Framework for a savanna burning emissions abatement methodology applicable to fire-prone miombo woodlands in southern Africa

Background to the ‘savanna burning’ emissions abatement methodology

The essential premise underlying Australia’s iterative savanna burning methods (CoA (Commonwealth of Australia) 2013, 2015, 2018) involves undertaking strategic prescribed burning of cured ground fuels (grass, litter, coarse woody debris – CWD, heavy woody fuels – HWF) through the EDS to reduce the risk of extensive wildfires under relatively severe LDS fire-weather conditions (high temperatures, low humidity, gusty winds), thereby significantly reducing GHG emissions and achieving a variety of other benefits (e.g. biodiversity conservation, pasture management, asset protection). The intent is not to eliminate the incidence of LDS fires but, rather, to allow safe deployment of relatively intense fires within a matrix of EDS fire-treated fuel-breaks for required purposes (e.g. habitat management, woody plant control). The method is founded on a long regional tradition of Aboriginal customary fire and land management practice (Yibarbuk et al. 2001; Russell-Smith et al. 2003; Garde et al. 2009; Vigilante et al. 2009).

Following Seiler and Crutzen (1980) and IPCC (Intergovernmental Panel on Climate Change) (1997) in their most basic form, GHG emissions from savanna fires for a defined project area or region can be calculated as the product of the area burnt, the mass of available fuel components (grass litter, CWD, HWF, shrubs) that are consumed and EFs of respective GHG species taking into account their climate warming potential relative to CO₂. For emissions of carbonaceous species, EFs are defined relative to fuel carbon; nitrogenous species are expressed relative to fuel nitrogen. Fuel carbon mass is determined from fuel mass multiplied by fuel carbon content, while fuel nitrogen is derived from the fuel mass by the product of either the carbon content and the fuel nitrogen to carbon ratio, or the fuel mass and fuel nitrogen content. GHG emissions accounting from savanna fires conventionally consider only emissions of methane (CH₄) and nitrous oxide (N₂O), ignoring the significant impacts of other GHGs including CO₂ itself, together with more reactive species (the ozone precursors comprising CO, volatile organic compounds and oxides of nitrogen).

For this assessment, we adapt, with fairly minor modifications, a currently applied version of Australia’s savanna burning methodology that accounts solely for emissions of GHGs (CoA 2015). Under IPCC guidelines, the method is Tier 2, disaggregated spatially, empirically derived with regional-specific EFs and parameters. The method takes into account contrasting emissions profiles produced under different seasonal fire-weather conditions – accounting for less fuel consumption under relatively mild EDS conditions associated with enhanced spatial unburnt patchiness and less fuel pyrolisation, versus generally more intense wildfires under LDS conditions and typically increased fuel loads given progressive dry season litterfall (Cook 2003; Yates et al. 2020). This dry seasonal contrast in fire-weather conditions also applies generally across the distributional range of miombo woodlands (e.g. Frost 1996; Russell-Smith et al. 2021), including the accession of increased LDS fuel loads associated with the dominance of seasonally deciduous miombo taxa (Malaisse et al. 1975; Frost 1996; Ribeiro et al. 2013). Studies both in Australian savanna and African miombo systems demonstrate that EFs of CH₄, N₂O, CO are more or less equivalent from burning cured fuels under EDS or LDS conditions (Hurst et al. 1994; Meyer et al. 2012; Russell-Smith et al. 2021; Vernooij et al. 2023). The methodological approach is only applicable where there is an extended (e.g. 5–6 months) annual dry season period with negligible rain and fully cured ground fuels.

In Australian savanna burning methods, the EDS period is defined as pre-August and the LDS thereafter. Although this temporal division is evidently somewhat arbitrary given inter-regional and -annual variability (Perry et al. 2020; Eames et al. 2023), it reflects generally the time of year in the dry season after which fires begin to burn through the night (Maier and Russell-Smith 2012), and accords well with time-tested Aboriginal seasonal calendars recognising the onset of severe fire-weather conditions (Russell-Smith et al. 2003; Garde et al. 2009; Vigilante et al. 2009). Given markedly equivalent climatic seasonality across the ranges of both Australian savanna and African miombo systems (Frost 1996; Archibald et al. 2009, 2010; Russell-Smith et al. 2021; Eames et al. 2023), for convenience the same pragmatic late July cut-off default is applied here to account for differential seasonal fire behaviour processes (see also Chidumayo 1997a, 1997b).

For the undertaking of savanna burning projects, the Australian emissions accounting method compares mean fire emissions (measured as t CO₂-e) produced during a defined pre-project period (10 or 15 years, depending on MAR variability) against annual emissions reductions resulting from prescribed fire management activities. A generalised description of the application of the emissions methodology at a project scale, including associated emissions accounting and operational procedures, is given in Russell-Smith et al. (2013a). The current savanna burning emissions abatement method is due to be superseded in the near future by a Tier 3 modelled method accounting for either or both GHG emissions abatement and complementary sequestration in woody biomass stocks. Current (and proposed) savanna burning accounting methods directly link with and are accounted for in Australia’s National Greenhouse Gas Inventory.

Southern African miombo woodlands

We focus on development of a robust savanna burning emissions abatement methodology for southern African miombo woodlands covering an extensive portion of southern Africa (Fig. 1a), ranging across 11 countries bounded by the Democratic Republic of Congo and Tanzania in the north, Mozambique in the east, Angola in the west, and Zimbabwe, Botswana and Namibia in the south. Broadly defined, miombo woodlands are included within the Miombo Ecoregion (Fig. 1b; Timberlake and Chidumayo 2011) spanning 3.8 million km², occurring mostly at 800–1250 m above sea level (asl) on typically infertile, acidic, freely draining sandy soils of the ancient African Plateau, under markedly southern hemisphere summer seasonal (November–April) rainfall conditions ranging generally between 600 and 1600 mm MAR. Recent reappraisal of the distribution of miombo vegetation suggests the current extent of miombo woodland in the strictest sense is ~1.9 million km² (Fig. 1c; Dziba et al. 2020). Fires recur frequently throughout the range of miombo ecoregion (Fig. 1d), occurring especially under relatively severe LDS fire-weather conditions (Supplementary Fig. S1), resulting in globally significant GHG emissions (van der Werf et al. 2017; Russell-Smith et al. 2021).

Fig. 1.

Distribution of miombo woodlands: (a) African major vegetation types (Ribeiro et al. 2015 – after White 1983); (b) Miombo Ecoregion vegetation (Timberlake and Chidumayo 2011); (c) Miombo vegetation (Dziba et al. 2020); (d) fire frequency, 2001–2022, derived from MODIS (Moderate Resolution Imaging Spectroradiometer) 500 m automated mapping product (Giglio et al. 2018).

Miombo woodlands are characterised by the dominance of members of the leguminous (subfamily Deltoideae) tree genera Brachystegia (~20 spp.), Julbernardia (2 spp.) and Isoberlinia (1 sp.), typically over well-developed C4 grassy understoreys (White 1983; Frost 1996; Chidumayo 1997a; Timberlake and Chidumayo 2011; Ribeiro et al. 2020b). Although characterised as being generally floristically homogenous across their range, more floristically diverse (notably including many other legume tree species) and structurally better developed (e.g. taller, higher biomass) miombo woodlands occur under higher rainfall conditions (White 1983; Frost 1996; Godlee et al. 2021), especially in Zambia (White 1983). By convention, miombo formations occurring under >1000 mm MAR conditions are described as Wet Miombo, and those occurring under lower (<1000 mm MAR) rainfall conditions as Dry Miombo (e.g. White 1983; Frost 1996; Ribeiro et al. 2020b). Seasonal deciduousness in miombo tree species, hence leaf litter fuel input in the LDS, increases generally with declining rainfall/seasonal soil moisture availability (Frost 1996; Ribeiro et al. 2020b).

As well as the miombo woodlands defined above, for methods development, we have also included extensive legume-dominated woodlands adjoining and intergrading with Dry Miombo in southern, low rainfall regions, dominated by the slow-growing, deciduous tree Baikiaea plurijuga (Zambezi teak) (Fig. 1b, c). Baikiaea woodlands are developed typically on deep, infertile, free-draining Kalahari sands (White 1983) and, like miombo woodlands generally, are very susceptible to severe fire regimes (Frost and Robertson 1987; Holdo 2005; Russell-Smith et al. 2021; Cassidy et al. 2022). Conversely, we have excluded fairly geographically restricted woodland formations from the Miombo Ecoregion as described by White (1983), Timberlake and Chidumayo (2011), Ribeiro et al. (2020b), especially those dominated by Colophospermum mopane (mopane) and Acacia/Senegalia–Combretum (Acacia), and a variety of other Zambezian restricted regional types noted by White (1983). Legume-dominated Acacia and deciduous mopane types typically both occur on heavier-textured, nutrient-rich, less well-drained soils (in the case of mopane, often with deflocculated sodic subsoils) than Miombo and Baikiaea vegetation fuel types. Further field studies would be required to assess the fuel accumulation and emissions properties of these restricted fuel types.

Other vegetation types excluded from our methods application include Afromontane, Evergreen dry forest, Itigi thicket formations (after Timberlake and Chidumayo 2011) and more localised embedded vegetation types associated with termitaria, riparian systems, omuramba (dune swales), and seasonally flooded (dambo) and freely draining grasslands systems. Dambos, omuramba and natural grasslands occur extensively in various southern African landscape settings but are excluded from the proposed emissions abatement methodology given that prescribed more conservative EDS fire management activities would likely result in woody encroachment and thickening, and hence significant deleterious impacts on floristic and wildlife species diversity associated with loss of open grazing (O’Connor et al. 2014; Parr et al. 2014) and avian habitat (Sirami and Monadjem 2012). However, as noted in the discussion, EDS prescribed fire management in adjacent woodland savannas can allow safe application of more intense fires in embedded open grassy habitats.

Sampling rationale and study sites

Given the recognised strong relationship between miombo floristic and vegetation structure with MAR described above, our field assessment program was designed to provide a cost-efficient and -effective strategy for sampling core miombo vegetation structural and fuel load accumulation, combustion and emissions attributes focused on the central geographic miombo rainfall/productivity gradient, from low-rainfall conditions ~600 mm MAR in northern Botswana to more productive, higher rainfall sites ~1500 mm MAR in north-central Zambia (Fig. 2). The field sampling program was undertaken progressively between 2019 and 2023, with an enforced hiatus in 2020 due to COVID (Table 1). Sampling of fuel accumulation and combustion parameters was undertaken also in the Niassa Special Reserve, Mozambique, in 2019, but these unreplicated data (17 individual transects, reported in Russell-Smith et al. 2021) are not utilised here. A similar rainfall gradient-focused sampling approach was taken with developing Australia’s original savanna burning methods, on the basis that sampled fire-prone structural and floristic vegetation types were effectively representative of north Australian savanna vegetation types in general (Russell-Smith et al. 2009; CoA 2013).

Fig. 2.

Location of miombo study sites in Botswana and Zambia, with reference to MAR gradient. Refer text for details.

Field sampling was undertaken at sampling locations (sites) that were unimpacted by contemporary agricultural, pastoral and logging practices, comprising extensive structurally representative vegetation under ambient fire regime and grazing/browsing conditions in designated Forest Reserves and conservation areas (e.g. National Parks and equivalent, Tsodilo World Heritage site; Table 1). For safety reasons, site sampling was undertaken adjacent to access tracks typically within sight of vehicles, mostly under armed guard supervision. Selection of plot locations was based initially on time-since-last-burnt mapping derived from 5 years of Landsat 30 m fire history mapping and, for analysis purposes as presented here, ultimately checked with reference to 10 years prior manual Landsat (30 m pixel) fire mapping following procedures described by Evans and Russell-Smith (2019). The Collection 6 MODIS 500 m burnt area data (after Giglio et al. 2018) were used for identifying potential sites for EDS 2022 fieldwork in northern Zambia, followed by checking from the Landsat archive following plot establishment. It is well recognised that the MODIS 500 m product significantly under-represents (by at least 50%) fire extent in African contexts, especially burnt patches <100 ha (Roteta et al. 2019; Ramo et al. 2021).

Field assessment and analyses

Woody vegetation structural and floristic relationships

Sampled miombo woodland (WDL) and open woodland (OWDL) woody structural attributes are summarised in Table 3 for two MAR classes, following conventional partition of Wet Miombo and Dry Miombo types under >1000 and ~600–1000 mm MAR conditions, respectively (e.g. White 1983; Frost 1996). Under higher MAR conditions, mean WDL structural attributes were observed to be taller, along with substantially larger tree stem densities (including stems >20 cm DBH), shrub densities (especially <50 cm tall), BA and number of species per plot. Although no significantly fire-impacted OWDL miombo vegetation was sampled under >1000 mm MAR conditions, mean OWDL structural attributes in <1000 mm MAR samples were typically substantially less than for equivalent WDL vegetation, notably save relatively taller tree heights and densities of shrubs >100–200 cm tall.

Ordination and associated analyses of a floristic dataset comprising tree species sampled at five or more of 160 plots illustrated that sampled vegetation strongly described a floristic continuum corresponding closely with the rainfall gradient, significantly positively correlated with CCI and BA structural variables, and negatively with FF (Fig. 3, Supplementary Fig. S2 and Table S2). Typical miombo tree species (Brachystegia longifolia, B. spiciformis, Dalbergiella nyasae, Julbernardia paniculata, Pterocarpus angolensis) were representative of higher MAR conditions, whereas Amblygonocarpus andongensis, Baikiaea plurijuga and the tall shrub to small tree Baphia massaiensis were representative of lower rainfall conditions (Supplementary Fig. S2a) – notably, all these representative taxa are legumes. dbRDA analysis confirmed that rainfall explained most variability (23%; P = 0.001) in tree assemblages, followed by CCI (19%; P = 0.001), live BA (12%; P = 0.001) and FF (3.6%; P = 0.001) (Fig. 3, Supplementary Fig. S2b). The sampled frequencies of commonly occurring tree species in the three structural MAR classes are given in Supplementary Table S3.

Fig. 3.

Ordination (dbRDA) based on a distance-based linear model of the relationship between tree floristics assemblages (presence/absence of 62 tree taxa occurring at five or more plots) and predictor variables, at 160 plots. The first axis of the ordination explained 88.5% of the fitted and 23.8% of the total tree floristics variability, and the second axis explained 6% of the fitted and <2% of the total variability. Key: Botswana plots (all <1000 mm MAR): C1W,  Chobe site woodland; C1O,  Chobe site open woodland; T1W,  Tsodilo site woodland; T1O,  Tsodilo site open woodland. Zambia plots: Z2W,  woodland <1000 mm MAR; Z3W,  woodland >1000 mm MAR.

Field observations

Fuel accumulation

Relationships describing accumulation of respective fuel components with time since fire for three vegetation structure–rainfall classes are presented graphically in Fig. 4. Statistically significant accumulation relationships were observed for fine fuels in all three vegetation classes and shrub fuels in WDL classes. Where fuel component accumulation relationships were not found to be statistically significant (P > 0.05), parameter values are given as means (Table 4a) following Russell-Smith et al. (2009) and Yates et al. (2015). Fuel loads in all three vegetation classes were dominated initially by fine fuels and subsequently by shrub fuels, with lesser contributions from coarse and heavy woody fuels. Fine fuels comprised mostly litter, with an increased proportion of litter in LDS samples, in all three vegetation classes (Table 5).

Fig. 4.

Relationship between accumulation of respective fuel components (a–c) and time since fire, in three vegetation structure–rainfall classes. The regression lines, P values and R² values refer to the results of linear regressions of fuel load (natural log-transformed) vs time since fire (natural log-transformed). Refer to text for additional details.

Table 4.Key miombo vegetation emissions calculation parameters assembled from field studies in miombo woodland (WDL) and open woodland (OWDL): (a) fuel accumulation; (b) combustion; (c) burnt patchiness, (d) fuel C and N content; (e) emission factors (CH₄, N₂O, CO); (f) Global Warming Potential.

Parameters and variables	600–1000 mm MAR				>1000 mm MAR		Notes
	OWDL		WDL		WDL
(a) Fuel accumulation (t ha−1)
Fine fuels	y = 0.2137x + 0.8822		y = 0.209x + 0.8248		y = 0.1877x + 0.9501		Significant linear relationships of natural log-transformed data; refer to Fig. 4
	EDS	LDS	EDS	LDS	EDS	LDS	Seasonal litter contributions adjusted for by difference in the proportions of litter: fine fuels determined respectively from EDS and LDS field samples; refer to Methods
1 year since last fire	2.27	2.56	2.16	2.40	2.42	2.75
2 years	2.63	2.97	2.50	2.78	2.75	3.14
3 years	2.87	3.24	2.72	3.02	2.97	3.39
4 years	3.05	3.45	2.88	3.21	3.14	3.57
5 years	3.20	3.61	3.02	3.36	3.27	3.73
6+ years	3.33	3.76	3.14	3.50	3.38	3.86
Shrub fuels	3.37 ± 0.10 (29)		y = 0.5108x + 0.458		y = 0.6876x + 0.1275		Significant linear relationships of natural log-transformed data for WDL classes; refer to Fig. 4
1 year since last fire			1.58		1.14
2 years			2.25		1.83
3 years			2.77		2.42
4 years			3.21		2.95
5 years			3.60		3.44
6 +years			3.95		3.89
Coarse woody debris	0.55 ± 0.08 (30)		0.52 ± 0.04 (68)		0.39 ± 0.03 (50)		No significant linear relationships; refer to Fig. 4
Heavy woody fuels	0.91 ± 0.13 (32)		0.99 ± 0.11 (76)		1.6 ± 0.21 (53)		No significant linear relationships; refer to Fig. 4
(b) Consumption (%)
Early Dry Season
Fine fuels	65.94 ± 2.49 (44)		63.78 ± 2.78 (53)		33.62 ± 10.35 (8) [59.82 ± 3.03 (61)]		Combining WDL observations given apparently anomalous mean and low n
Shrubs	4.90 ± 2.00 (31)		5.65 ± 1.58 (48)		4.65 ± 0.49 (8) [5.42 ± 1.53 (56)]		Combining WDL observations given low n
Coarse woody debris	10.18 ± 1.80 (44)		6.09 ± 1.40 (57)		3.49 ± 1.47 (8) [5.77 ± 1.23 (65)]		Combining WDL observations given apparently anomalous mean and low n
Heavy woody fuels	1.05 ± 0.52 (43)		0.61 ± 0.40 (57)		0.00 ± 0.00 (8) [0.51 ± 0.35 (65)]		Combining WDL observations given apparently anomalous mean and low n
Late Dry Season
Fine fuels	76.42 ± 3.97 (13)		70.00 ± 2.57 (51)		73.04 ± 1.32 (73)
Shrubs	9.33 ± 2.89 (13)		18.90 ± 3.82 (40)		13.76 ± 1.96 (78)
Coarse woody debris	19.23 ± 4.68 (14)		21.18 ± 2.40 (50)		8.65 ± 1.00 (77)
Heavy woody fuels	23.33 ± 8.77 (14)		8.73 ± 2.25 (50)		10.37 ± 1.90 (78)
(c) Burnt patchiness (%)
Early dry season	67.54 ± 3.45 (44)		68.08 ± 3.26 (57)		39.93 ± 6.03 (8) [64.62 ± 3.16 (65)]		Combining WDL observations given apparently anomalous mean and low n
Late dry season	81.89 ± 4.03 (14)		94.28 ± 1.39 (50)		96.32 ± 0.69 (78)
(d) C and N content (%)							Refer to Russell-Smith et al. (2021)
Early Dry Season
Fuel C content	49.01 ± 0.25 (12)		43.48 ± 2.73 (10)		45.43 ± 2.73 (13)
Fuel N:C ratio	0.020 ± 0.00 (12)		0.015 ± 0.005 (10)		0.014 ± 0.005 (13)
Late Dry Season
Fuel C content	48.59 ± 0.04 (14)		46.44 ± 0.06 (31)		46.28 ± 0.37 (22)
Fuel N:C ratio	0.020 ± 0.00 (14)		0.020 ± 0.00 (31)		0.012 ± 0.001 (22)
(e) Emission factors							Refer to Vernooij et al. (2023)
Early Dry Season
CH4 (% C)	0.19 ± 0.04 (2)		0.25 ± 0.02 (5)		0.26 ± 0.02 (11)
N2O (% N)	0.60 ± 0.05 (2)		0.45 ± 0.08 (5)		0.43 ± 0.08 (11)
CO (% C)	5.0 ± 0.8 (2)		5.8 ± 0.2 (5)		5.9 ± 0.4 (11)
Late Dry Season
CH4 (% C)	0.18 ± 0.01 (5)		0.31 ± 0.02 (16)		0.35 ± 0.02 (10)
N2O (% N)	0.70 ± 0.11 (5)		0.45 ± 0.05 (16)		0.54 ± 0.06 (10)
CO (% C)	4.8 ± 0.2 (5)		5.8 ± 0.4 (16)		6.6 ± 0.4 (10)
(f) Global Warming Potential
CH4	34		34		34		Refer to IPCC AR5 (Myhre et al. 2014)
N2O	298		298		298		Refer to IPCC AR5 (Myhre et al. 2014)
CO	2.2		2.2		2.2		Chosen by us as a conservative estimate; refer to text for details

Errors given as s.e.m.; n given in parentheses.

Table 5.Proportion of litter in fine fuel samples; values of n given in parentheses.

Sample	Mean litter fraction in fine fuels (%)
	OWDL (<1000 mm MAR)	WDL (<1000 mm MAR)	WDL (>1000 mm MAR)
EDS	55.3 (24)	64 (52)	65.7 (22)
LDS	75.5 (8)	78.1 (24)	85 (31)
Annual	59.8 (32)	67.3 (76)	76.4 (53)

For WDL classes, total fuels increased from 5.5 to 9.0  and 5.9 to 9.7 t ha⁻¹ after 6 years in Dry Miombo (<1000 mm MAR) and Wet Miombo (>1000 mm MAR) samples respectively, especially associated with developing shrub fuels. OWDL fuels exhibited limited accumulation with time since fire, from 7.4 to 8.6 t ha⁻¹, attributable to relatively large sampled shrub fuel loads in years immediately following fire (Table 4a).

Fire management and emissions scenarios

We applied assembled accumulation, combustion and EF parameters to a 2716 km² regionally characteristic case study site with a recent (last 10 years) fire history dominated by LDS fires (annual mean  35.6%) and little prescribed EDS burning (annual mean  3.2%). Based on emissions calculations for 2022 (Supplementary Fig. S3), GHG emissions from EDS fires are approximately half (56%) those of LDS fires in OWDL, and one-third (34%) those in both WDL types (Table 7). Under enhanced EDS fire management, emissions reductions of between 30 and 40% are realistically achievable in respective vegetation structural types under fire frequencies of 20% prescribed EDS fire and 20% LDS fire, and 36–50% emissions reductions with 25% EDS and 10% LDS fire frequencies (Table 7).

Emissions abatement framework

Building on earlier technical exploration of the feasibility of applying a savanna fire management emissions abatement methodology derived from Australian conditions to fire-prone southern African woody savanna sites at Tsodilo, northern Botswana, and Niassa, northern Mozambique (Russell-Smith et al. 2021), the present paper extends that assessment towards developing a robust emissions accounting framework pertinent to expansive woodlands covering ~2 million km² of the southern African miombo ecoregion. Based on Australian precedent (CoA 2015), our proposed framework is Tier 2 (IPCC 2000), empirically derived, comprising spatially stratified and disaggregated parameters (e.g. different vegetation fuel types, fuel components, combustion and emissions under different seasonal conditions). One advantage of such an essentially bottom–up Tier 2 approach is that emissions calculation components and processes are readily transparent; by contrast, a Tier 3 approach typically involves complex, non-linear carbon cycle modelling such as will be applied in the next generation of Australian savanna burning methods, due from 2024. Although this proposed framework can be applied also to meet national and United Nations Framework Convention on Climate Change (UNFCCC) emissions accounting and reporting commitments as part of the Agriculture, Forestry and Other Land Use (AFOLU) sector, our primary focus here is to address project-scale accounting needs – i.e. to account for mitigated GHG emissions delivered through enhanced site-based fire management activities. Here, we address core components of the proposed miombo emissions accounting framework; limitations and ongoing methodological challenges associated with assembled plot data and derived framework parameters are summarised in Table 8.

Our field program sampled miombo vegetation structural and associated fuel load accumulation, combustion and emissions attributes, focused on the recognised central geographic miombo floristics and productivity gradient, from low-rainfall conditions in northern Botswana to higher-rainfall sites in north-central Zambia (White 1983; Frost 1996; Godlee et al. 2021). An equivalent centralised rainfall gradient-focused sampling approach was undertaken with developing Australia’s original savanna burning methods (Russell-Smith et al. 2009). Ordination confirmed the strong floristic and associated vegetation structural continuum with the rainfall/productivity gradient described by our assembled common tree species dataset.

Following regional convention (e.g. White 1983; Frost 1996; Timberlake and Chidumayo 2011; Ribeiro et al. 2020b), we have recognised Wet and Dry Miombo fuel types occurring under conditions of >1000 and <1000 mm MAR, respectively. Similarly, in the Australian savanna burning method, respective vegetation fuel types in High and Low Rainfall Zones are delimited by the 1000 mm MAR isohyet. From an emissions methodology perspective, however, definitive geographic separation of Wet and Dry Miombo types presents practical difficulties given the paucity of meteorological stations and associated unreliability of rainfall surfaces (Trisos et al. 2022); here, we have attributed MAR per plot from the widely applied European Centre for Medium-Range Weather Forecasts Copernicus global data set (https://cds.climate.copernicus.eu/). Unlike with the Australian approach (Whitehead et al. 2014), here we do not define a priori the lower MAR limits applicable to the method; rather, we propose that project proponents should select prospective project areas after assessment of burnt area mapping and associated emissions estimates, and eligible miombo VFTs as described below.

Structurally, we have distinguished betweenWDL and OWDL miombo VFTs where, following Australian convention, OWDL exhibits canopy cover <20% and WDL >20% (CoA 2012; Fig. 5). Unlike for Australian savannas, we do not include an Open Forest class (OF canopy cover >50–80%) given that southern African studies report that fire occurrence is effectively truncated where percentage tree cover exceeds ~40% (Scholes 2003; Archibald et al. 2009). We note, however, that assessments reported by Archibald et al. (2009, 2010) were undertaken with burnt area mapping products derived from fairly coarse-scale MODIS 500 m imagery, which, subsequently, has been shown to significantly underestimate the extent of especially smaller fires <100 ha in African contexts (Roteta et al. 2019; Ramo et al. 2021). In eucalypt OF savannas, fires occur extensively in litter-dominated fuel loads (e.g. Russell-Smith et al. 2009).

Fig. 5.

Miombo Vegetation Fuel Types on typical sandy substrates: (a) deciduous miombo WDL, Kasane Forest Reserve, Botswana – LDS, 680 mm MAR; (b) remnant fire-affected shrubby OWDL, Tsodilo Hills World Heritage Site, Botswana – EDS, 640 mm MAR; (c) shrubby OWDL, north-west Ngamiland, Botswana – LDS, 660 mm MAR; (d) miombo WDL, Kafue National Park, Zambia – EDS, 970 mm MAR; deciduous miombo WDL, Chisamba, Zambia – LDS, 950 mm MAR; (f) miombo WDL, Lualaba Forest Reserve, Zambia – EDS, 1340 mm MAR; (g) semi-deciduous miombo WDL, East Lunga Forest Reserve, Zambia – LDS, 1100 mm MAR.

To distinguish between OWDL and ineligible grassland types, we adapted the definition of White (1983) that eligible OWDL should comprise at least 10% cover of typical miombo tree and shrub taxa. Data presented in Table 3 illustrate that OWDL plot samples typically supported (with 95% confidence) >30 tree stems ha⁻¹, and >1000 shrubs taller than 1 m ha⁻¹.

All OWDL plots sampled in our assessment were located under <1000 mm MAR conditions, typically comprising fire-disturbed shrubby vegetation with sparse tree stem recruitment. Although we did not sample heavily disturbed fire- and agriculturally affected (e.g. slash and burn) miombo sites under higher rainfall conditions, for example severely fire-disturbed early successional chipya vegetation types in northern Zambia (Lawton 1964, 1978), disturbed miombo vegetation is widespread and rapidly expanding (Syampungani et al. 2020; Zulu et al. 2020). For present methodological purposes, we suggest that emissions accounting parameters determined for disturbed OWDL sites under <1000 mm MAR conditions apply also to higher-rainfall OWDL sites provided that such sites meet essential eligibility criteria including the potential for functional miombo vegetation regeneration from extant root sucker and coppice resources (Stromgaard 1985; Frost 1996; Chidumayo 1997a, 2004, 2013; Williams et al. 2008; Syampungani et al. 2016). Although further assessment is evidently warranted (Table 8), such an approach follows conservative emission accounting principles given observed enhanced productivity, including of grass fuels in disturbed sites (Chidumayo and Kwibisa 2003), under higher rainfall conditions.

Sampling of fuel loads and fuel combustion was undertaken under ambient grazing, browsing, termite harvesting conditions mostly in forest and conservation reserves that included substantial herds of larger ungulates and elephants, especially in the Chobe region (Botswana) and Kafue NP (Zambia). Similarly, field sampling in Australia was conducted under ambient grazing, browsing, termite harvesting conditions (avoiding heavily impacted sites) involving macropods, cattle and Asian water buffalo. Fuel loads comprised mostly fine fuels and shrubs in all three miombo VFTs (WDL > 1000 mm MAR; WDL and OWDL < 1000 mm MAR), with the proportion of shrub fuels increasing with time since fire (TSF) in WDL types and exhibiting no relationship with TSF in OWDL. The contribution of mean shrub fuels in miombo fuel types is three to four times greater than that observed for comparable VFTs in Australian savannas, for example: 3.37 t ha⁻¹ in OWDL vs 1.12 t ha⁻¹ OWLMixed grasses (Yates et al. 2015); 2.23 t ha⁻¹ in WDL > 1000 mm MAR vs 0.5 t ha⁻¹ WLMixed grasses (Russell-Smith et al. 2009). Other than for shrubby fuels, means of other fuel load components are observed to be more or less comparable in the two savanna systems, with the exception of substantially greater (2 t ha⁻¹) fine fuels in higher-rainfall eucalypt woodlands. Overall, mean fuel loads (unadjusted for seasonal inputs) in comparable miombo and eucalypt savanna VFTs are as follows: 7.9 t ha⁻¹ in OWDL vs 5.86 t ha⁻¹ OWLMixed grasses (Yates et al. 2015); 7.2 t ha⁻¹ in WDL <1000 mm MAR vs 5.9 t ha⁻¹ WLMixed grasses (Yates et al. 2015); 7.4 t ha⁻¹ in WDL >1000 mm MAR vs 8.9 t ha⁻¹ WLMixed grasses (Russell-Smith et al. 2009).

Despite grassy fuels being considered the archetypic fuel type of savannas (Scholes and Archer 1997) and recognition that grass production is inversely related to woody cover in more mesic savannas (Cook 2003; Scholes 2003; Dohn et al. 2013), fine fuel loads were dominated by leaf litter in all three miombo VFTs. By contrast, litter fuels are the dominant component of fine fuels only in more wooded eucalypt VFTs (Russell-Smith et al. 2009; Yates et al. 2015). Litter inputs in miombo systems exhibit strong seasonality, especially under more xeric conditions, including OWDL with extensive deciduous shrubby components (Malaisse et al. 1975; Frost 1996; Ribeiro et al. 2013; Russell-Smith et al. 2021). Such seasonal litter inputs have significant implications for resultant fire intensities and, as noted below, EFs. In the absence of empirical seasonal litter accession data, following Australian precedent (Russell-Smith et al. 2009; Yates et al. 2015), here, we have accounted for increased LDS litter fuels by adjusting the difference in the proportions of litter:fine fuels determined respectively from EDS and LDS field samples. This adjustment is likely very conservative and warrants further field assessment under different rainfall/productivity gradient conditions (Table 8). By contrast, grassy fuel load components, once cured, may be anticipated to be aseasonal except where extensively grazed or consumed by termites (Scholes et al. 1996; Hély et al. 2003). Given the absence of apparent accumulation relationships between CWD and HWF with TSF, respective inputs are most likely associated with stochastic events such as severe wet season storms and fire events (Yates et al. 2020) and, in African contexts, destructive impacts of large animals such as elephants (Mosugelo et al. 2002; Ribeiro et al. 2008).

Fuel combustion and associated fire patchiness characteristics were observed to be strongly related to fire severity, and thereby seasonal conditions, as widely reported in the savanna literature (e.g. Frost and Robertson 1987; Shea et al. 1996; Chidumayo 1997b; Williams et al. 1998; Russell-Smith et al. 2009; Oliveira et al. 2015). Major contributions to resultant quantities and seasonal differences in emissions are associated especially with consumption of fine fuels and burnt patchiness (Table 4). Although acknowledging that imposed experimental fire treatments were designed to emulate prescribed low-severity EDS and LDS wildfire conditions respectively, the resultant fire severity frequency distributions of both are generally similar to those observed under Australian ambient savanna fire regime conditions, for example: miombo – EDS 89% low severity, LDS 89% moderate + high severity (Table 6); eucalypt – EDS 74% low severity, LDS 84% moderate + high severity (Edwards et al. 2015). Such observations reinforce the practical feasibility of implementing prescribed strategic EDS fire management where appropriate.

In the absence of current means for reliably mapping fire severity or intensity at appropriate spatial and temporal scales to enhance combustion estimates, for convenience we have applied the end of July as a default cut-off between the EDS and the LDS. Although this suffices for our study sites given the seasonal distribution of fire severity observations, we recognise that some variation may be appropriate to cater for regional differences, e.g. for parts of northern Angola and southern Mozambique where the driest 6-month period commences 1 month earlier (Russell-Smith et al. 2021, fig. 1c).

Our methodological approach addressing combustion calculations differs in two notable, albeit probably quantitatively minor, regards to that applied in previous Australian assessments. First, here we have applied EDS and LDS combustion parameters representing the mean of respective EDS and LDS treatment observations, whereas in Australian assessments, EDS and LDS combustion parameters account for the seasonal proportions of low, moderate and severe fires as observed at long-term, ambient fire regime monitoring plots (Russell-Smith and Edwards 2006; Edwards et al. 2015). Second, for combustion calculation purposes, here we have not distinguished between unburnt and typically very small quantities of burnt residual materials (e.g. ash, char fractions) in post-fire sample quadrats but, rather, combined both on the basis that (a) experience with separating the two components in field assessments has proved practically difficult, and (b) resultant gaseous emissions estimates are therefore conservative.

The limited seasonal fluctuations of EFs observed here in seasonally dry woodlands are consistent with previous studies from northern Australian savanna (Meyer et al. 2012), miombo types (Hoffa et al. 1999; Russell-Smith et al. 2021) and Brazilian cerrado (Vernooij et al. 2021). The increase of CH₄ EF in both miombo WDL types in the LDS is consistent with findings from Brazilian cerrado (Vernooij et al. 2021). This, and observed LDS increases also of N₂O and CO EFs, may be related to a larger contribution of smouldering combustion-prone fuels (e.g. leaf litter) in the LDS, combined with more efficient combustion of litter and grass. Importantly, these results reinforce earlier observations (Russell-Smith et al. 2021) that prescribed EDS fire management undertaken in cured miombo fuels affords conservative emissions benefits.

To compare our EF measurements with previous literature, here we adopt the widely used format of grams emitted per kilogram of dry matter consumed, departing from the percentage-based representation given in Table 4. Our EFs were computed using the carbon mass balance methodology, following Vernooij et al. (2023). In OWDL, our CH₄ and CO EFs were slightly lower than global savanna averages; EF(CH₄) = 1.85 vs 1.22 (EDS), 1.45 (LDS); EF(CO) = 64 vs 57 (EDS), 55 (LDS). In both WDL types, our values exhibited a nuanced temporal pattern, with lower EFs during the EDS and higher EFs during the LDS.

Conversely, for N₂O, compared with the global savanna average EF of 0.16 (Vernooij et al. 2023), EFs exhibited slightly higher values in OWDL (0.24 EDS; 0.3 LDS) albeit from low numbers of observations, and lower values in WDL types (<1000 mm MAR: 0.09 EDS, 0.1 LDS; >1000 mm MAR: 0.1 EDS, 0.11 LDS). Although our study provides fixed EFs for EDS and LDS fires, EFs are influenced by various factors including fuel characteristics like moisture content and deciduous litter accumulation, and meteorological variables such as wind speed and humidity, changing progressively throughout the fire season (Vernooij et al. 2023).

Application of the proposed fire management emissions abatement framework to a realistic project-scale case study illustrates that EDS prescribed fire management can substantially reduce LDS emissions in all three miombo VFTs (Table 7, Supplementary Fig. S3). In Australian projects, net emissions from EDS fires have been shown to be ~50% less per unit area than LDS fires under high rainfall conditions (Russell-Smith et al. 2009), and ~25% under lower rainfall conditions (Russell-Smith et al. 2015). Given current predominance of LDS fires in miombo systems, it is realistically feasible to significantly reduce both LDS and annual BA extent and realise a variety of climate change, hydrological, ecological, grazing management and socio-economic benefits (e.g. McNaughton 1985; Chidumayo 1997a; Ribeiro et al. 2020a; Trisos et al. 2022).

Complementary approaches

In addition to abatement of GHG emissions, strategic prescribed EDS fire management projects can also deliver significant sequestration in miombo woody biomass and possibly SOC. With respect to woody biomass sequestration, it is widely observed that fire regimes dominated by fairly intense fires have significant destructive impacts on miombo woodland systems, especially under more xeric conditions (e.g. Trapnell 1959; Frost 1996; Chidumayo 1997b; Holdo 2005; Ryan and Williams 2011; Ribeiro et al. 2020b, 2021; Cassidy et al. 2022). As such, there is considerable opportunity for developing complementary sequestration methods. To date, two studies have assessed the effects of fire regimes on woody tree biomass sequestration potential in more or less intact miombo systems, both based on gap phase modelling simulating photosynthesis and carbon allocation processes: one predicting that annual burning (of whatever intensity) would result in almost total loss of tree biomass (Ryan and Williams 2011); the other predicting that annual fires (based on an ambient fire regime dominated by LDS fires) would result in woody biomass stabilising at a level substantially less than that under no burning (Ribeiro et al. 2021). Further assessment of the sequestration effects (and benefits) of low-severity fire regimes is warranted, however, given contrasting results from a long-term replicated fire experiment conducted in miombo woodland at Ndola (now Zambia) where it was observed that annual EDS low-intensity fires resulted in negligible stem mortality, similar to that observed on unburnt plots (Trapnell 1959).

Significant opportunities for relatively rapid woody biomass sequestration have been identified in studies of miombo regeneration following clearing and cutting in agricultural settings. Given availability of substantial miombo root sucker and coppice recruitment stocks following initial clearing activities, and low-intensity disturbance impacts especially from infrequent burning thereafter, re-establishment of site woody biomass can be attained within as little as a few decades (e.g. Chidumayo 1997a, 1997b, 2019; Williams et al. 2008; Gumbo et al. 2018). As Syampungani et al. (2020) observe, such vigorous regeneration potential on degraded sites offers substantially enhanced carbon sequestration opportunities compared with essentially moribund old-growth stands.

Sequestration of SOC also has been shown to have possible, if variable, potential in southern African savanna systems under conditions of: (a) longer-term (multi-decadal) absence of, or infrequent, low-intensity burning; (b) commensurate maintenance of tree canopy cover to provide sustained litter inputs in addition to inputs from grasses; (c) in soils with higher clay contents (Bird et al. 2000; Williams et al. 2008; Holdo et al. 2012; Pellegrini et al. 2018, 2020; Zhou et al. 2022; Coetsee et al. 2023). SOC sequestration potential in sandy miombo soils with characteristically very low organic content and base-exchange capacity is further limited given the complementary impacts of recurrent fires and intense harvesting on leaf litter inputs (Trapnell et al. 1976; Frost 1996). In contrast to woody sequestration potential, this applies also to cultivated sites given very slow rates of identifiable SOC accumulation following agricultural abandonment, especially when associated with ongoing fire disturbance (Chidumayo and Kwibisa 2003; Walker and Desanker 2004; Williams et al. 2008). Based on evidence from a multi-decadal study of SOC sequestration potential under different experimental fire regimes, Coetsee et al. (2023) conclude that frequent low-severity fires offer a conservative management approach – at least in habitat conditions not undergoing detrimental woody encroachment.

Prescribed fire management emissions abatement projects can also complement other developing and potential market-based NBS (Nature-Based Solutions) and PES (Payments for Environmental Services) approaches applicable to miombo systems. The most established of these, albeit at early stages of development in different southern African countries, concerns REDD+ projects that aim to Reduce Emissions from Deforestation and forest Degradation through sustainable management of forests and sequester carbon for climate-change mitigation. Although Syampungani et al. (2020) note the substantial potential of REDD+ for ameliorating miombo degradation and community benefit, to date there have been mixed results and emerging governance concerns, including reliability of measurement and monitoring processes, and equitable economic benefits for participating communities (Zulu et al. 2020). Current experience with a foundational REDD+ project at Kariba, Zimbabwe, highlights both these challenges and, in the context of the present study, the importance of the integrity of certification schemes (Blake 2023).

Applying the framework

Credible carbon emissions markets are reliant on generating high-integrity, transparently accounted credits. Incorporation of our proposed miombo fire management emissions abatement framework within a formal accreditation process, or Standard, potentially can involve application of two (non-exclusive) pathways: registering a project under a voluntary carbon market standard; utilising emerging bilateral (or plurilateral) government and private sectors arrangements under especially Article 6.2 of the Paris Agreement (Minas 2022). An instructive example of the former approach is provided by the Nakau programme, which utilises a commercial certification scheme to underpin community-owned, high-integrity carbon projects in the Pacific that support nature-based, cultural and climate resilience values (https://nakau.org). The second pathway draws on experience with Australian Government and UNFCCC-sanctioned savanna burning projects as a basis for establishing ethical terms and responsibilities in contractual agreements between participating communities, governments and purchasers of carbon credits, with an annex that outlines the relevant Standard to be applied for the project.

Application of appropriate emissions accounting methods and robust Standards is reliant also on independent MRV (Monitoring, Reporting, Verification) processes and structures, ideally servicing both national (e.g. Nationally Determined Commitments to the UNFCCC under the Paris Agreement) and linked project-scale requirements. In Australia, the regulated savanna burning MRV system, SAVBAT (https://savbat.environment.gov.au), currently fulfils both roles, where mitigation achieved through emissions abatement projects links directly with and is accounted for in Australia’s National Greenhouse Gas Inventory. Development of a Tier 2 MRV system for REDD+ applications in southern African countries has been initiated through a partnership involving SADC (Southern African Development Commission) and the German development agency GIZ (Deutsche Gesellschaft für Internationale Zusammenarbeit).

As identified in the introductory remarks, there are two significant types of caveats that need to be considered when implementing incentivised savanna burning projects. First, application of a fairly conservative, low-severity EDS fire management approach is not appropriate in grassland situations at risk of exacerbated woody encroachment and associated loss of habitat and dependent biodiversity (e.g. Parr et al. 2014; Veldman et al. 2015). However, strategically undertaken savanna burning projects can provide a funded, operational framework to assist with delivering a range of objectives including, for example, safe application of intense fires in a matrix of reduced fuel loads for woody plant control and maintaining open grassy habitats (Edwards et al. 2021). Second, as cautioned in related studies (e.g. ISFMI 2015; Russell-Smith et al. 2013b, 2021; Zulu et al. 2020), there are a variety of significant economic, societal and political implementation challenges that require informed and sustained consideration if the benefits of market-based carbon offset projects are to be realised. We recognise that addressing both these conditional challenges requires considered site-based assessment.