite comprising only ANOVA and Tukey’s HSD tests to look for signifi- 2.6 ± 2.2% of the trees in each plot. Overall, half cant differences in tree structure and diversity be- of the total measured biomass (across the 1-ha tween groups of plots after testing the data for plots) was stored in the 484 largest trees, which normality using Shapiro–Wilk tests. comprised 3.7% of the total trees measured. Carbon and Tree Diversity in an African Landscape 745 Table 1. Top 5 Dominant Species Within Plots of Broadly Similar AGC Stocks Ranked by Their Contribution to the Total Carbon Stock and Total Tree Abundance -1) -1 -1 -1 Rank Low AGC (0–10 tC ha Low to moderate AGC (15–25 tC ha ) Moderate to high AGC (25–40 tC ha ) High AGC (>45 tC ha ) -1 AGC (tC ha ) 1 Diospyros quiloensis Dalbergia melanoxylon Julbernardia globiflora Hymenocardia ulmoides 2 Sclerocarya birrea Pseudolachnostylis maprouneifolia Brachystegia spiciformis Hymenaea verrucosa 3 Combretum apiculatum Julbernardia globiflora Combretum apiculatum Rytigynia sp. 4 Dalbergia melanoxylon Combretum apiculatum Burkea africana Pteleopsis myrtifolia 5 Burkea africana Brachystegia spiciformis Diplorhynchus condylocarpon Euphorbia nyikae % of total 53.6 44.3 60.9 49.7 -1 Stocking density (trees ha ) 1 Combretum apiculatum Diplorhynchus condylocarpon Diplorhynchus condylocarpon Hymenocardia ulmoides 2 Spirostachys africana Combretum apiculatum Combretum apiculatum Suregada zanzibariensis 3 Acacia nilotica Dalbergia melanoxylon Pseudolachnostylis maprouneifolia Euphorbia nyikae 4 Burkea africana Pseudolachnostylis maprouneifolia Hymenocardia ulmoides Uvaria lucida 5 Bauhinia petersiana Bridelia scleroneura Julbernardia globiflora Strychos spinosa % of total 62.8 55.4 64.7 52.0 n plots 7 7 8 3 Species richness 15 (6) 26 (8) 32 (7) 42 (4) Fisher’s a 4.2 (2.3) 6.4 (2.3) 7.5 (1.7) 8.2 (0.7) Total species richness 56 74 95 87 a b b c Bray–Curtis 0.77 (0.11) 0.69 (0.14) 0.52 (0.10) 0.61 (0.10) Number of unique species 9 10 26 32 Plots with a moderate and high AGC density are further separated to better highlight changes in tree species dominance over the gradient, particularly in our three highest AGC plots (> 60% tree canopy cover) which are marked out as floristically distinct from other high AGC plots (Figure 4). Additional information includes the mean tree species richness and Fisher’s a in each plot (±SD), the total number of species recorded in each group, as well as the number that are unique to each group, and the average species dissimilarity between plots [Bray–Curtis Index (±SD)]. The letters in superscript next to the Bray–Curtis index indicate the results of the PerMANOVA which tested whether trees species composition significantly differed between groups of plots. 746 I. M. McNicol and others Table 2. Diversity Indices for Group of Plots Separated by Broad Size Class Size class Small trees (5–15 cm DBH) Medium trees (15–40 cm DBH) Large trees (40 cm + DBH) Low AGC Species richness 14 (7) 6 (3) 7 (4)* Fisher’s a 3.2 (2.0) 3.8 (3.7) 3.4 (2.0) a a a Bray–Curtis Index 0.77 (0.11) 0.89 (0.12) 0.77 (0.12) Moderate AGC Species richness 22 (6) 15 (5) 15 (5)* Fisher’s a 5.5 (1.7) 6.0 (2.3) 4.9 (1.6) b b b Bray–Curtis Index 0.66 (0.14) 0.67 (0.13) 0.64 (0.17) High AGC Species richness 28 (9) 19 (4) 17 (4)* Fisher’s a 6.2 (1.1) 5.6 (1.7) 5.0 (1.1) c b b Bray–Curtis Index 0.73 (0.14) 0.74 (0.15) 0.74 (0.16) As in Table 1 information includes the average species richness and Fisher’s a (±SD) for different size classes within each plot. The Bray–Curtis Index is used to highlight difference in floristic composition within plots. The letters in superscript indicate the results of the PerMANOVA which tested whether the composition of small, medium and large trees significantly varied between groups of plots. Includes the measured trees from the 9-ha plot meaning that comparisons of large tree species richness are only valid between groups, and not between size classes due the larger sample area for large trees compared to medium and smaller trees. Figure 3. Relationships between tree species richness and aboveground woody carbon stocks. Ordinary least squares (OLS) regression models are fitted to the data; A tree species richness (y = 1.15–6.67, r = 0.63, P = <0.001) and B rarefied richness (y = 1.95–5.12, r = 0.22, P = 0.01). The distribution of carbon stocks among tree size between our moderate and high AGC density plots classes differed significantly between our low AGC (P = 0.51), despite a clear trend towards greater -1 density plots (<10 tC ha ) and those with a tree size (that is, >80 cm DBH) at the upper end of moderate and high AGC density (Kolmogorov– the gradient, where these very large trees had a Smirnov; P = <0.001 in both cases). In the low disproportionate contribution to plot AGC (10%) -1 AGC density, typically grassland savanna plots, the relative to their abundance (1 ± 1ha ) (Fig- majority of AGC (42%) was contributed by the ure 2). smallest diameter classes (5–15 cm) (Figure 2), whereas in moderate density savanna ‘woodlands’ Patterns in Tree Species Composition and higher AGC density ‘forest’ plots, the propor- and Diversity tion of AGC stored in small trees was relatively low We identified 158 morphospecies across the (15%), despite the greater number of trees in 25 9 1 ha plots by their local species name, of these areas. There were no significant differences in which 91 were fully identified to species level the distribution of AGC among different size classes Carbon and Tree Diversity in an African Landscape 747 Figure 4. A Plot-pair differences in tree species composition with differences in plot-level AGC stocks; B NMDS ordi- nation based on the Bray–Curtis Index which is used to uncover the main compositional patterns across the gradient in AGC storage. (57%) and a further 16 to genus (10%), with 32 biomass grassland savannas and savanna wood- taxonomic families present. In the surrounding 9- lands (39; 21%), compared to the ‘forests’ (25; 6%) ha plots (>40 cm DBH trees only), 79 morphos- where they were few in number, but large. This pecies were identified, including 26 not found in pattern was also true for potentially nodulating the 1-ha plots, with 54 (68%) of these identified to legumes (Caroline Lehmann and others unpubl. species level, and 3 (4%) to genus, with a further data.) which were almost absent in high AGC three families represented. In both 1- and 9-ha areas, yet gradually more common as AGC stocks plots, the identified taxa contributed 96% of the decreased, comprising 40% of trees in low density total measured trees and AGC across all sites. The plots. data presented in the following sections are from A small number of species were both abundant the 1-ha plots unless otherwise stated. and widespread, with 8 species collectively con- Tree species richness ranged from 9 to 45 per plot tributing over 50% of the trees measured, includ- with both richness and Fisher’s a significantly ing Diplorhynchus condylocarpon (15.9% of all trees; higher in the moderate and high AGC density plots n plots = 17), Combretum apiculatum (10.6%; compared to the lowest density plots (ANOVA + n = 21), and to a lesser extent, Hymenocardia ul- Tukey HSD, P < 0.01) (Table 1). The results were moides (9.9%; n = 8) and Pseudolachnostylis the same when comparing small, medium and maprouneifolia (3.6%; n =16). A similar level of large trees separately (Table 2). Tree species rich- dominance was observed when assessing species ness exhibited a positive linear relationship with contributions to the total carbon stock, with just 9 AGC storage (r = 0.63, P < 0.001) (Figure 3). The species, including the four aforementioned species, significant trend was maintained when controlling containing over half (52.5%) of the total AGC. The for tree density (rarefied richness), though the remaining biomass dominant species were Jul- relationship was markedly weaker (r = 0.22, bernardia globiflora (15.4% of total measured bio- P = 0.01) (Figure 3), indicating that differences in mass), Brachystegia spiciformis (7%), Burkea Africana tree density partly drive this relationship. (4.5%), Pteleopsis mytifolia and the priority conser- Euphorbiaceae was the dominant family across vation and timber species, Dalbergia melanoxylon, the plot network, comprising 39% of the total with the remainder either commonly used for measured AGC and 17% of trees, followed equally charcoal (P. myrtifolia), or occasionally harvested by Combretaceae and Fabaceae (each 21% of for timber. A similar level of species dominance was AGC and 11% of trees), and Apocynaceae (12; observed within each of the broad vegetation types, 17%). Familial dominance differed among vegeta- with approximately 5 species contributing over half tion types with trees in the family Euphorbiaceae of the AGC stocks and trees (Table 1). more common in areas with an AGC density The large majority of species were considerably -1 greater than 40 tC ha (39; 24%), with those in less abundant, with 49 species (31% of total) con- Fabaceae proportionally more dominant in lower tributing fewer than 50 individuals. Many of the 748 I. M. McNicol and others recorded species were restricted to particular habi- with more disturbed systems. The results highlight tats, with nine restricted to the low AGC plots, 36 the obvious importance of maintaining a low DBH to plots with a moderate AGC density, with 32 threshold (that is, 5 cm) in lower biomass stands in species only found in the three highest AGC ‘forest’ order to capture and quantify the majority of AGC plots (Figure 4). Species turnover (b-diversity) stocks. among plots was therefore relatively high, with In the more carbon dense savanna woodlands some areas of similar AGC found to contain entirely and dry forest plots, a greater proportion of AGC different species assemblages (Figure 4). The lowest was contained in larger trees, with the relative AGC plots were the most heterogeneous (Table 1), proportion contained in different size classes sta- as shown by the NMDS ordination plot tistically similar between plots in moderate (10– -1 -1 (stress = 0.12, n dimensions = 3) and were floris- 35 tC ha ) and high AGC (>40 tC ha ) stands. tically distinct to both the moderate and high AGC We therefore conclude that the variations in AGC plots, both when considering all tree together stocks between these areas are due to differences in (>5 cm) (PerMANOVA, P < 0.001; Figure 4; Ta- tree abundance in each size class, although there is ble 1) and small, medium and large trees separately some evidence to suggest that these differences (Table 2). may also reflect the greater density of very large Despite the wider range of AGC storage, we ob- trees (‡80 cm) in forests, which typically numbered served a greater compositional similarity among the only one per hectare in the most carbon dense -1 moderate density ‘woodland’ plots (15–40 tC ha ), ‘forest’ plots (>50% canopy cover), yet con- which tend to be dominated terms of AGC contri- tributed on average 8% of the measured AGC. bution by two of the defining miombo woodland These very large trees were comparatively rare in species—J. globiflora and B. spiciformis—and in the low density, typically grassland savanna plots; number by D. condylocarpon and C. apiculatum (Ta- however, where a very large tree was present on a ble 1). At the upper end of the gradient, species plot (>94.9 cm, Diospyros quiloensis), its contribu- characteristic of wet miombo woodland and coastal tion to the total measured AGC was considerable forest was common, including Suregada zanz- (50%). ibariensis and Hymenaea verrucosa. This shift in tree The concentration of biomass in a small number composition is reflected in the NMDS plot with the of trees has been previously observed in other three highest AGC plots—two of which were located moist forest ecosystems (Bastin and others 2015; at relatively high elevations along an escarpment Fauset and others 2015; Slik and others 2013)and (Figure 1)—exhibiting clear differences in compo- has clear implications for the development of ra- sition (Figure 4), both when considering all trees pid, low-cost forest monitoring protocols. In more -1 together, and when comparing trees in different size wooded areas (that is, >10 tC ha /% canopy classes (PerMANOVA; P < 0.001; Tables 1, 2). cover), large trees—that is, those larger than 40 cm—comprised approximately 40% of the biomass measured in each plot, with half the plot DISCUSSION AGC contained in the top 4.9% of trees (range Links Between Vegetation Structure and 2.7–9%; n trees = 9–64; minimum DBH = 24– Aboveground Carbon Storage 46 cm). These results are consistent with the re- sults of Bastin and others (2015)who detected a Our landscape-level estimates of aboveground similar concentration (that is, 50%) of plot bio- -1 carbon (AGC) stocks (24 ± 16 tC ha ) are similar mass in a similar proportion of trees (5% of total) to those recorded using similar approaches in across Central African moist forests. Similar results Mozambique by Ryan et al. (2011) were also found across an identical plot network -1 (21 ± 11 tC ha ) and Woollen and others (2012) in the miombo woodlands of Mozambique (Ryan -1 (21 ± 10 tC ha ), but lower than the regional 2009; Ryan and others 2011), where approxi- -1 average (28.7 ± 19.1 tC ha ) (Ryan and others mately 50% of plot AGC was contained in trees 2016) which includes many plots from protected larger than 40 cm DBH, suggesting this is a com- areas which are unlikely to be representative of the mon feature of miombo-dominated woodlands. wider miombo eco-region. Our lowest AGC plots, Our results contrast with those of Marshall and defined as areas with a tree canopy cover (%) and others (2012) who found that in the moist forests -1 AGC stock (tC ha ) of less than 10, were charac- of the Eastern Arc Mountains, trees larger than terised by a lower tree density, with the majority of 40 cm stored a much higher proportion (75–80%) trees (80%), and thus AGC (42%) contained in of plot AGC. smallest size classes (5–15 cm DBH), as is common Carbon and Tree Diversity in an African Landscape 749 The tendency towards greater tree size in plots at their maximum tree height (Nzunda and others the upper end of the gradient may be due to their 2014) and shade tolerance. In contrast, the noted location at moderate to high elevations (Marshall compositional similarities among the moderate and others 2012), suggesting a possible topo- density plots mean it is unlikely that differences in graphic, and/or edaphic influence on AGC storage composition are driving the within-vegetation type (Woollen and others 2012). These plots were also heterogeneity in AGC storage. Our results therefore more remote from human populations (Figure 1), suggest that compositional/functional differences meaning that historically lower levels of distur- may be more important in explaining the variation bance (human and ‘natural’) in these areas may between, rather than within vegetation types. have allowed larger trees to persist and AGC to Despite this diversity in tree species composition, accrue over longer periods. In the moderate AGC we find that total tree abundance and biomass is -1 density plots (10–35 tC ha ), we found no trees skewed strongly towards a relatively few locally larger than 75 cm DBH, yet in the surrounding 9- dominant species (Shirima and others 2011), with ha plots, several trees (n = 12) surpassed this limit 8 species (5.7% of the total) accounting for over (max. 112 cm), suggesting that in some cases, even half the measured trees and 9 species for greater 1-ha plots are unable to fully capture the stem size than 50% of biomass. A larger degree of biomass- distribution of woodlands (Anderson and others and stem-‘hyperdominance’ is found in the more 2009). This in turn may lead to high sampling er- diverse rainforests of both Amazonia (Fauset and rors when scaling AGC estimates across the land- others 2015; ter Steege and others 2013), and to a scape (Fisher and others 2008;Re´jou-Me´chain and lesser extent, Central Africa (Bastin and others others 2014), or remote sensing data of coarser 2015), although these results are derived from resolutions than the plots, such as the European much larger regional plot networks. In our study Space Agency’s Biomass mission, which will oper- area, the relatively large proportion of biomass lo- ate at a resolution of 4 ha (Scipal and others 2010). cated in such a small number of trees (90% is This mismatch again highlights the importance of contained in 38 species) suggests that most biomass sampling on a sufficiently large scale, either productivity in these seasonally dry ecosystems is through sampling many smaller plots, or a few also channelled through a relatively small number larger plots, to account for the inherent patchiness of tree species. The additional finding that greater of these ecosystems and presence of rare large trees. than 50% of the biomass is contained in moderate to high value timber suitable trees also highlights Relationship Between AGC Storage, Tree the future sensitivity of woody carbon stocks, and potentially productivity, in this area to logging and/ Species Diversity and Composition or charcoal production (Ahrends and others 2010). The inclusion of biodiversity as a co-benefit in From a conservation standpoint, our finding that carbon sequestration projects necessitates an more carbon dense areas also harbour the greatest assessment on how the two co-vary to assess tree species diversity suggests a ‘win–win’ scenario potential trade-offs, or co-benefits of conservation for forest conservation projects operating under the initiatives. From an ecological perspective, exam- umbrella of REDD+. Among the recorded species ining these linkages along with the extent to which were a number that are endemic to the remaining certain species contribute to carbon storage in these fragments coastal forest in the region, including H. systems, will help with efforts to reveal a more verrucosa and Uvaria kirkii, which is recorded as deterministic relationship between these two vari- ‘Near Threatened’ on the IUCN red list. Lower ables, and likely resilience of these ecosystems to biomass stands, particularly the miombo (Jul- future changes in land use (Hinsley and others bernardia—Brachystegia)-dominated ‘woodlands’, 2014). also contained a relatively diverse assemblage of We find clear differences in tree species compo- trees, including a number of high value timber sition along our AGC gradient, with the lowest species, such as Pterocarpus angolensis which is AGC stands and our three highest biomass plots commercially extinct in many parts of Tanzania marked out as being floristically distinct from the (Jew and others 2016) and classified as ‘Near spatially extensive, and moderate AGC density Threatened’, and the priority conservation species miombo-dominated ‘woodlands’. The composi- Dalbergia melanoxylon. A large number of species tional patterns suggest that the associated varia- were also found to be constrained to either mod- tions in AGC storage along the gradient may be erate or high density stands resulting in localised partially explained by differing functional traits patterns of species endemism. As such, the ‘win– between the dominant species in each area, such as win’ scenario indicated by our results does not 750 I. M. McNicol and others mean that comparatively low biomass areas should tions over whether tree diversity does indeed have be excluded from conservation efforts, as these a mechanistic effect on AGC storage and produc- areas may retain many locally and biologically tivity in these systems, which is important for important species, particularly in the understory understanding how changes in biodiversity will (that is, woody plats < 5 cm), and herbaceous affect these important ecosystem functions (Liang layers, as well as in faunal communities (Murphy and others 2016). It is also unclear whether more and others 2016), none of which were sampled in diverse tree communities help to create greater this study. diversity across multiple trophic levels, and whe- The preservation of biodiversity may have ther these communities also increase the ecosystem additional benefits if higher tree species diversity services provided to humans such as timber re- also results in higher AGC storage. Our finding of a sources and medicinal products (Maestre and oth- positive relationship between diversity and AGC ers 2012), both of which are important areas of storage is consistent with other observational future research. studies from both the miombo eco-region (Shir- ima and others 2015) and other forests globally Potential Implications for Future Tree (Ruiz-Jaen and Potvin 2010; Ruiz-Benito and Measurement and Monitoring others 2014;Vila and others 2007; Maestre and The need to acquire data on AGC stocks has taken others 2012; Liang and others 2016; Poorter and on added significance due to the rise in carbon others 2015). This positive relationship is consis- sequestration initiatives such as REDD+. The col- tent with theories of (1) niche complementarity, lection of species data also needs to be included in where a higher tree species richness leads to a any future measurement campaign to allow co- more functionally diverse community and thus variation between AGC and biodiversity to be ex- greater resource capture and biomass production; plored in the context of forest conservation (Venter and (2) selection effects, which posit that in al- and 2009; Liang and others 2016; Ahrends and ready dense stands there is a greater chance that 2011). Expanding the current network of perma- oneorafewhighlyproductivespecies arepresent nent inventory plots is a necessity, and a stan- (Fridley 2001). The absence of any clear saturation dardised methodology based on existing data sets is in the relationship at higher biomass levels, which crucial to rapidly facilitate the establishment of new would be suggestive of species redundancy or plots in the region and aid cross-plot comparisons. competitive exclusion, indicates that relatively To date, no studies have presented a clear view on dense patches of vegetation are still capable of the most appropriate and efficient strategy (that is, efficiently utilising available resources to allow sample size, plot size, appropriate DBH threshold) many species and high AGC stocks to coexist, for accurately measuring carbon stocks and/or suggesting that some form of complementarity or biodiversity in savanna woodlands (that is, Bar- facilitation is operating in these areas. Yet despite aloto and others 2013), a fact which is evidenced by the statistical significance of the relationships, the wide variety of sampling methodologies used to there was considerable variability in tree diversity for tree measurement (Ribeiro and others 2008; between plots, particularly after accounting for NAFORMA 2010; Chidumayo 2013; Ryan and differences in tree density. Recent studies from others 2011; Willcock and others 2014). The moist tropical forests indicate that diversity con- RAINFOR manual has provided some consistency trols on AGC storage operate at much smaller based on data collected in Amazonian forests scales than the ones observed here (0.1 ha) (Phillips and others 2009; Phillips and others 2003); (Chisholm and others 2013; Poorter and others however, there is no equivalent methodology for 2015; Sullivan and others 2016), which may ex- the dry tropics which are very different in terms of plain the lack of explanatory power. An alterna- their tree structure, diversity and composition tive explanation is that the greater diversity of tree (Fauset and others 2015; ter Steege and others species at higher AGC densities is the result of 2013). The results here provide some insights in more heterogeneous environmental conditions how sampling could be tailored in future to suit the within these areas, leading to greater species aims of a given project and its available resources. turnover related to habitat specialisation in certain For example, we show that in more wooded patches. High AGC may also occur in areas that -1 areas (>10 tC ha , >10% canopy cover), where have fewer major disturbances, allowing species stem size distribution is broadly consistent across less adapted to disturbance to persist. sites, measuring only those trees larger than 10 cm A full assessment of the biomass–diversity rela- DBH would have captured on average 93% of the tionship over larger scales will help answer ques- Carbon and Tree Diversity in an African Landscape 751 total AGC in each plot, yet would have required aloto and others 2013). Based on our data set, it is measuring 40% of the trees, or skipping on average unclear which of these sampling strategies (‘‘few -1 approximately 600 trees ha in denser woodlands large’’ vs. ‘‘many small’’ plots) is more appropriate -1 and dry forests (>40 tC ha ) and approximately for accurately and cost effectively capturing tree -1 275 trees ha in more open canopy savanna species diversity and composition in these areas. -1 woodlands (10–35 tC ha ). Raising the threshold Such information will be important for facilitating to 15 cm would still have captured 86% of the total conservation planning and implementation and AGC stocks in only 20% of the trees. We suggest will likely require the intensive (sub)-sampling of that such an approach would be ideal for con- very large plots to properly address this question ducting rapid inventories of AGC, such as for the (Baraloto and others 2013). calibration of earth observation data. The issue of plot size has additional importance Measuring for biodiversity and species composi- for measuring biomass, with smaller plots more tion would have very different requirements with likely to either overestimate, or completely miss the 50% of the species sampled here likely to be missed presence of rare, large trees, thus creating signifi- when measuring trees larger than 10 cm. These cant small scale variations in AGC stocks (Re´jou- species are likely to be among the rarest; therefore, Me´chain and others 2014; Fisher and others 2008; sampling at a higher DBH threshold will have little Chave and others 2004). Indeed, we find that even value when assessing the biodiversity or conser- the 0.5-ha plots produce highly variable AGC -1 vation value of these areas. Our results also suggest densities (tC ha ) relative to the corresponding that for a given site, the use of smaller inventory 1 ha values (5–95th percentile; 40–120%), tending plots (that is, <0.5 ha) (Willcock and others 2014; towards underestimation (median = 90%) (Chave NAFORMA 2010; Shirima and others 2015), which and others 2003). These sampling errors were are ideally suited for rapid sampling and often used exacerbated when using progressively smaller sub- for species measurement across the tropics plots, with 0.25 ha (25–150%) and 0.1 ha (14– (Stohlgren and others 1995; Baraloto and others 200%) plots generating an ever-larger range of 2013; Phillips and others 2003), are potentially possible AGC values relative to the 1-ha estimates. more sensitive to species clustering and/or likely to The 0.1-ha plots also produced anomalously high -1 exclude rare tree species (Baraloto and others values above 100 tC ha where a large tree(s) is 2013). For example, in the 9-ha plots, we find 26 present. For this reason, we would caution against species not in the 1-ha plots, despite measuring the use of very small plots (that is, <0.25 ha) for only those trees larger than 40 cm in these areas, measuring biomass as they can create large uncer- suggesting that even 1-ha plots fail to fully capture tainties on AGC stocks for a given site. However, if the species diversity at certain sites. We explored replicated in sufficient number, smaller plots may this potential issue further by sub-sampling the 1- still be suitable for estimating the average AGC ha plots which showed that the use of smaller plots density across the landscape, although such esti- would have captured on average 36 ± 13% mates may be less precise (Chave and others 2004). (0.1 ha), 53 ± 14% (0.25 ha) and 71 ± 14% This issue of plot size has clear relevance when (0.5 ha) of the plot-level tree species richness. considering the suitability of the plots for the cali- Hence, smaller plots clearly sample a smaller pro- bration of remotely sensed data; particularly radar portion of tree species for a given site than the 1-ha (for example, ALOS PALSAR) and LiDAR sensors, plots (Phillips and others 2003). However, sampling which in future will be the primary method for 0.5-ha plots instead of the 1-ha plots at each site upscaling ground based AGC estimates to the would still have captured a large majority landscape scale. Smaller plots (for example, (80 ± 2%) of the tree species found across the <0.25 ha) tend to be unsuitable for this purpose entire 1-ha network in only half the sample area, due to the aforementioned scaling issues, but also highlighting that the use of smaller plots may be their larger relative geo-location errors which may more efficient for gathering large-scale floristic be of similar size to the field plot (Ryan and 2012). data. The issue of many potentially rare tree species As a result, AGC stocks measured in larger plots are being missed in the smaller plots could be avoided if often found to exhibit a much stronger relationship sampling a larger number of these across the wider with the remotely sensed observation (Carreiras landscape; however, the physical and financial and others 2013;Re´jou-Me´chain and others 2014; challenges associated with repeat plot establish- McNicol 2014; Robinson and others 2013; Mauya ment and accessing typically remote areas may and others 2015). The mismatch in spatial scale outweigh the costs associated with establishing a between many of the current field inventory plots smaller number of well stratified larger plots (Bar- (Shirima and others 2011; Willcock and others 752 I. M. McNicol and others 2014; Ryan and others 2011) and the larger pixels License (http://creativecommons.org/licenses/by/ of future sensors such as the European Space 4.0/), which permits unrestricted use, distribution, Agency’s Biomass mission (4 ha) (Scipal and others and reproduction in any medium, provided you 2010) also has the potential to introduce consid- give appropriate credit to the original author(s) and erable errors when scaling plot even our 1 ha AGC the source, provide a link to the Creative Commons values to the size of the radar pixel (Re´jou-Me´chain license, and indicate if changes were made. and others 2014). The use of higher DBH thresh- olds would allow for larger areas (that is, >1ha) to REFERENCES be sampled in a more time and cost-efficient Ahrends A et al. 2010. Predictable waves of sequential forest manner, as was achieved in this study with the 9- degradation and biodiversity loss spreading from an African ha plots which were typically sampled in two-third city. Proc Natl Acad Sci USA 107(33):14556–61. of the time taken to sample the 1-ha plots; how- Ahrends A et al. 2011. Conservation and the botanist effect. Biol ever, this would clearly be at the detriment of Conserv 144(1):131–40. doi:10.1016/j.biocon.2010.08.008. biodiversity assessment. As shown here, the sam- Anderson M. 2001. A new method for non-parametric multi- variate analysis of variance. Aust Ecol 26:32–46. [online] URL: pling of large plots (that is, >1 ha) also has the http://onlinelibrary.wiley.com/doi/10.1111/j.1442-9993.2001. additional benefit of capturing of suitable of num- 01070.pp.x/full. Accessed December 13, 2013. ber larger trees, which will be useful for the anal- Anderson L, Malhi Y, Ladle R. 2009. Influence of landscape ysis of large tree mortality. heterogeneity on spatial patterns of wood productivity, wood The development of a standardised field protocol specific density and above ground biomass in Amazonia. that appropriately incorporates measurements of Biogeosciences 6:1883–1902. [online] URL: http://www. biogeosciences.net/6/1883/2009/bg-6-1883-2009.html. Ac- tree species diversity and aboveground carbon cessed August 5, 2013. stocks, but is also suitable for the calibration of Avitabile V et al. 2016. An integrated pan-tropical biomass map earth observation data, is urgently needed in order using multiple reference datasets. Glob Change Biol to ensure the best use of time and resources. For 22(4):1406–20. doi:10.1111/gcb.13139. this reason, we would suggest that larger sample Baraloto C et al. 2013. Rapid simultaneous estimation of plots (that is, ‡1 ha) should be favoured where aboveground biomass and tree diversity across Neotropical possible to capture potentially important variations forests: a comparison of field inventory methods. Biotropica 45(3):288–98. doi:10.1111/btp.12006. in large tree densities, and thus AGC stocks, Barlow J. et al. 2007. Quantifying the biodiversity value of whereas at the same time, allowing the plots to be tropical primary, secondary, and plantation forests. Proc Natl used as a calibration points for earth observation Acad Sci USA 104(47):18555–60. [online] URL: http://www. data, and facilitating cross-project comparisons pubmedcentral.nih.gov/articlerender.fcgi?artid=2141815&tool= (that is, RAINFOR). These plots may form part of pmcentrez&rendertype=abstract. nested sampling strategy to account for the differ- Bastin J.-F. et al. 2015. Seeing Central African forests through their ent data requirements, including the use of smaller largest trees. Sci Rep 5:1–8. [online] URL: http://www.nature. com/srep/2015/150817/srep13156/full/srep13156.html. plots (for example, 0.5 ha) for the sampling of tree Brooks TM et al. 2006. Global biodiversity conservation priori- species diversity, and potentially even smaller plots ties. Sciences 313(August):58–61. for sampling the understory and herbaceous layer, Carreiras J, Melo J, Vasconcelos M. 2013. Estimating the above- which was not sampled at all in this study, yet is a ground biomass in Miombo Savanna Woodlands (Mozam- major store of diversity in these ecosystems (Mur- bique, East Africa) using L-band synthetic aperture radar data. phy et al. 2016). Remote Sens 5(4):1524–48. [online] URL: http://www.mdpi. com/2072-4292/5/4/1524/. [Accessed July 22, 2014]. ACKNOWLEDGEMENTS Chave J et al. 2003. Spatial and temporal variation of biomass in a tropical forest: results from a large census plot in Panama. J We thank Nicholas Berry for his work on the pro- Ecol 91(2):240–52. ject. We also thank Deogratias Ndossi and Juvenal Chave J et al. 2004. Error propagation and scaling for tropical Pantaleo for co-managing the field data collection. forest biomass estimates. Philos Trans R Soc Lond Ser B Biol Sci 359(1443):409–20. IMM was supported as part of his PhD by the UK Chidumayo E. 2002. Changes in miombo woodland structure Natural Environment Research Council (NERC) under different land tenure and use systems in central Zam- and MCDI under their REDD Pilot Project funded bia. J Biogeogr 29:1619–26. [online] URL: http://online by the Royal Norwegian Embassy in Tanzania. library.wiley.com/doi/10.1046/j.1365-2699.2002.00794.x/full. [Accessed December 12, 2012]. OPEN ACCESS Chidumayo EN. 1997. Miombo ecology and management: an introduction. London: IT Publications in Association with the This article is distributed under the terms of the Stockholm Environment Institute. Creative Commons Attribution 4.0 International Carbon and Tree Diversity in an African Landscape 753 Chidumayo EN. 2013. Forest degradation and recovery in a Maestre FT. et al. 2012. Plant species richness and ecosystem Miombo woodland landscape in Zambia: 22 years of obser- multifunctionality in global drylands. Science 335(6065):214– vations on permanent sample plots. For Ecol Manag 291:154– 8.http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid 61. [online] URL: http://linkinghub.elsevier.com/retrieve/pii/ =3558739&tool=pmcentrez&rendertype=abstract. [Accessed S0378112712007074. [Accessed August 7, 2013]. March 19, 2014]. Chisholm RA. et al. 2013. Scale-dependent relationships be- Marshall AR. et al. 2012. Measuring and modelling above- tween tree species richness and ecosystem function in forests. ground carbon and tree allometry along a tropical elevation J Ecol 101(5):1214–24. [online] URL: http://doi.wiley.com/ gradient. Biol Conserv 154:20–33. http://linkinghub.elsevier. 10.1111/1365-2745.12132. [Accessed November 6, 2013]. com/retrieve/pii/S0006320712001607. [Accessed November 5, 2012]. Coates-Palgrave M. 2002. In: Moll E, Ed. Keith Coates-Palgrave trees of Southern Africa 3rd ed. Cape Town: Random House Mauya E. et al. 2015. Effects of field plot size on prediction Struik Publishers. accuracy of aboveground biomass in airborne laser scanning- assisted inventories in tropical rain forests of Tanzania. Carbon Dı´az S, Hector A, Wardle DA. 2009. Biodiversity in forest carbon Balance Manag 10(1):10. [online] URL: http://www.scopus. sequestration initiatives: not just a side benefit. Curr Opin com/inward/record.url?eid=2-s2.0-84928886552&partnerID= Environ Sustain 1(1):55–60. [online] URL: http://linkinghub. tZOtx3y1%5Cnhttp://www.scopus.com/inward/record.url?eid elsevier.com/retrieve/pii/S1877343509000177. [Accessed =2-s2.0-84928886552&partnerID=40&md5=83db6c72f8a246e November 1, 2012]. 4454e8e8e5c48a9d5. FAO. 2001. Global forest resources assessment 2000. [online] Mayaux P. et al. 2004. A new land cover map of Africa for the URL: https://www.soils.org/publications/sssaj/abstracts/73/6/ year 2000. J Biogeogr 31:861–77. [online] URL: http:// onlinelibrary.wiley.com/doi/10.1111/j.1365-2699.2004.0107 Fauset S. et al. 2015. Hyperdominance in Amazonian forest car- 3.x/full. [Accessed March 20, 2013]. bon cycling. Nat Commun 6:6857. [online] URL: http://www. McNicol I. 2014. The biomass and biodiversity of African sa- nature.com/doifinder/10.1038/ncomms7857%5Cnhttp://www. vanna woodlands: spatial patterns, environmental correlates pubmedcentral.nih.gov/articlerender.fcgi?artid=4423203&tool= and responses to land-use change. PhD thesis, University of pmcentrez&rendertype=abstract. Edinburgh. Fisher JI. et al. 2008. Clustered disturbances lead to bias in large- Miles L, Kapos V. 2008. Reducing greenhouse gas emissions scale estimates based on forest sample plots. Ecol Lett from deforestation and forest degradation: global land-use 11(6):554–63. [online] URL: http://www.ncbi.nlm.nih.gov/ implications. Science (New York, NY) 320(5882):1454–5. pubmed/18373680. [Accessed October 8, 2012]. Mittermeier RA. et al. 2003. Wilderness and biodiversity con- Fridley J. 2001. The influence of species diversity on ecosystem servation. Proc Natl Acad Sci USA 100(18):10309–13. [online] productivity: How, where, and why? Oikos 93:514–26. [on- URL: http://www.pubmedcentral.nih.gov/articlerender. line] URL: http://onlinelibrary.wiley.com/doi/10.1034/j. fcgi?artid=193557&tool=pmcentrez&rendertype=abstract. 1600-0706.2001.930318.x/full. [Accessed April 2, 2014]. Mugasha WA, Eid T, Bollandsas OM, Malimbwi RE, Chamshama Frost P. 1996. The ecology of Miombo woodlands. In: Campbell SAO, Zahabu E, Katani JZ. 2013. Allometric models for predic- B, Ed. The Miombo in transition: woodlands and welfare in tion of above- and belowground biomass of trees in the miombo Africa. Bogor: CIFOR. p 11–55. woodlands of Tanzania. Forest Ecol Manag 310:87–101. Halperin J et al. 2016. Model-based estimation of above-ground Murphy BP, Andersen AN, Parr CL. 2016. The underestimated biomass in the Miombo ecoregion of Zambia. For Ecosyst biodiversity of tropical grassy biomes. Philos Trans R Soc B 3(1):14. Biol Sci 371(1703):20150319. doi:10.1098/rstb.2015.0319. Hinsley A, Entwistle A, Pio DV. 2014. Does the long-term success NAFORMA. 2010. National forestry resources monitoring and of REDD+ also depend on biodiversity? Oryx 49(2):1–6. [on- assessment of Tanzania (NAFORMA). Dar es Salaam: Field line] URL: http://www.journals.cambridge.org/abstract_ Manual—Biophysical Survey. S0030605314000507. NAFORMA. 2011. NAFORMA species list. Dar es Salaam: Na- Huffman GJ. et al. 2007. The TRMM multisatellite precipitation tional Forestry Resources Assessment Monitoring and (NA- analysis (TMPA): quasi-global, multiyear, combined-sensor FORMA) of Tanzania. precipitation estimates at fine scales. J Hydrometeorol 8(1):38–55. [online] URL: http://journals.ametsoc.org/doi/ Nzunda EF, Griffiths ME, Lawes MJ. 2014. Resource allocation abs/10.1175/JHM560.1. [Accessed January 27, 2014]. and storage relative to resprouting ability in wind disturbed coastal forest trees. Evolut Ecol 28(4):735–49. [online] URL: Jew EKK et al. 2016. Miombo woodland under threat: conse- http://link.springer.com/10.1007/s10682-014-9698-7 [Ac- quences for tree diversity and carbon storage. For Ecol Manag cessed September 10, 2014]. 361:144–53. Oksanen J. 2013. Multivariate analysis of ecological communi- Khatun K, Corbera E, Ball S. 2016. Fire is REDD+: offsetting ties in R: Vegan tutorial. carbon through early burning activities in south-eastern Tanzania. Oryx 51(1):1–10. [online] URL: http://www. Phillips OL. et al. 2003. Efficient plot-based floristic assessment journals.cambridge.org/abstract_S0030605316000090. of tropical forests. J Trop Ecol 19(6):629–45. [online] URL: http://eprints.whiterose.ac.uk/237/. Lewis SL. et al. 2013. Above-ground biomass and structure of 260 African tropical forests. Philos Trans R Soc Lond Ser B Biol Phillips O. et al. 2009. RAINFOR: field measurement for plot Sci 368(1625):20120295. [online] URL: http://www.ncbi.nlm. establishment and remeasurement. nih.gov/pubmed/23878327. Platts PJ. et al. 2010. Can distribution models help refine Liang J. 2016. Positive biodiversity–productivity relationship inventory-based estimates of conservation priority? A case predominant in global forests. Science 354(6309):aaf8957. study in the Eastern Arc forests of Tanzania and Kenya. Divers 754 I. M. McNicol and others Distrib 16(4):628–42. [online] URL: http://doi.wiley.com/10. Scipal K. et al. 2010. The BIOMASS Mission—an ESA earth 1111/j.1472-4642.2010.00668.x [Accessed February 26, explorer candidate to measure the BIOMASS of the Earth’s 2014]. forests. In: International geoscience and remote sensing symposium (IGARSS). pp 52–5. Poorter L et al. 2015. Diversity enhances carbon storage in tropical forests. Glob Ecol Biogeogr 24(11):1314–28. Shirima DD et al. 2011. Carbon storage, structure and compo- sition of Miombo woodlands in Tanzania’s Eastern Arc R Core Team. 2014. R: a language and environment for statistical Mountains. Afr J Ecol 49(3):332–42. computing. [online] URL: http://www.r-project.org. Shirima DD et al. 2015. Relationships between tree species Re´jou-Me´chain M. et al. 2014. Local spatial structure of forest richness, evenness and aboveground carbon storage in mon- biomass and its consequences for remote sensing of carbon tane forests and Miombo woodlands of Tanzania. Basic Appl stocks. Biogeosci Discuss 11(4):5711–42. [online] URL: http:// Ecol 16(3):239–49. www.biogeosciences-discuss.net/11/5711/2014/ [Accessed April 28, 2014]. Slik JWF. et al. 2013. Large trees drive forest aboveground bio- mass variation in moist lowland forests across the tropics. Glob Ribeiro NS. et al. 2008. Aboveground biomass and leaf area in- Ecol Biogeogr 22(12):1261–71. [online] URL: http://doi.wiley. dex (LAI) mapping for Niassa Reserve, northern Mozambique. com/10.1111/geb.12092 [Accessed August 7, 2013]. J Geophys Res 113(G3):G02S02. [online] URL: http://www. agu.org/pubs/crossref/2008/2007JG000550.shtml [Accessed ter Steege H et al. 2013. Hyperdominance in the Amazonian tree March 14, 2013]. flora. Science 342(6156):1243092. Robinson C. et al. 2013. Impacts of spatial variability on Stohlgren TJ, Falkner MB, Schell LD. 1995. A modified-Whit- aboveground biomass estimation from L-band radar in a taker nested vegetation sampling method. Vegetatio temperate forest. Remote Sens 5(3):1001–23. [online] URL: 117(2):113–21. http://www.mdpi.com/2072-4292/5/3/1001/ [Accessed Au- Sullivan MJ et al. 2016. Diversity and Carbon storage across the gust 26, 2014]. tropical forest biome. Sci Rep 6(39102):1–12. doi:10.1038/ Ruiz-Benito P et al. 2014. Diversity increases carbon storage and srep39102. tree productivity in Spanish forests. Glob Ecol Biogeogr Venter O et al. 2009. Harnessing carbon payments to protect 23(3):311–22. doi:10.1111/geb.12126. biodiversity. Science (New York, NY) 326(5958):1368. Ruiz-Jaen MC, Potvin C. 2010. Tree diversity explains variation Vila` M. et al. 2007. Species richness and wood production: a in ecosystem function in a Neotropical forest in Panama. positive association in Mediterranean forests. Ecol Lett Biotropica 42(6):638–46. 10(3):241–50. [online] URL: http://www.ncbi.nlm.nih.gov/ Ryan CM. 2009. Carbon cycling, fire and phenology in a tropical pubmed/17305807 [Accessed April 2, 2014]. Savanna woodland in Nhambita, Mozambique. PhD Thesis, White F. 1983. The vegetation of Africa: a descriptive memoir to University of Edinburgh. University of Edinburgh. accompany the Unesco/AETFAT/UNSO vegetation map of Ryan CM, Williams M, Grace J. 2011. Above- and belowground Africa, Paris. carbon stocks in a Miombo woodland landscape of Mozam- Willcock S. et al. 2014. Quantifying and understanding carbon bique. Biotropica 43(4):423–32. storage and sequestration within the Eastern Arc Mountains Ryan CM. et al. 2012. Quantifying small-scale deforestation and of Tanzania, a tropical biodiversity hotspot. Carbon Balance forest degradation in African woodlands using radar imagery. Manag 9:2. [online] URL: http://www.pubmedcentral.nih. Global Change Biol 18(1):243–57. [online] URL: http://doi. gov/articlerender.fcgi?artid=4041645&tool=pmcentrez&rende wiley.com/10.1111/j.1365-2486.2011.02551.x [Accessed July rtype=abstract. 13, 2012]. Woollen E, Ryan CM, Williams M. 2012. Carbon stocks in an Ryan CM et al. 2016. Ecosystem services from Southern African African woodland landscape: spatial distributions and scales of woodlands and their future under global change. Philos Trans variation. Ecosystems 15(5):804–18. R Soc Lond Ser B Biol Sci 371:20150312. Scharlemann JPW et al. 2010. Securing tropical forest carbon: the contribution of protected areas to REDD. Oryx 44(March):352–7.

Journal

EcosystemsSpringer Journals

Published: Sep 6, 2017

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off