Methane gas emissions from savanna fires : What analysis of local 1 burning regimes in a working West African landscape tell us 2 3

3 Paul Laris ; Moussa Koné; Fadiala Dembélé ; Christine M. Rodrigue ; Lilian Yang ; Rebecca Jacobs ; 4 Quincy Laris 5 1 Geography Department, California State University Long Beach, 1250 Bellflower Blvd. Long Beach CA 90840. 6 USA. 7 2 Institut de Géographie Tropicale (IGT), UFR-SHS, Université FHB de Cocody-Abidjan, Côte d'Ivoire. 8 3 Institut Polytechnique Rural de Formation et de Recherche Appliquée de Katibougou, Mali. 9 4 Department of Civil and Environmental Engineering, University of California, Berkeley. 10 11 Correspondence to Paul Laris paul.laris@csulb.edu 12 13


Introduction
In sum, while savannas undoubtedly harbor great theoretical potential to sequester more carbon, and emit 116 less through a change in fire regime there exists a great deal of uncertainty as to what the actual carbon shifts might 117 be, should regimes change. Fire regimes are themselves complex; we define them as the characteristic fire activity 118 prevailing in a region, typically determined by frequency, intensity, seasonality, size distribution, type of fire and 119 fuels consumed (Pausas and Keeley 2021). Changes in one or more of these factures can alter fire emissions. We 120 suggest the key sources of uncertainty in terms of carbon emissions arise largely from the spatiotemporal complexity 121 of savanna vegetation patterns and fire regimes combined with many unknowns or biases associated with a lack of 122 consideration of human fire setting and land management practices in these complex landscapes (Laris 2021  falling, and humidity rising, which limits fire intensity (Laris et. al. 2020). Lower intensity fires tend to self-151 extinguish at the edge of moister vegetation patches and in the evening; they have lower flame heights reducing the reach of fires into leafy tree canopies (Laris et. al. 2021). Later in the fire season, fires are less likely to be 153 purposefully set and are more likely to burn as intense, uncontrolled head-fires.

154
It is clear that any effort to predict future changes in emissions or to implement policy to reduce emissions 155 requires more detailed information on how emissions vary according to the key factors noted above, many of which 156 are a function of human land management practices (see appendix). Specifically, given the spatiotemporal 157 complexity of savanna environments, whether a shift to an earlier fire regime will result in a decrease in methane 158 emissions for a given savanna must be determined empirically and proposed policies to apply generalized findings 159 from one continent to another may not achieve desired emissions reductions.

160
This study aims to fill a knowledge gap by incorporating data on human burning practices, the 161 characteristics of the fire regimes they produce, the vegetation conditions on the landscapes they burn and the 162 resulting emissions of key GHG gases. Through a novel geographic approach, we designed our experiments to 163 gather data in ways that reflect actual on-the-ground burning practices of people living in working landscapes at two 164 mesic savanna sites in Mali, West Africa. By "working landscapes," we mean savanna lands that are occupied and 165 worked by people as opposed to areas managed as reserves (e.g., Charnley et al., 2014); the latter are most often 166 used in fire research. The biomass (fuels) in working landscapes are a function of land use practices including 167 rotational agriculture, annual burning, and animal grazing and can differ significantly from those found on non-168 working lands (Figures 1d and 1e), which can affect fire intensity, combustion completeness and combustion 169 efficiency with implications for gas emissions. The burning regimes studied, which are determined by such factors 170 as seasonality, time of day, (ambient weather), fire type (with or counter to the wind), grass type and woody 171 vegetation cover, were selected to reflect local practices and based on over a decade of field and remotely sensing 172 research.

173
To determine the factors that influence fire emissions of methane gas from anthropogenic fires we 174 conducted experimental fires using a field-based method to measure key factors. We collected canister samples of 175 smoke emissions for 36 fires during the early and middle seasons, which we report on here. We also collected data 176 for savanna type, grass type, biomass composition and amount consumed; scorch height, speed of fire front, fire type 177 and ambient air conditions for two mesic savanna sites in Mali. 178 179

180
We based our research in two working landscapes located in the southern Sudanian savanna of southern 181 Mali (Fig. 2). We chose areas with annual precipitation over 900 mm because they burn frequently and are typical of

195 196
The vegetation is southern Sudanian savanna and is predominantly composed of a mixture of grasses, trees, 197 and shrubs in a complex mosaic. The landscape heterogeneity is a function of topography, underlying soil and 198 hydrology, as well agricultural uses, the combinations of which produces unique patterns of land cover (Duvall, 199 2011;Laris, 2011). Ferricrete outcrops on hard pan cover considerable areas. Soil in these areas generally has high 200 gravel content and is very shallow, creating xeric conditions. Vegetation is dominated by short, annual grasses 201 (principally Loudetia togoensis but also Andropogon pseudapricus) and usually have few widely scattered trees.

238
We selected plots to represent an array of savanna vegetation types dominated by different amounts of 239 woody cover and grass species. To aid in the selection of the burn plots, we used a long-term fire database to select 240 sites with known fire seasonality-fires known to burn during the early, mid, or late fire season on an annual basis 241 (Laris, 2011). We divided the sites into plots of 10 x10 meters and applied fire treatments of head and back burns.

290
We calculated EF as: EFx is the emissions factor for species x (g/kg). Fc is the mass fraction of carb in in the fuel for which we use the 295 value of 0.5 (the majority of studies find the carbon fraction to vary between 0.425 and 0.50; the latter is used most 296 often for purposes of comparison (Ward et al., 1996)   Here, BA is burned area, FL is fuel load, CC is combustion completeness, EFx is emission factor of species x. We

372
The characteristics of the fires also vary by season ( Table 2)

384
The mean EF for methane was 3.47 g/kg and the mean MCE was 0.90, which is considered on the cusp 385 of flaming and smoldering (Table 2). Our study found that methane EFs ranged from 3.83 g/kg in the EDS to 3.18 in 386 the MDS. These differences yield a weak effect size of 0.25 (Cohen's d) but the small sample did not provide 387 enough power (1-β=0.11) for this effect to be significant (p=0.45). The results indicate that fire type has a larger 388 impact on methane EF than fire season. Head-fires had nearly double the CH4EF of backfires (5.12 g/kg to 2.74g/kg)   We found no significant relationship between Byram's intensity and CO EF ( Figure 5) and no significant 410 relationship between EFCH4 and either total moisture or calculated Viney moisture or percent grass in the biomass.

411
We did find a negative and significant relationship between MCE and EFCH4 as expected (P=0.000001; R 2 adj = 412 0.436), however, the effects of fire type can be seen here as well. When head-and backfires are examined 413 separately, the relationship between CH4 EF and MCE for back-fires is much stronger than head-fires ( Figure 6).

414
Similarly, for MCE and methane density we found a stronger relationship for back-than head-fires.

426
Our study finds that methane EF means were highest for EDS as expected and dropped by about 20% by 427 the MDS. We found, however, that fire type had a greater (and more significant) impact on methane EF than season; 428 head-fire methane EFs were nearly double those for backfires (5.12 g/kg to 2.74 g/kg). In general, methane EFs 429 increased as fire intensity increased and head fires, which have higher fire intensity, had higher methane EF 430 regardless of season. Increased fire intensity results in taller flame heights, which reach into the tree canopies of the 431 numerous small trees and burn greater amounts of fresh green leaves (Figure 7). Indeed, our field observations 432 recorded the highest methane emissions (over 5000 ppm) during the combustion of green leaves on small trees. We 433 were not able to determine the amount of leaves on trees that were combusted in this study, although it is reasonable 434 to estimate that more green leaves would burn on trees in the EDS than other seasons. Interestingly, we did not find

490
Contrastingly, they found little difference by season in methane EF, for dry wooded savannas and a surprisingly 491 higher EF in late season for more wooded savannas, which contrasts with our findings and those of Korontzi (2005) 492 for wooded savannas. We must note, however, that the date chosen for Russell-Smith's EDS is more comparable 493 with the MDS date used in our study-both dates represent the "middle" of the dry season. Indeed, as Russell-Smith West Africa. These so-called "Gulliver" trees are often less than 2-meters tall because they repeatedly burned back 529 to the rootstock by annual fires (Laris and Dembele 2012). We argue that burning of small trees contributes 530 significantly to methane release. As such, we cannot support the policy suggestion put forth by Lipsett-Moore 531 (2018) who promote increased early burning in African savannas to reduce methane emissions. While it is 532 theoretically possible that very early fires would burn a lower fraction of the landscape than we have observed, we 533 argue that such a policy is just as likely to cause an increase in methane emissions due to higher methane EF of 534 earlier burning, which may be a function of green leaf combustion (see Korontzi 2005). It is also important to note 535 that higher intensity head fires would be required to increase the burned area of moist perennial grasses in the EDS 3 536 and because head-fires have a methane EF nearly double that of backfires, burning with head fires would likely 537 2 We note that results from our larger study of 97 fires found a less dramatic rise in BP and CC from EDS to MDS to LDS than for the sample of 36 fires used here. In the larger study, BP increased marginally as the dry season progressed to a near complete burn by the late season (85.3% to 92.3 to 99.2). CC increased very slightly from early to mid before increasing substantially in late season (85.1% to 86.4 to 92.8) (Laris et al., 2020). These findings suggest a stronger emissions trade off than reported here. 3 We made several attempts to burn perennial grasses in December and could not get them to ignite. Only under windy, head-fire conditions will perennial grasses burn in the EDS.

559
This study finds that when fires are set in working landscapes in accordance with well-documented burning 560 practices of West African people, methane EFs decreased from early dry season to mid-dry season (although the 561 results were not significant). We also found that methane emission density increased only marginally from EDS to 562 the MDS a difference that was not significant. We found that fire type had a much greater effect on methane 563 emissions than fire season with head-fires having significantly higher methane EF compared to backfires and 564 significantly higher methane densities due to higher fire intensity. We note that we are unaware of any estimates for 565 area burned according to fire type for any of the world's savannas.

566
It is important to reiterate that several key findings of this study arise from documented burning practices of 567 people living in working landscapes. People set fires in West Africa later in the day resulting in fires with lower 568 intensity due to lower wind and air temperature, and higher humidity; and people set predominantly backfires all of 569 which contribute to lower intensity burning, which results on lower methane emissions. In addition, we note that the 570 fuel loads we recorded are nearly 50% lower on working savanna lands compared to reserve lands used in some 571 other studies (Laris et al. 2020). Finally, the number of fires peaks in the West African region in the MDS and 572 although the methane emissions density values for the EDS were slightly lower than for the MDS, a significant reason for this was the increased fuel load from leaf litter in the MDS. We should note that EDS fires tend to burn 574 more green leaves on trees, which are not accounted for in this study.

575
In conclusion, our study finds that several factors influence the emissions from savanna fires including the 576 fire season, fuel load and type, and, most importantly, fire type. Each of these factors are a function of human land 577 and fire management practices. We also conclude there is an emissions trade-off in setting fires earlier and, as such, 578 a policy to increase the amount of early burning in West African would be very difficult to implement because much 579 burning is already "early" and because earlier burning of uncured grasses and green leaves would likely result in 580 higher methane EFs. Moreover, any policy aimed at increasing the amount of early burning would likely require 581 setting head fires, which would decrease burn patchiness and increase combustion completeness further negating the 582 effects of any reduction in burned area while also causing an undesired increase in uncontrolled fires.
583 584 People play a critical role determining the point at which fires are set often according to fuel moisture level of grasses at fine spatial resolution.
High. Fuel moisture is often considered to be a function of seasonality; however, there is high spatial heterogeneity in savannas. A single fire can burn one type of grass with high fuel moisture and another with low moisture with implications for CH4 EF.

Patchiness
High. Fires tend to burn in a patchy manner especially when vegetation is not uniformly dry and when burning as a backfire. Patchiness created by earlier fires, prevents spread of later ones.
People create a patch-mosaic by systematically burning the driest patches on the landscape first fragmenting the landscape and creating a patch-mosaic with new, old and unburned patches.
Low. Advances in remote sensing and image processing algorithms have improved estimates of patchy burning although the smaller, often earlier, fires are still most often underestimated. Higher-resolution data eliminates this problem.

Dry or
Green Leafy

Biomass
High. Green leaves burned on trees have high CH4 EF. Leaf fall commences in mid dry season adding to the fuel load, altering fuel composition, increasing fuel connectivity while reducing airflow through the fuel bed affecting combustion.
People determine the timing of fires which has implications for whether leaves are burned green (early dry season) or dry (later dry season) High. Amounts of leaf litter vary by savanna type and season. While amounts of dry leaf litter have been estimated in some cases, green leaf combustion on standing trees and shrubs is relatively understudied.

Fire Type
High. Head-fires burn more intensively, with higher flame lengths scorch heights causing more of the tree canopy to burn.
People purposefully set backfires although fires can change direction and accidental fires may more often burn as head-fires.
High. There is a potentially large and unknown impact on emissions of methane. There are few studies of fire type for savannas but remotely sensed methods offers potential. People determine the time of day to set fires, most often late afternoon.
Low. Although rarely considered in the literature, satellite data can provide an estimate of fire timing.

Grass Type
Low. Perennial grasses hold moisture longer and are often taller than annuals. Grass types vary dramatically on savanna landscapes.
Human actions modify grass species over the short and long term. Perennials are highly valued, but are being replaced by annuals.
High. Few studies consider variations in grassy vegetation cover at fine resolution. Remotely sensed methods can potentially distinguish between annuals and perennials.

Type
Medium. Savannas are highly heterogeneous with varying levels of tree cover, which affects CH4 EF especially when small trees burn.
Woody vegetation type is partially a function of longterm human land use patterns of agriculture and grazing.
Medium. Improved remote sensing techniques can increase accuracy of vegetation mapping including canopy cover.