How have carbon stocks in central and southern Africa’s miombo woodlands changed over the last 50 years? A systematic map of the evidence

Literature searches

Article screening and study inclusion criteria

Before screening began, a Kappa analysis [74] was done to compare reviewer agreement on a random sample of 100 titles and abstracts. This was to measure inter-rater agreement in applying the inclusion criteria to studies (where all reviewers must achieve a score of > 0.6 in agreement of 100 articles). Randolph’s free marginal multi-rater Kappa [75] was used to score agreement for 2 rounds of Kappa analysis between 9 and 8 reviewers, respectively. Both rounds had over 0.6 agreement, and reported 0.67 (9 reviewers) and 0.64 (8 reviewers) free marginal Kappa scores. During each round, disagreements in applying the inclusion criteria were discussed and these notes were saved in a screening guidance sheet that was shared with all reviewers. Once the second round of Kappa test was completed and discrepancies discussed, screening of articles began.

The review team screened articles through two main stages: title/abstract and full-text levels. During the first stage, seven reviewers screened 11,565 titles and abstracts simultaneously with the online software tool Abstrackr.² Abstrackr allows multiple reviewers to screen titles and abstracts independently and collates results. After the first round was completed, a smaller review team (five people) re-screened abstracts with Abstrackr. This was partly because of over-cautiousness by the larger review team in reviewing abstracts, but also because of the large number of records remaining and the limited resources of the team. The reviewers followed the same inclusion criteria but excluded articles that were highly likely to be ineligible for the map. These results were then imported into a Microsoft Excel file for full-text review.

Full text screening was the second stage of screening. A small review team (four people) screened 1014 full-texts online according to the inclusion criteria. Records which passed this stage were downloaded and included in the map for further eligibility screening in study validity assessment and data coding stages.

Study validity

Study coding strategy

Study inclusion

We screened 11,565 records by title and abstract (after removing duplicates, Fig. 2). Many records were excluded because they did not show the presence of the defined miombo tree species, or failed to directly link land use change to impacts upon plant biomass or above or below-ground carbon. Title/abstract and full-text screening excluded 11,284 records, leaving 281 records (221 peer-reviewed articles and 60 additional sources) that were appraised for eligibility. During this process, 227 records were excluded because of low validity (see Additional file 4 for excluded studies). What particularly limited the number of included records was poor documentation of information on the Population—failure to demonstrate that studies were in miombo areas as defined by the species outlined in the protocol—and Outcomes—poor or missing information on how land use changes impacted carbon or biomass in defined measurements. In total, 54 records were found to have usable data for the map, comprising 48 quantitative and eight qualitative records (two records are double-counted as they reported both quantitative and qualitative data).

Descriptive statistics

Of the 54 records included, three records were dissertations, three records were reports, and 48 records were journal articles, covering the years between 1962–2016 (Fig. 3). All records contained only one unique study. Two studies were assessed as having high validity; 51 studies were assessed as medium; and one study was assessed as having low validity. These factors are discussed further below.

Qualitative studies

Eight qualitative studies directly related miombo land use practices to impacts on woody biomass and carbon. The study locations were as follows: Malawi (n = 2), Zambia (n = 2), Tanzania (n = 2), Zimbabwe (n = 1) and one global study. Most studies were journal articles (n = 7) and one study was a dissertation. Three studies used mixed methods (quantitative and qualitative approaches including interviews, and measured transects and plots), while five studies employed qualitative methods only. In the assessment of study validity, one had high study validity, and six studies were scored as medium and one low.

The majority of qualitative studies (n = 7) focused on exposures around the use of local forest resources for livelihoods. Only one study focused on agricultural practices in forest areas. One study made direct links to forest loss and carbon, and two studies attempted to measure forest resource extraction. The other studies (n = 5) used narrative evidence and description to describe land and woodland practices. All studies linked land and woodland uses (i.e. exposures) to impacts on woodland or biomass (i.e. outcomes) but were extremely site specific and therefore difficult to generalise.

All eight studies documented how the miombo woodlands were important for local livelihoods, and thus advocated local level forest management. For example, one study argued for the use of local knowledge in woodland management [77], and almost all studies (n = 7) called for decentralisation of management to local levels. Mbwambo et al. [78] most clearly demonstrated the critical issues affecting woodland governance at different scales, by showing that local forestry reserves (e.g. State) had as much tree loss as open (e.g. communal) woodlands in Tanzania. This reflects the finding by Roe et al. [79] that all miombo countries have some form of community forestry management scheme captured in forest policies, but that the failure to deploy these policies has not been fully understood. However, in terms of broader governance aspects of miombo woodland management, the evidence base was limited. No studies discussed the implications for adopting regional forestry management models (i.e. SADC and COMESA) or global models such as REDD+, and the necessary land use or forest policy changes needed [80].

Quantitative studies

Studies were from 87 locations in Malawi, Mozambique, Tanzania, Zambia and Zimbabwe (Fig. 4). Five of these sites lie outside the main miombo woodland range as it is typically known but were included because the flora of the sites include species of Brachystegia and Julbernardia as defined by the inclusion criteria [67]. The sites extend over a large geographical region and differ in some biophysical conditions. Notably, no sites were located in Angola, northern Mozambique or the DRC, and more of the studies occurred in dry (n = 31) compared to moist miombo (n = 18) (one article had plots in both dry and moist miombo) (see Fig. 5).

Research design

Data in the relevant studies (Table 5) were collected using diverse sampling designs at sites with heterogeneous geographical, ecological and land use conditions. Few studies (n = 17) used permanent sampling designs; most used chronosequence study designs in old and re-growth plots (n = 494) (Fig. 6).

Data type	Method description	No of measurements	Studies included
Basal area	m2 ha−1 at 1.3 m above-ground	181	Lees [81]; Boaler and Sciwale [12]; Endean [11]; Strang [54]; Chidumayo [56, 82–85]; Kwibisa [86]; Chamshama et al. [87]; Banda et al. [88]; Williams et al. [89]; Ryan and Williams [90]; Mbwambo et al. [78]; Chomba et al. [91]; Kashindye et al. [92]; Sawe et al. [93]; Treue et al. [94]
Biomass	Mg ha−1	187	Guy [95]; Stromgaard [96]; Chidumayo [97, 98]; Grundy et al. [99]; Smith [100]; Sambane [101]; Kutsch et al. [102]; Kashindye et al. [92]; Ando et al. [103]; Ryan et al. [104]; Sawe et al. [93]; Hofstad and Araya [105]; Jew et al. [17]
Above-ground carbon	Mg C ha−1	105	Stromgaard [96]; Guy [95]; Chidumayo [97]; Grundy et al. [99]; Smith [100]; Sambane [101]; Kutsch et al. [102]; Kashindye et al. [92]; Ando et al. [103]; Sawe et al. [93]; Hofstad and Araya [105]; Shirima et al. [106]; Jew et al. [17]
Tree AGB & BGB	kg tree−1	34	Ryan et al. [25]; Chidumayo [107]; Mugasha et al. [108]
Soil organic matter	% per soil volume	361	Araki [109]; Kwibisa [86]; Chidumayo and Kwibisa [110]; Mapanda et al. [111, 112]; Muposhi et al. [113]
Soil organic carbon	Mg C ha−1	108	Jenkinson et al. [114]; Chidumayo and Kwibisa [110]; Walker and Desanker [115]; Anonymous [116]; Williams et al. [89]; Rossi et al. [117]; Kutsch et al. [102]; Ryan et al. [25]; Woollen et al. [66]; Chidumayo [98]; Mapanda et al. [112]; Ando et al. [103, 118]; Muposhi et al. [113]; Shelukindo et al. [119]; Winowiecki et al. [120]
Policy and land use change			McGregor [77]; Abbot [121]; Hogan et al. [122]; Grace et al. [123]; Culas [124]; Davies et al.[125]; Mbwamdo et al. [78]; Chomba et al. [91]

Studies in italic reflect those listed more than once

AGB & BGB, above- and below-ground biomass respectively

The majority of the data reported on above-ground biomass, basal area and soil organic matter (Table 5 and Fig. 7). In terms of woodland cover, the majority of study plots were in re-growth (n = 298) compared to old-growth (n = 213) woodland. Re-growth study sites were further classified into either fallow re-growth (n = 163) (i.e., woodland regenerating at a site/plot following abandonment of crop cultivation); or woodland re-growth (n = 135) (woodland regenerating at a site/plot following clearing, e.g. for charcoal production or tsetse fly control, without experiencing crop cultivation). In terms of conservation area status, the data were distributed as follows: 5 in GMAs, 15 in NPs, 72 in FRs and 419 in OAs (see Fig. 8).

Woodland management practices (harvesting, fire regime and herbivory) were not explicitly described for the majority of sites, which would constrain any analysis of effects of these practices on carbon stocks and sequestration (Fig. 9).

Carbon stocks in old-growth miombo woodlands

Average above-ground carbon stock in old-growth miombo woodland across all studies was 33.9 ± 1.3 Mg C ha⁻¹, with a wide range (1.3–95.7 Mg ha⁻¹). However, variation between years (i.e. different data sets) and between rainfall zones was large, and the wide range suggests that sampling occurred randomly across the region. Given this large variation, future assessments would need to differentiate by rainfall/aridity zones, history of land use and human interventions.

A number of studies have asserted that the level of disturbance in miombo woodlands increases with decreasing distance from human settlements and road infrastructure [17, 126–128], and that miombo woodlands are losing their potential to hold C stocks, rather becoming net emitters of CO₂ into the atmosphere [102]. A global study found human pressure (measured through human population density, land transformation and electrical power infrastructure) on terrestrial ecosystems, including protected areas, to have increased by 64% since the 1990s [129].

However, there is scant evidence in the available literature of miombo biomass variation with distance from settlements. It is also known that biomass depends on the aridity levels [130], drainage and effective rooting depth [51], as well as rainfall [131]. Data for these factors in the studies we mapped are not available in a uniform way, and this would severely limit future meta analysis testing this hypothesis. A further limitation of studying biomass across sites and years from the a posteriori compilation of available data sets comes from the different sampling designs and spatial confounding factors between studies. Further, the studies do not represent any systematic sampling across the whole miombo region and its environmental and climatic gradients. Some countries have been completely omitted from the map because of the lack of data. Long-term observational data in the same stands under controlled intervention regimes of different intensity would be needed to address specifically the question of stand biomass degradation over time.

Below-ground carbon

Miombo woodland soils contain considerable amounts of organic matter, which forms a large below-ground pool of carbon [115]. Two soil qualities have an important bearing on soil carbon. First, total soil organic matter (SOM) in miombo soils is often concentrated in the top 30 cm of the soil, with contents ranging from 1 to 2% [89, 132, 133]. Second, soil bulk density (SBD), which ranges from 1.2 to 1.4 in miombo soils [89, 110]. Bulk density was not reported in all the studies reviewed [115].

Few studies have estimated below-ground plant biomass in the miombo. In general the root:shoot (R/S) ratio is used to estimate below-ground biomass and this ratio varies with tree size in miombo woodlands [25, 107, 108]. Since young and small trees tend to have higher below-ground to above ground biomass ratio than old and large ones, it is important to use different R/S ratios for re-growth miombo dominated by small trees and for old-growth miombo dominated by large trees. Appropriate ratios given in [107] of 0.77 and 0.54 for re-growth and old-growth miombo woodland, respectively, would need to be applied to raw data to estimate below-ground biomass and carbon effectively across studies.

Loss of miombo woodland area

Miombo woodland area losses have been observed to varying degrees of intensity at landscape level, largely driven by land clearing for agriculture and wood extraction for energy, both processes often going hand in hand [21]. Governments have been working to address tree loss in general as noted in the promulgation of new forest policies, acts and regulations in the wake of the developments after the Rio Earth Summit (UNCED, Rio de Janeiro, June 1992) [134, 135]. While regulations exist, the situation is not likely to change soon without sufficient budgetary support, enforced regulations, the participation of communities, and awareness and capacity building.

Carbon stocks along a conservation gradient

Wegmann et al. [136] found that vegetation cover change is faster outside than inside protected areas, and Geldmann et al. [129] observed that human impact on protected areas is correlated to their location and IUCN management category. Cumming et al. [137] observed that woodland degradation in protected areas in southern Africa is largely caused by elephants and fire. Studies included in the map report varying figures for C stocks along a conservation gradient in miombo woodlands; future meta analysis could test whether the data are in line with findings by Banda et al. [88] in western Tanzania. Indeed, given the increasing number of reports of human encroachments into national parks located within miombo woodlands (see [121, 126, 138]), human and elephant pressure in protected areas may have made C stocks in conservation areas just as vulnerable to loss as in unprotected areas. This would present considerable challenges to efforts to conserve C stocks in the miombo region of central and southern Africa, e.g. for initiatives to reduce greenhouse gas emissions and increase carbon sequestration that are under way in parts of the region [139–141]. These initiatives include REDD+ (Reduced Emissions from Deforestation and forest Degradation plus enhancement of forest stocks), which creates incentives for developing countries to protect, better manage and wisely use their forest resources and thereby contribute to the global fight against climate change while providing development benefits to communities on the ground. Countries like Tanzania, Mozambique and Zambia have gone through REDD+ readiness and planning phases, and are moving towards the development and adoption of national REDD+ strategies [142–144]. It is important therefore that such initiatives take cognisance of the current dynamics in C stocks in miombo woodlands during their development phase. To some extent, data and information generated in the planning phases have been incorporated in country reports to UNFCCC and in some cases used to bolster regional climate change negotiations (see [145]).

Carbon stock recovery in re-growth woodland

The studies report sufficient data on trends in C stock recovery in re-growth miombo that could be analysed to assess possible differences in recovery trajectories on post-clearing re-growth and fallow re-growth sites. In shifting cultivation systems most fallow re-growth is converted back to cropland before the age of 30 years [96, 146–148]; reported declines in C stocks after 20 years would need further research.

Studies report C stock increments in young re-growth (< 15 years old) and for intermediate ages (15–40 years), but few data points for older trees. Chidumayo [84] found that stem density in re-growth miombo of 49 years was 13–25% of that in re-growth of 11 years and the decrease in stem density was attributed to competition in the intervening period. These changes often indicate a socio-economic portrait of the landscape born out of the interplay between land use measures and decisions by people and governments [149, 150] and in the case of southern Africa, land contestations of 1990s and encroachments in response to Structural Adjustment Programmes (SAPs) of the same period have been behind forest loss in general [151]. New policy and management strategies are therefore required to reverse this negative trend if new initiatives, such as REDD+, are to achieve their objectives.

Effects of fire management on carbon stocks and dynamics

Wild fires affect carbon stocks and dynamics in miombo woodlands [25, 98] and there were studies reporting this effect. Previous findings indicate that production and C accumulation in miombo woodland is higher under fire protection and early burning than under late burning [11, 56, 84, 152]. However, complete fire protection is risky because accidental fires occurring under fire protection can find more fuel to burn and undo the gains made in carbon stocks accumulated over long periods. Chidumayo [56] reported that in a 36-year old fire protected re-growth plot, an accidental fire reduced carbon stocks to 13.9 Mg ha⁻¹ compared to a stock of 72.5 Mg ha⁻¹ that had accumulated under early burning. However, after 36 years of the experiment, both these carbon stocks were higher than that of 2.3 Mg ha⁻¹ under late burning.

Limitations of the map

While our search strategy aimed to minimise publication bias by consulting a wide range of peer-reviewed and grey literature sources, we did not have the time nor resources to review studies in languages other than English. As such, we are likely to have missed relevant studies in Portuguese and other languages from the southern African region. The lack of systematic biomass assessments across the regions’ climatic and environmental gradients, while accounting carefully for human intervention, is another important limiting factor.

Most of the data were chronosequence studies that used space for time and may have spatial confounding factors. In addition to conservation area status, data reported included age and type of re-growth (i.e. fallow re-growth or woodland re-growth) and cultivation period that provides information indicative of temporal trends in carbon stocks and sequestration.

Studies reported above- and below-ground carbon stocks in miombo woodlands in a wide variety of ways. Most studies estimated above-ground wood biomass using allometric models based on tree diameter measurements at breast height. These models ranged from non-linear power functions to linear logarithmic models and therefore the estimates may have contained biases due to the allometric model used. This would be a serious challenge for future anlaysis of reported data. Meta analysis of the data in these studies would require that all the data be converted to carbon units at the same spatial scale (e.g. Mg C ha⁻¹) using an appropriate equation [68].

Implications for research in the miombo and future systematic review topics

The systematic map reveals an inadequate evidence base to address the four questions originally posed in our protocol. An insufficient number of robust studies was found that meet our inclusion criteria from across the miombo region. There is a clear knowledge gap for Angola, northern Mozambique and the Democratic Republic of Congo, at least from the wide sources we consulted, and while we acknowledge that this may due in large part to language bias in the mainstream academic literature, it points to a need to focus on this region in future primary research. There was more literature on dry miombo sites compared with moist sites, and the latter may also be a useful focus for future research. The relative dearth of qualitative studies highlights a potential source of future work to develop policy based on community engagement in the miombo region.

Much further research is needed to understand C stocks and their trends in miombo old-growth woodland, because pooling the data is not sufficiently informative for localised management suggestions, given the variability (i.e., rainfall, aridity, soil depth and stand age) of miombo across the region. Understanding miombo biomass stocks, losses and gains and carbon sequestration in light of the variability across the regional gradients is crucial for future studies. It would require better assessment of both above- and belowground biomass across the region in a systematic way, accounting for spatial, climatic and land use differences.