Precipitation and temperature drive woody dynamics in the grasslands of sub-Saharan Africa

Precipitation and temperature drive woody dynamics in the grasslands of sub-Saharan Africa | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Precipitation and temperature drive woody dynamics in the grasslands of sub-Saharan Africa Francesco D'Adamo, Rebecca Spake, James M. Bullock, Booker Ogutu, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3914432/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Understanding the drivers of ecosystem dynamics, and how responses vary spatially and temporally, is a critical challenge in the face of global change. Here we used structural equation models and remote sensing datasets to understand the direct and indirect effects of climatic, environmental, and anthropogenic variables on woody vegetation dynamics across four grasslands regions (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) of sub-Saharan Africa. We focus on African grasslands given their importance for biodiversity and ecosystem services, the lack of clarity on how they are likely to respond to changes in disturbances, and how such responses vary geographically. This is particularly the case of grass-dominated ecosystems – the focus of our study – rather than more mixed grass-tree regions (e.g., savannas). Rainfall (β = 0.148 [-0.111, 0.398]) and temperature (β = -0.109 [-0.387, 0.133]) showed consistently opposing effects on woody vegetation (average standardised regression coefficients and 95% confidence interval range during 1997–2016) across the four bioregions. Other variables showed overall negligible effects including, for instance, dry season rainfall, soil moisture and, notably, fire. Other relationships were more context-dependent. Only Greater Karoo and Kalahari drylands showed a negative relationship between woody vegetation and fire (β = -0.031 [-0.069, 0.021]). Similarly, in Madagascar we observed strong negative effects of temperature (β = -0.429 [-1.215, -0.259]) and population density (β = -0.354 [-0.651, -0.015]) on burned area, yet these did not result in any significant indirect effects on woody vegetation. Our results clarify the contribution of environmental and anthropogenic variables in controlling woody dynamics at broad spatiotemporal scales and reveal that the widely documented negative feedback between fire and woody vegetation does not necessarily apply across all African grasslands. Terrestrial Ecology grassland ecosystems grassland-forest transition rainfall-vegetation-fire feedback disturbances structural equation modelling sub-Saharan Africa woody dynamics Figures Figure 1 Figure 2 Figure 3 Figure 4 1 Introduction Understanding how global environmental change will affect natural ecosystem dynamics is a central challenge in ecology. Such understanding is complicated by the complexity of natural systems, with changes that are often context-dependent across space and time (Spake et al. 2022 ) due to cross-scale interactions among drivers (Spake et al. 2019 ). A key biome for which context dependency effect on ecosystem dynamics is poorly understood are grasslands, which are relatively understudied compared to, e.g., savannas and forests (Bardgett et al. 2021 ). Grasslands comprise ca. one-third of terrestrial ecosystems by area and are subject to widespread degradation (ca. half of global grasslands) that is leading to increasing concern for both biodiversity and human well-being (Bardgett et al. 2021 ; Strömberg and Staver 2022). This is particularly true for the African continent, where grasslands (together with savannas) constitute the dominant vegetation cover, host diverse endemic faunas and flora, and provide a multitude of material and non-material benefits to humans (Osborne et al. 2018 ). Here, dominant drivers of grassland loss and degradation include climate change, conversion to cropland, fire suppression, and overgrazing (Stevens et al. 2022 ). In addition, cross-scale interactions among these drivers often trigger other forms of degradation. Woody plant encroachment, for instance, impoverishes grassland ecosystems as woody vegetation replacing the grass layer pushes the system toward a secondary state (e.g., shrubland, woodland) (Nerlekar and Veldman 2020) with lower herbaceous plant diversity and palatability (Wieczorkowski and Lehmann 2022 ; Venter, Hawkins, and Cramer 2017). Characterising the drivers of grassland dynamics and the interactions over time and space is needed to improve predictions of grassland responses to global environmental change and ultimately limit degradation (Abdi et al. 2022 ). This is a complex task because of the many processes, both direct and indirect, linked to climate, soil, and disturbances (Pausas and Bond 2020) and given the functional differences between grasslands and savannas. Grasslands are defined as ecosystems dominated by indigenous or natural grasses and other herbaceous species, differing from savannas which are transitional woody-herbaceous systems between grassland and forest (Allen et al. 2011 ). Grass-dominated states generally exist below 650–1000 millimetres of annual rainfall (mm/yr) while forest ecosystems tend to dominate above 2500 mm/yr (Mayer and Khalyani 2011). Between these rainfall ranges, the grassland-forest transition is largely determined by feedback processes between rainfall, vegetation, and fire (Pausas and Bond 2020). Frequent fires and rainfall seasonality maintain an open system and promote the colonization of shade-intolerant flammable grasses, which in a positive feedback enhances fire and excludes woody plants (Wei, Wang, Brandt, et al. 2020 ). By contrast, shady, moist environments inhibit flammability and hamper grass growth, suppressing fire and promoting woody plant establishment (Charles-Dominique et al. 2018 ). Accordingly, the feedback between fire and woody vegetation is generally negative (i.e., each suppresses the other), with the likelihood of closed woody canopies increasing with rainfall. However, the interplay between rainfall, vegetation, and fire is more complex and often influenced by other drivers. Intensive rainfall before or after the main wet season, for instance, can promote woody over grassy vegetation (Brandt et al. 2019 ). The rise in atmospheric CO 2 is favouring the C 3 woody component over C 4 grasses (Osborne et al. 2018 ) but the concomitant increase in global temperature is increasing tree mortality by limiting photosynthesis via higher stomatal closure rates (McDowell et al. 2022 ). Shrub communities appear less constrained, and shrub encroachment may have increased because of a warming climate (D’Odorico, Okin, and Bestelmeyer 2012). At the same time, increasing temperature could also trigger fires either by facilitating seasonal fuel curing or as C 4 grasses have high-temperature photosynthesis and growth optima, which leads to rapid biomass build-up in warmer ecosystems (Lehmann et al. 2014 ). Yet the opposite has also been reported across Africa, as a recent sequence of hot years caused a decrease in fuel loads and hence fire events (Wei, Wang, Fu, et al. 2020 ). Further, rainfall can also increase fire likelihood by promoting grass production and hence fuel availability, yet the amount of rain that maximises fuel availability may vary with different rainfall accumulation periods prior to the fire season (van der Werf et al. 2008 ; Archibald et al. 2009 ). Anthropogenic actions may also increase (e.g., tree planting schemes) or decrease (e.g., deforestation) regional woody vegetation density (Hansen et al. 2013 ), and increase (via ignition) or decrease (e.g., fragmentation) fire activity (Alvarado et al. 2020 ). These examples highlight that the key drivers of woody dynamics on grasslands in Africa, and the degree to which their effects change across space and time, remain unclear. Part of this inconsistency may relate to the fact that most studies have assessed grasslands and savannas together, conflating any potentially different responses that these two ecosystems may have to changes in climate, disturbances, and their interactions. Here we quantify the relative importance of environmental and anthropogenic disturbances on woody vegetation dynamics across all major grasslands in sub-Saharan Africa. Our goal was to increase understanding of the conditions determining the likelihood that grasslands transition towards a woody state. To this end, we used continental-scale datasets to explore the degree to which different ecologically reasonable specifications of pertinent variables could explain woody vegetation dynamics across sub-Saharan grasslands. More specifically, we used structural equation models (SEMs) to test a series of a priori formulated hypotheses of both direct and indirect (i.e., mediated by fire) effects of environmental and anthropogenic disturbances on woody dynamics. Analyses were run separately for four bioregions (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) to account for the ecological heterogeneity that exists across Africa and the distinct socio-ecological histories that may exist among regions. We expected a dominant direct role of rainfall on woody vegetation and fire, consistent with previous studies. Similarly, we anticipated negative feedback between fire and woody vegetation and dry season rainfall to promote woody vegetation. We did not have any specific expectations regarding direct or indirect effects of temperature and population density on woody vegetation. We did not try to anticipate to what extent support for each hypothesis varied across space and time. 2 Materials and methods 2.1 Study area We defined grasslands by using the European Space Agency (ESA) Climate Change Initiative (CCI) land cover map (version 2.0.7), which adopts the United Nations Land Cover Classification System (UN-LCCS) and is available at 300 m spatial resolution (ESA 2017 ). From this product, we selected classes 110 (mosaic herbaceous cover > 50% and tree and shrub < 50%) and 130 (grassland), and excluded areas dominated by croplands, trees, or where the herbaceous cover is lower than 50%. By doing so, we produced a grassland mask that matches the classification of grassland of the Intergovernmental Panel on Climate Change (ESA 2017 ) and fits our interest in grass-dominated ecosystems (as opposed to more mixed grass-tree savanna regions). Given the ecological and climatic heterogeneity that exists across Africa, we opted to subdivide grasslands by applying the One Earth Bioregions framework (One Earth 2020 ). This framework aggregates similar Dinerstein et al. ( 2017 ) terrestrial ecoregions into larger-scale ecological systems that are therefore better suited for broad regional assessments (e.g., our study). The four One Earth Bioregions are Sahel Acacia savannas (hereafter Sahel grasslands for simplicity of terminology), Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar (Fig. 1 ). 2.2 Data Vegetation optical depth (VOD) describes the attenuation of the microwave signal by the vegetation layer (Meesters, DeJeu, and Owe 2005). It is proportional to the vegetation water content of aboveground vegetation, so that higher VOD values indicate high vegetation water content and more energy attenuation (e.g., dense vegetation), whereas lower VOD values refer to limited vegetation water content, little attenuation, and higher transmissivity (e.g., sparse vegetation). Compared to optical-based products, VOD is insensitive to atmospheric haze and dust, cloud cover, or sun illumination (Li et al. 2021 ). We used the VOD Climate Archive (VODCA) Ku band (18.70-19.35 GHz, 1987–2017, 25 km × 25 km) (Moesinger et al. 2020 ) to exploit the longest available time series, and took annual minimum values to reduce the effects of the green herbaceous layer and produce a VOD signal that is more representative of the general woody cover community (e.g., shrubs, small trees, large trees) (Brandt et al. 2017 ; D’Adamo et al. 2021 ). The VODCA dataset has been shown to agree well with other VOD, leaf area index, and vegetation continuous field global products (Moesinger et al. 2020 ). Burned area data were obtained from the Global Fire Emissions Database (GFED), version 4s (1997–2016, 25 km × 25 km). This product provides burned area from GFED4 complemented with the contribution of small fires (s), among other data (e.g., fire carbon, dry matter emissions) (van der Werf et al. 2017 ). Burned area represents a direct estimate of fire impacts on ecosystems and has the advantage of persisting on the land surface, thus preventing potential fire data gaps due to cloud and smoke cover spells (Andela et al. 2017 ). We first converted monthly burned area fraction (dimensionless) to monthly burned area (ha) using the ancillary grid map (m 2 ) that is embedded with the data and, second, we aggregated monthly burned area into annual sum composites. Daily rainfall data from Climate Hazards group Infrared Precipitation with Stations (CHIRPS v2.0) (1981–2023, 5 km × 5 km) (Funk et al. 2015 ) were used to produce annual sums (mm/yr) and to calculate the dry season rainfall following Liebmann et al. (2012) (Appendix S1). Quality assessments based on mean absolute error have shown how CHIRPS data perform better than many other products including or not including gauge stations as well as reanalysis data (Funk et al. 2015 ). Temperature data were obtained from the Climatologies at High resolution for the Earth’s Land Surface Areas (CHELSA v2.1) (1979–2019, 1 km × 1 km) (Karger et al. 2017 , 2021 ). Monthly data from daily means of synoptic hourly temperature at 2 metres were converted from Kelvin to Celsius and then averaged to produce annual mean composites. Soil moisture data were taken from the ESA CCI program (Dorigo et al. 2018 ). Produced as active, passive, and active-passive merged product, we used the merged product (version v04.2, 1978–2016, 25 km × 25 km) as it combines the advantages of active (better for averagely vegetated areas) and passive (preferable over sparse vegetation and at distinguishing between wet and dry soils) observations (Dorigo et al. 2017 ). We created annual soil moisture composites by summing only good quality daily data (i.e., pixels without issues) each year (m 3 m − 3 ). Soil moisture data were used to assess the role of soil moisture on woody vegetation and the effect of moisture availability on fire (Lehmann et al. 2014 ). The gridded Population of the World (GPWv4) dataset provides spatially explicit global distribution of the human population at ca. 1 km spatial resolution (CIESIN 2018). GPWv4 data do not rely on ancillary data sources (e.g., land cover, vegetation indices), thus precluding potential problems of collinearity with VOD (Brandt et al. 2017 ). We used population density data (persons/km 2 ) adjusted to the 2015 revision of the United Nations World Population Prospects to investigate the effect of people on woody vegetation and fire dynamics (Archibald et al. 2010 ; Brandt et al. 2017 ). A continuous population density time series was produced using the available data (i.e., 2000, 2005, 2010, 2015, and 2020) to interpolate missing years (Abel et al. 2020 ). 2.3 Statistical analysis Structural equation modelling (SEM) is a probabilistic tool that allows the inclusion of multiple dependent and independent variables with different distributions in a single framework (Lefcheck 2016 ). Unlike standard regression, SEM is capable of evaluating both direct and indirect effects among variables, which is typical within complex natural ecosystems (Fan et al. 2016 ). Here we used a SEM approach as it is a useful tool for testing causal relationships hypothesised from theory and knowledge (Lehmann et al. 2014 ). Our statistical modelling workflow consisted of four steps: 1 - Hypotheses. We reviewed the literature on direct and indirect relationships among variables in the grassland-forest transition (Fig. 2 a) (see Introduction). Based on this, we created an initial most plausible hypothesis (represented by a causal structure), which we then modified to both simpler and more complex hypotheses to account for all ecologically reasonable combinations among variables (corresponding to alternative causal structures) (Fig. 2 b). This exercise also allowed us to assess whether simpler SEMs are able to explain grassland dynamics than more complex ones. We ended up with a total of 37 plausible hypotheses, each of which was formalized as an individual causal SEM structure (Appendix S2). All SEMs have VOD as the main response variable, with the effect of the other variables quantified directly and/or indirectly via burned area. Two exceptions were dry season rainfall and previous year VOD, which we deemed as reasonable predictors of VOD only (e.g., Brandt et al. 2019 ) and were not included as indirect predictors of VOD to avoid any circular causality problems (Bowman, Perry, and Marston 2015). Annual rainfall was specified as a predictor of both VOD and burned area. We acknowledge that fire regimes may be better explained by rainfall preceding the fire season, yet we only used annual rainfall as it was strongly collinear (Pearson’s r > 0.774) with rainfall metrics accumulated over 6, 12, 18, and 24 months before the fire season (Appendix S3). 2 - Model runs. All SEMs were run using the psem function in the R package ‘piecewiseSEM’ (Lefcheck, Byrnes, and James 2020). Rather than fitting continent-wide models with all years, separate models were fit for each bioregion (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) and year (i.e., 1997–2016) to be able to account for both the spatial and temporal dependency structures of the data (e.g., Walker et al. 2020 ) and, importantly, to assess to what extent support for each hypothesis varied across space and time. Preliminary model runs revealed strong spatial autocorrelation (SAC) in the residuals, indicated by statistically significant (p < 0.001) Moran’s I values and correlograms (Dormann et al. 2007 ). To account for SAC, SEMs were embedded with spatial autoregressive error models produced with the errorsarlm function in the R package ‘spatialreg’ (Bivand, Piras, et al. 2022 ). Errorsarlm models handle SAC through spatial weighting matrices computed as the Euclidean distance between neighbouring sites (i.e., pixels) (Bivand and Wong 2018). We calculated spatial weighting matrices using the dnearneigh function from the R package ‘spdep’ (Bivand, Altman, et al. 2022 ), setting the lower and upper distance bounds at 0 km and 39 km, respectively (39 km is the distance that allows all pixels to be linked to at least another pixel, including some farther pixels at the edge of the bioregions). The upper distance bound of 39 km did not vary significantly across bioregions as the data are regularly gridded (Appendix S4). SEM path coefficients were calculated as standardised regression coefficients (β) to enhance comparability across responses of different units. The indirect effect of a variable on VOD was calculated by multiplying the standardized regression coefficients of the two respective paths. 3 - Model selection. We used Fisher’s C and Chi-squared (χ 2 ) goodness of fit measures to identify SEM specifications that best reproduce the relationships among the variables in the sample data (hereafter well-fitted SEMs). Fisher’s C is calculated by summing the p-values of all unspecified paths, and p > 0.05 suggests that the model is well structured (i.e., no paths are missing) (Haynes et al. 2022; Shipley 2009 ). Similarly, χ 2 tests whether there is a discrepancy between the model-implied and observed covariance matrices, and also in this case p > 0.05 is recommended (Fan et al. 2016 ). In each year and each bioregion, we therefore selected only SEMs showing p > 0.05 for Fisher’s C and χ 2 (complete Fisher’s C and χ 2 statistics, p-value, and degrees of freedom for each statistical test are provided in Appendix S5). The number of well-fitted SEMs was then refined by calculating the difference in the Akaike Information Criterion (ΔAIC) between each SEM and the SEM with the lowest AIC (Cade 2015 ), and selecting only SEMs with ΔAIC < 3 (hereafter final SEMs) (Burnham, Anderson, and Huyvaert 2011). We present the results for each bioregion by natural model averaging the final SEMs β path coefficients each year, and then averaging over 1997–2016. We took this approach as the values of the statistically significant standardised regression coefficients did not change appreciably over time (coefficient of variations < 5%). Full result tables are provided in Appendix S5. 4 - Model evaluation. Although final SEMs were selected via satisfactory Fisher’s C, χ 2 , and ΔAIC values, we further evaluated them by calculating Nagelkerke r 2 and plotting residuals against fitted values for each variable included and not included in the model, latitude and longitude, and time (Zuur and Ieno 2016) (Appendix S6). This step is important to assess the goodness of fit of any saturated SEMs (i.e., all variables are linked), as these have no degrees of freedom (Cortina et al. 2017 ). The four-step statistical modelling workflow is shown in Fig. 3 . The preprocessing we computed before the statistical analysis is detailed in Appendix S7. All analyses were performed within the R environment (R Core Team 2021 ). 3 Results From the initial 37 SEM structures, the mean number of final SEMs (i.e., models with non-significant χ 2 and Fisher’s C statistics and ΔAIC < 3) per year varied from 5.3 in Sahel grasslands, 5.7 in Greater Karoo and Kalahari drylands, 4.7 in Southeast African subtropical grasslands, to 2.4 in Madagascar, with a maximum of eleven (2009 in Greater Karoo and Kalahari drylands and 2005 in Southeast African subtropical grasslands) and a minimum of one (nine different years across all bioregions) (Appendix S5: Tables S1, S3, S5, S7). In other words, only ca. 12.2% of our hypothesized SEM causal structures could satisfy the criteria determining final SEMs each year. Further, our simple SEM structures were systematically excluded from the final SEMs, meaning that the observed direct and indirect relationships resulted from more complex structures (except for SEM 3 and SEM 8) (Appendixes S2). In Sahel grasslands, rainfall showed a positive direct effect on both VOD (average β = 0.216 [95% confidence interval range during 1997–2016 {CI}: 0.100, 0.398]) and burned area (β = 0.256 [0.024, 0.537]) (Fig. 4 a). VOD showed a negative direct relationship with temperature (β = -0.216 [-0.387, -0.050]) while being largely unaffected by other variables. As expected, previous year VOD was a strong predictor of VOD (β = 0.511 [0.385, 0.758]) (Fig. 4 a and Appendix S5: Table S2). The Greater Karoo and Kalahari drylands and Southeast African subtropical grasslands showed overall similar relationships regarding the effects of rainfall on VOD and burned area, temperature on VOD, and previous year VOD on VOD (path coefficients of the remaining relationships were largely negligible) (Figs. 4 b and 4 c and Appendix S5: Tables S4 and S6). In addition, in the Greater Karoo and Kalahari drylands, we also found a weak negative direct effect (β = -0.031 [-0.069, 0.021]) of burned area on VOD. This is likely because the Greater Karoo and Kalahari drylands is the only bioregion where VOD significantly increased during 1997–2016, which indicates a potential increase in woody vegetation and, therefore, stronger feedback with burned area (Appendix S8: Figure S2). Madagascar also featured a positive relationship between rainfall and both VOD (β = 0.126 [-0.069, 0.267]) and burned area (β = 0.208 [-0.063, 0.599]) and a negative effect of temperature on VOD (β = -0.037 [-0.137, 0.133]), yet here, we observed a strong negative effect of both temperature (β = -0.429 [-1.215, -0.250]) and population density (β = -0.354 [-0.651, -0.015]) on burned area (Fig. 4 d and Appendix S5: Table 8). Noticeably, both these variables showed a sharp change during 1997–2016 (Appendix S8: Figure S4). Three key features were shared by all bioregions. First, we did not observe any significant indirect effects (i.e., mediated by burned area) of the variables on VOD. In Sahel grasslands, for instance, the indirect effect of rainfall on VOD was much weaker (β = 0.256 × 0.012 = 0.0031) than its direct effect (β = 0.216 [0.100, 0.398]) (Fig. 4 a). This finding applied also in case of strong direct effects of variables on burned area (e.g., temperature and population density in Madagascar) (Fig. 4 d). Second, we noticed that the direct effects of precipitation and temperature on VOD had similar magnitudes but opposite signs. Given the marginal role of other variables, these diametrically opposite effects might have constrained changes in VOD, except for the Greater Karoo and Kalahari drylands (Appendix S8). Finally, all final SEMs showed both satisfactory residual plots (Appendix S6) and explained a high degree of variation in VOD as indicated by high Nagelkerke r 2 values, i.e., r 2 = 0.88 in Sahel grasslands, r 2 = 0.96 in Greater Karoo and Kalahari drylands, r 2 = 0.84 in Southeast African subtropical grasslands, and r 2 = 0.94 in Madagascar (Figs. 4 a-d and Appendix S5: Tables S1, S3, S5, S7). The high Nagelkerke r 2 values are likely related to the strong spatial autocorrelation component observed in the gridded data and, hence, to the high explanatory power of the spatial weighting distances in our models. 4 Discussion Ecological systems are complex because of the many interactions among biotic and abiotic components that occur at different spatiotemporal scales. As such, scientists tend to characterize ecological relationships as context-dependent (Spake et al. 2023 ) or with narrative descriptions of relationships (Zellmer, Allen, and Kesseboehmer 2006). Our results, however, help disentangle this complexity for grassland systems by clarifying the role of temperature, rainfall, fire, and population density on woody dynamics at broad spatiotemporal scales. In addition, they also provide empirical evidence of complexity. The low number of final SEMs each year, for instance, indicates that only specific combinations of variables yield meaningful explanations of woody dynamics. Further, final SEMs are systematically described by more complex structures, suggesting that simple models including only woody vegetation, burned area, and one or two other variables are not able to characterize the ecological relationships in time and space. Yet complexity does not mean that ecosystems have no common features. Rainfall controls both woody vegetation and burned area, which is expected as water availability is known to be the key resource for both plant growth (Ogutu, D’Adamo, and Dash 2021; Sankaran et al. 2005 ) and fuel availability (Lehmann et al. 2014 ; Staver, Botha, and Hedin 2017) in arid and semi-arid regions of Africa. However, and contradicting our initial hypothesis, the amount of dry season rainfall consistently did not predict woody vegetation dynamics, suggesting that the establishment of woody plants in these bioregions is more related to rainfall totals (García Criado et al. 2020 ) than rainfall timing (Brandt et al. 2019 ). In contrast to rainfall, temperature showed a consistent negative effect on woody vegetation. While this is unsurprising as rainy days are generally colder, climate change-driven rising aridity may cause woody plants to be more prone to hydraulic failure (Abel et al. 2023 ). Importantly, precipitation and temperature showed overall comparable but diametrically opposite effects on woody vegetation, which might have contributed to the prevention of encroachment during 1997–2016 (i.e., no significant changes in VOD during 1997–2016) (see Appendix 8). In turn, these results indicate that any future imbalances between these two variables might push the system toward a woody state. There is evidence that this may already be the case for the Greater Karoo and Kalahari drylands, which is the sole bioregion with significant positive trends in VOD (i.e., woody encroachment), and where weaker effects of temperature might have favoured rainfall-driven plant proliferation (see Appendix S8: Figure S2). This increase in VOD likely explains why the Greater Karoo and Kalahari drylands is the only bioregion where we observed the expected negative feedback between woody vegetation and fire, i.e., an increase in woody plants suppressed fire activity (Andela et al. 2017 ). As such, it seems plausible that within the grasslands of sub-Saharan Africa (as opposed to grasslands and savanna together) the negative feedback between woody vegetation and fire is not pronounced and can only be detected at higher VOD. Madagascar also displayed a few distinct features, i.e., the effect of temperature and human population density on fire (see Appendix S8: Figure S4). This may relate not only to the different geographical setting (i.e., island vs. continental), but also to widespread human activities in the country (Ralimanana et al. 2022 ). Unsustainable agriculture and overexploitation, for instance, are likely explanations for our results, as fragmented landscapes reduce fuel connectivity and, therefore, fire spread (Bowman et al. 2020 ). Further, these anthropogenic activities (e.g., clearing woody vegetation) may explain why fire decline did not trigger woody encroachment (Phelps et al. 2022 ). The similarities in some relationships suggest that general rules of drivers of woody vegetation dynamics exist across the four bioregions investigated here. However, local changes or interactions for one or more variables may lead to context-dependent patterns that need to be accounted for ahead of effective grassland management (Bullock et al. 2021). However, it is also possible that some unexplained patterns may relate to important variables not investigated in this study (e.g., atmospheric CO 2 , grazing pressure, etc.) (Stevens et al. 2017). Another caveat to our findings is that the coarse spatial resolution of VOD data did not allow us to capture finer-grain processes and changes. For instance, a surprisingly large number of non-forest trees was observed in African drylands using 10 m to 0.5 m spatial resolution data (Reiner et al. 2023 ). Similarly, the contribution of small fires undetectable from coarse spatial resolution products has been shown to be decisive for appropriate burned area and fire emission estimations in Africa (e.g., up to + 80% of burned area detected with 20 m spatial resolution Sentinel 2 data as compared to MODIS data) (Ramo et al. 2021 ). These examples suggest that the consistent availability of long-term high-resolution data will be essential to better understand finer-grain woody and fire dynamics and, ultimately, grassland distribution across sub-Saharan Africa. Declarations Acknowledgements We thank the groups that produced and shared the datasets used here. We thank J Lefcheck for his resources and guidance with the CRAN package ‘piecewiseSEM’, R Bivand for his help with the CRAN package ‘spdep’, and P Fibich for the fruitful discussions on errorsarlm models. This work was funded by the European Research Council Starting Grant SCALEFORES (515825101) granted to F Eigenbrod and the UKCEH National Capability project (06895) to J M Bullock. Conflict of interest statement The authors declare no conflicting interests. References Abdi, Abdulhakim M, Martin Brandt, Christin Abel, and Rasmus Fensholt. 2022. “Review Article Satellite Remote Sensing of Savannas: Current Status and Emerging Opportunities.” Journal of Remote Sensing, 2022:2022. https://doi.org/DOI: 10.34133/2022/9835284. Abel, Christin, Abdulhakim M Abdi, Torbern Tagesson, Stephanie Horion, and Rasmus Fensholt. 2023. “Contrasting Ecosystem Vegetation Response in Global Drylands under Drying and Wetting Conditions.” Global Change Biology, 29: 3954–69. https://doi.org/10.1111/gcb.16745. Abel, Christin, Stéphanie Horion, Torbern Tagesson, Wanda De Keersmaecker, Alistair W R Seddon, Abdulhakim M Abdi, and Rasmus Fensholt. 2020. “The Human–Environment Nexus and Vegetation–Rainfall Sensitivity in Tropical Drylands.” Nature Sustainability, 4: 25–32. https://doi.org/10.1038/s41893-020-00597-z. Allen, Vivien G, Caterina Batello, Elbio J Berretta, John Hodgson, Mort Kothmann, Xianglin Li, John McIvor, et al. 2011. “An International Terminology for Grazing Lands and Grazing Animals.” Grass and Forage Science 66: 2–28. https://doi.org/10.1111/j.1365-2494.2010.00780.x. Alvarado, Swanni T, Niels Andela, Thiago S F Silva, and Sally Archibald. 2020. “Thresholds of Fire Response to Moisture and Fuel Load Differ between Tropical Savannas and Grasslands across Continents.” Global Ecology and Biogeography 29: 331–44. https://doi.org/10.1111/geb.13034. Andela, Niels, Douglas C Morton, Luis Giglio, Yang Chen, Guido R van der Werf, Prasad S Kasibhatla, Rurth S DeFries, et al. 2017. “A Human-Driven Decline in Global Burned Area.” Science 356: 1356–62. https://doi.org/10.1126/science.aal4108. Archibald, Sally, Alecia Nickless, Nicolin Govender, Robert J Scholes, and Veiko Lehsten. 2010. “Climate and the Inter-Annual Variability of Fire in Southern Africa: A Meta-Analysis Using Long-Term Field Data and Satellite-Derived Burnt Area Data.” Global Ecology and Biogeography 19: 794–809. https://doi.org/10.1111/j.1466-8238.2010.00568.x. Archibald, Sally, David P Roy, Brian W van Wilgen, and Robert J Scholes. 2009. “What Limits Fire? An Examination of Drivers of Burnt Area in Southern Africa.” Global Change Biology 15: 613–30. https://doi.org/10.1111/j.1365-2486.2008.01754.x. Bardgett, Richard D, James M Bullock, Sandra Lavorel, Peter Manning, Urs Schaffner, Nicholas Ostle, Mathilde Chomel, et al. 2021. “Combatting Global Grassland Degradation.” Nature Reviews Earth and Environment 2: 720–35. https://doi.org/10.1038/s43017-021-00207-2. Bivand, Roger, Micah Altman, Luc Anselin, Renato Assunção, Olaf Berke, Guillaume Blanchet, Marilia Carvalho, et al. 2022. “Package ‘Spdep.’” CRAN. Bivand, Roger, Gianfranco Piras, Luc Anselin, Andrew Bernat, Eric Blankmeyer, Yongwan Chun, Virgilio Gómez-Rubio, et al. 2022. “Package ‘Spatialreg.’” CRAN. Bivand, Roger S, and David W Wong. 2018. “Comparing Implementations of Global and Local Indicators of Spatial Association.” Test 27: 716–48. https://doi.org/10.1007/s11749-018-0599-x. Bowman, David M J S, Crystal A Kolden, John T Abatzoglou, Fay H Johnston, Guido R van der Werf, and Mike Flannigan. 2020. “Vegetation Fires in the Anthropocene.” Nature Reviews Earth and Environment 1: 500–15. https://doi.org/10.1038/s43017-020-0085-3. Bowman, David M J S, George L W Perry, and J B Marston. 2015. “Feedbacks and Landscape-Level Vegetation Dynamics.” Trends in Ecology and Evolution 30: 255–60. https://doi.org/10.1016/j.tree.2015.03.005. Brandt, Martin, Pierre Hiernaux, Kjeld Rasmussen, Compton J Tucker, Jean-Pierre Wigneron, Abdoul Aziz Diouf, Stefanie M Herrmann, et al. 2019. “Changes in Rainfall Distribution Promote Woody Foliage Production in the Sahel.” Communications Biology 2: 133. https://doi.org/10.1038/s42003-019-0383-9. Brandt, Martin, Kjeld Rasmussen, Josep Peñuelas, Feng Tian, Guy Schurgers, Aleixandre Verger, Ole Mertz, John R B Palmer, and Rasmus Fensholt. 2017. “Human Population Growth Offsets Climate-Driven Increase in Woody Vegetation in Sub-Saharan Africa.” Nature Ecology & Evolution 1: 0081. https://doi.org/10.1038/s41559-017-0081. Burnham, Kenneth P, David R Anderson, and Kathryn P Huyvaert. 2011. “AIC Model Selection and Multimodel Inference in Behavioral Ecology: Some Background, Observations, and Comparisons.” Behavioral Ecology and Sociobiology 65: 23–35. https://doi.org/10.1007/s00265-010-1029-6. Cade, Brian S. 2015. “Model Averaging and Muddled Multimodel Inferences.” Ecology 96: 2370–82. https://doi.org/10.1890/14-1639.1. Charles-Dominique, Tristan, Guy F Midgley, Kyle W Tomlinson, and William J Bond. 2018. “Steal the Light: Shade vs Fire Adapted Vegetation in Forest–Savanna Mosaics.” New Phytologist 218: 1419–29. https://doi.org/10.1111/nph.15117. CIESIN. 2018. "Gridded Population of the World, Version 4 (GPWv4): Population Count Adjusted to Match 2015 Revision of UN WPP Country Totals, Revision 11". Palisades, New York: NASA Socioeconomic Data and Applications Center (SEDAC). https://doi.org/10.7927/H4PN93PB. Cortina, Jose M, Jennifer P Green, Kathleen R Keeler, and Robert J Vandenberg. 2017. “Degrees of Freedom in SEM: Are We Testing the Models That We Claim to Test?” Organizational Research Methods 20: 350–78. https://doi.org/10.1177/1094428116676345. D’Adamo, Francesco, Booker Ogutu, Martin Brandt, Guy Schurgers, and Jadunandan Dash. 2021. “Climatic and Non-Climatic Vegetation Cover Changes in the Rangelands of Africa.” Global and Planetary Change 202: 103516. https://doi.org/10.1016/j.gloplacha.2021.103516. Dinerstein, Eric, David Olson, Anup Joshi, Carly Vynne, Neil D Burgess, Eric Wikramanayake, Nathan Hahn, et al. 2017. “An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm.” BioScience 67: 534–45. https://doi.org/10.1093/biosci/bix014. D’Odorico, Paolo, Gregory S Okin, and Brandon T Bestelmeyer. 2012. “A Synthetic Review of Feedbacks and Drivers of Shrub Encroachment in Arid Grasslands.” Ecohydrology 5: 520–30. https://doi.org/10.1002/eco.259. Dorigo, Wouter, Wolfgang Wagner, Clement Albergel, Franziska Albrecht, Gianpaolo Balsamo, Luca Brocca, Daniel Chung, et al. 2017. “ESA CCI Soil Moisture for Improved Earth System Understanding: State-of-the-Art and Future Directions.” Remote Sensing of Environment 203: 185–215. https://doi.org/10.1016/j.rse.2017.07.001. Dorigo, Wouter, Wolfgang Wagner, Gruber, Alexander, Scanlon, Tracy; Hahn, Sebastian; Kidd, Richard, Paulik, C, Reimer, Christoph; Van der Schalie, Robin; De Jeu, Richard (2018): "ESA Soil Moisture Climate Change Initiative: Combined Product, Version 04.2." Centre for Environmental Data Analysis. https://doi.org/10.5285/3a8a94c3fa464d68b6d70df291afd457. Dormann, Carsten F, Jana M. McPherson, Miguel B Araújo, Roger Bivand, Janine Bolliger, Gudrun Carl, Richard G Davies, et al. 2007. “Methods to Account for Spatial Autocorrelation in the Analysis of Species Distributional Data: A Review.” Ecography 30: 609–28. https://doi.org/10.1111/j.2007.0906-7590.05171.x. ESA. 2017. “Land Cover CCI: Product User Guide Version 2.0.” UCL-Geomatics (Belgium). Fan, Yi, Jiquan Chen, Gabriela Shirkey, Ranjeet John, Susie R Wu, Hogeun Park, and Changliang Shao. 2016. “Applications of Structural Equation Modeling (SEM) in Ecological Studies: An Updated Review.” Ecological Processes 5. https://doi.org/10.1186/s13717-016-0063-3. Funk, Chris, Pete Peterson, Martin Landsfeld, Diego Pedreros, James Verdin, Shraddhanand Shukla, Gregory Husak, et al. 2015. “The Climate Hazards Infrared Precipitation with Stations—a New Environmental Record for Monitoring Extremes.” Scientific Data 2: 150066. https://doi.org/10.1038/sdata.2015.66. García Criado, Mariana, Isla H Myers-Smith, Anne D Bjorkman, Caroline E R Lehmann, and Nicola Stevens. 2020. “Woody Plant Encroachment Intensifies under Climate Change across Tundra and Savanna Biomes.” Global Ecology and Biogeography 29: 925–43. https://doi.org/10.1111/geb.13072. Hansen, Matthew C, Peter V Potapov, Rebecca Moore, Matt Hancher, Svetlana A Turubanova, Alexandra Tyukavina, et al. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342: 850–53. https://doi.org/10.1126/science.1244693. Haynes, Kyle J, Andrew M Liebhold, Jonathan S Lefcheck, Randall S Morin, and Guiming Wang. 2022. “Climate Affects the Outbreaks of a Forest Defoliator Indirectly through Its Tree Hosts.” Oecologia. https://doi.org/10.1007/s00442-022-05123-w. James M Bullock, Elisa Fuentes-Montemayor, Ben McCarthy, Kirsty Park, Rosie S Hails, Ben A Woodcock, Kevin Watts, Ron Corstanje, and Jim Harris. 2021. “Future Restoration Should Enhance Ecological Complexity and Emergent Properties at Multiple Scales.” Ecography, 1–11. https://doi.org/10.1111/ecog.05780. Karger, Dirk N, Olaf Conrad, Jurgen Böhner, Tobias Kawohl, Holger Kreft, Rodrigo W Soria-Auza, Niklaus Zimmermann, Peter Linder, and Michael Kessler. 2017. Climatologies at high resolution for the earth’s land surface areas. Scientific Data 4, 170122. https://doi.org/10.1038/sdata.2017.122. Karger, Dirk N, Olaf Conrad, Jurgen Böhner, Tobias Kawohl, Holger Kreft, Rodrigo W Soria-Auza, Niklaus Zimmermann, Peter Linder, and Michael Kessler. 2021. “Climatologies at high resolution for the earth land surface areas.” EnviDat. https://doi.org/10.16904/envidat.228.v2.1. Lefcheck, Jonathan S, Jarrett Byrnes, and Grace James. 2020. “Package ‘PiecewiseSEM.’” CRAN. Lefcheck, Jonathan S. 2016. “PiecewiseSEM: Piecewise Structural Equation Modelling in r for Ecology, Evolution, and Systematics.” Methods in Ecology and Evolution 7: 573–79. https://doi.org/10.1111/2041-210X.12512. Lehmann, Caroline E R, T M Anderson, M Sankaran, S I Higgins, Sally Archibald, W A Hoffmann, N P Hanan, et al. 2014. “Savanna Vegetation-Fire-Climate Relationships Differ Among Continents.” Science 343: 548–52. https://doi.org/10.1126/science.1247355. Li, Xiaojun, Jean Pierre Wigneron, Frédéric Frappart, Lei Fan, Philippe Ciais, Rasmus Fensholt, Dara Entekhabi, et al. 2021. “Global-Scale Assessment and Inter-Comparison of Recently Developed/Reprocessed Microwave Satellite Vegetation Optical Depth Products.” Remote Sensing of Environment 253:112208. https://doi.org/10.1016/j.rse.2020.112208. Mayer, Audrey L, and Azad H Khalyani. 2011. “Grass Trumps Trees with Fire.” Science 334: 188–89. https://doi.org/10.1126/science.1213908. McDowell, Nate G, Gerard Sapes, Alexandria Pivovaroff, Henry D Adams, Craig D Allen, William R L Anderegg, Matthias Arend, et al. 2022. “Mechanisms of Woody-Plant Mortality under Rising Drought, CO2 and Vapour Pressure Deficit.” Nature Reviews Earth & Environment 3: 294–308. https://doi.org/10.1038/s43017-022-00272-1. Meesters, Antoon G C, Richard A M DeJeu, and Manfred Owe. 2005. “Analytical Derivation of the Vegetation Optical Depth From the Microwave Polarization Difference Index.” IEEE Geoscience and Remote Sensing Letters 2: 121–23. https://doi.org/10.1109/LGRS.2005.843983. Moesinger, Leander, Wouter Dorigo, Richard De Jeu, Robin Van Der Schalie, Tracy Scanlon, Irene Teubner, and Matthias Forkel. 2020. “The Global Long-Term Microwave Vegetation Optical Depth Climate Archive (VODCA).” Earth System Science Data 12: 177–96. https://doi.org/10.5194/essd-12-177-2020. Nerlekar, Ashish N, and Joseph W Veldman. 2020. “High Plant Diversity and Slow Assembly of Old-Growth Grasslands.” Proceedings of the National Academy of Sciences of the United States of America 117: 18550–56. https://doi.org/10.1073/pnas.1922266117. Ogutu, Booker O, Francesco D’Adamo, and Jadunandan Dash. 2021. “Impact of Vegetation Greening on Carbon and Water Cycle in the African Sahel-Sudano-Guinean Region.” Global and Planetary Change 202: 103524. https://doi.org/10.1016/j.gloplacha.2021.103524. One Earth. 2020. “Bioregions.” 2020. URL https://www.oneearth.org/. Osborne, Colin P, Tristan Charles-Dominique, Nicola Stevens, William J Bond, Guy Midgley, and Caroline E R Lehmann. 2018. “Human Impacts in African Savannas Are Mediated by Plant Functional Traits.” New Phytologist 220: 10–24. https://doi.org/10.1111/nph.15236. Pausas, Juli G, and William J Bond. 2020. “Alternative Biome States in Terrestrial Ecosystems.” Trends in Plant Science 25: 250–63. https://doi.org/10.1016/j.tplants.2019.11.003. Phelps, Leanne N, Niels Andela, Mathieu Gravey, Dylan S Davis, Christian A Kull, Kristina Douglass, and Caroline E R Lehmann. 2022. “Madagascar’s Fire Regimes Challenge Global Assumptions about Landscape Degradation.” Global Change Biology, 1–17. https://doi.org/10.1111/gcb.16206. R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/. Ralimanana, Hélène, Allison L Perrigo, Rhian J Smith, James S Borrell, Søren Faurby, Mamy Tiana Rajaonah, Tianjanahary Randriamboavonjy, et al. 2022. “Madagascar’s Extraordinary Biodiversity: Threats and Opportunities.” Science 378: eadf1466. https://doi.org/10.1126/science.adf1466. Ramo, Ruben, Ekhi Roteta, Ioannis Bistinas, Dave van Wees, Aitor Bastarrika, Emilio Chuvieco, and Guido R van der Werf. 2021. “African Burned Area and Fire Carbon Emissions Are Strongly Impacted by Small Fires Undetected by Coarse Resolution Satellite Data.” Proceedings of the National Academy of Sciences of the United States of America 118: 1–7. https://doi.org/10.1073/pnas.2011160118. Reiner, Florian, Martin Brandt, Xiaoye Tong, David Skole, Ankit Kariryaa, Philippe Ciais, Andrew Davies, et al. 2023. “More than One Quarter of Africa’s Tree Cover Is Found Outside Areas Previously Classified as Forest.” Nature Communications 14: 2258. https://doi.org/10.1038/s41467-023-37880-4. Sankaran, Mahesh, Niall P Hanan, Robert J Scholes, Jayashree Ratnam, David J Augustine, Brian S Cade, Jacques Gignoux, et al. 2005. “Determinants of Woody Cover in African Savannas.” Nature 438: 846–49. https://doi.org/10.1038/nature04070. Shipley, Bill. 2009. “Confirmatory Path Analysis in a Generalized Multilevel Context.” Ecology 90: 363–68. https://doi.org/10.1890/08-1034.1. Spake, Rebecca, Chloe Bellamy, Laura J Graham, Kevin Watts, Tom Wilson, Lisa R Norton, Claire M Wood, Reto Schmucki, James M Bullock, and Felix Eigenbrod. 2019. “An Analytical Framework for Spatially Targeted Management of Natural Capital” Nature Sustainability 2: 90–97. https://doi.org/10.1038/s41893-019-0223-4. Spake, Rebecca, Diana E Bowler, Corey T Callaghan, Shane A Blowes, Patrick C Doncaster, Laura H Antão, Shinichi Nakagawa, Richard McElreath, and Jonathan M Chase. 2023. “Understanding ‘It Depends’ in Ecology: A Guide to Hypothesising, Visualising and Interpreting Statistical Interactions.” Biological Reviews, brv.12939. https://doi.org/10.1111/brv.12939. Spake, Rebecca, Rose E O’Dea, Shinichi Nakagawa, Patrick C Doncaster, Masahiro Ryo, Corey T Callaghan, and James M Bullock. 2022. “Improving Quantitative Synthesis to Achieve Generality in Ecology.” Nature Ecology & Evolution 6: 1818–28. https://doi.org/10.1038/s41559-022-01891-z. Staver, Carla A, Judith Botha, and Lars Hedin. 2017. “Soils and Fire Jointly Determine Vegetation Structure in an African Savanna.” New Phytologist 216: 1151–60. https://doi.org/10.1111/nph.14738. Stevens, Nicola, William Bond, Angelica Feurdean, and Caroline E R Lehmann. 2022. “Grassy Ecosystems in the Anthropocene.” Annual Review of Environment and Resources 47, 261–289. https://doi.org/10.1146/annurev-environ-112420-015211. Strömberg, Caroline A E, and Carla A Staver. 2022. “The History and Challenge of Grassy Biomes.” Science 377: 592–93. https://doi.org/10.1126/science.add1347. Venter, Zander S, Heidi Jayne Hawkins, and Michael D Cramer. 2017. “Implications of Historical Interactions between Herbivory and Fire for Rangeland Management in African Savannas.” Ecosphere 8: 1–14. https://doi.org/10.1002/ecs2.1946. Walker, Xanthe J, Brendan M Rogers, Sander Veraverbeke, Jill F Johnstone, Jennifer L Baltzer, Kirsten Barrett, Laura Bourgeau-Chavez, et al. 2020. “Fuel Availability Not Fire Weather Controls Boreal Wildfire Severity and Carbon Emissions.” Nature Climate Change 10: 1130–36. https://doi.org/10.1038/s41558-020-00920-8. Wei, Fangli, Shuai Wang, Martin Brandt, Bojie Fu, Michael E Meadows, Lixin Wang, Lanhui Wang, Xiaowei Tong, and Rasmus Fensholt. 2020. “Responses and Feedbacks of African Dryland Ecosystems to Environmental Changes.” Current Opinion in Environmental Sustainability 48: 29–35. https://doi.org/10.1016/j.cosust.2020.09.004. Wei, Fangli, Shuai Wang, Bojie Fu, Martin Brandt, Naiqing Pan, Cong Wang, and Rasmus Fensholt. 2020. “Nonlinear Dynamics of Fires in Africa over Recent Decades Controlled by Precipitation.” Global Change Biology 26: 4495–4505. https://doi.org/10.1111/gcb.15190. Werf, Guido R van der, James T Randerson, Louis Giglio, Nadine Gobron, and A J Dolman. 2008. “Climate Controls on the Variability of Fires in the Tropics and Subtropics.” Global Biogeochemical Cycles 22: 1–13. https://doi.org/10.1029/2007GB003122. Werf, Guido R van der, James T Randerson, Louis Giglio, Thijs T Van Leeuwen, Yang Chen, Brendan M Rogers, Mingquan Mu, et al. 2017. “Global Fire Emissions Estimates during 1997-2016.” Earth System Science Data 9: 697–720. https://doi.org/10.5194/essd-9-697-2017. Wieczorkowski, Jakub D, and Caroline E R Lehmann. 2022. “Encroachment Diminishes Herbaceous Plant Diversity in Grassy Ecosystems Worldwide.” Global Change Biology, 1–15. https://doi.org/10.1111/gcb.16300. Zellmer, Amanda J, Timothy F H Allen, and Kirsten Kesseboehmer. 2006. “The Nature of Ecological Complexity: A Protocol for Building the Narrative.” Ecological Complexity 3: 171–82. https://doi.org/10.1016/j.ecocom.2006.06.002. Zuur, Alain F, and Elena N Ieno. 2016. “A Protocol for Conducting and Presenting Results of Regression-Type Analyses.” Methods in Ecology and Evolution 7: 636–45. https://doi.org/10.1111/2041-210X.12577. Additional Declarations The authors declare no competing interests. Supplementary Files DAdamoetalSupplementary.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3914432","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":270290446,"identity":"337d3cbb-aafb-4cf1-83ba-31dd29e85615","order_by":0,"name":"Francesco D'Adamo","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+UlEQVRIiWNgGAWjYBAC9gYGAwYeIIMfzLUBYgkCWngOQLVINoC4aTAtCURoMThAtBYG5o0f3tRskzM+fvbhgx8JDIn9sxsYP/z8gU8LW7HknGO3jc3OpBsb9gC1zLhzgFmyB48t9gw8BtI8bLcTt91gY5Pg/cGQ2HAjgY2BB6/DeIx/8/y7nbh5Bhv7zz9AW+YDtTD+wa/FTJq37XbiBgk2Nmag4YkbgFqY8drCzFZmObfvtrHEmTRmaZkECeONdw42S8uk4dHC3rz5xptvt+X4248xfnyTYCM773bzwY9vbHBrYWBG5YIihbEBj/pRMApGwSgYBcQAAPtqTMfxxdo3AAAAAElFTkSuQmCC","orcid":"","institution":"CREAF","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Francesco","middleName":"","lastName":"D'Adamo","suffix":""},{"id":270290447,"identity":"58628c32-2e2d-4fbb-b60d-e51cd2c66161","order_by":1,"name":"Rebecca Spake","email":"","orcid":"https://orcid.org/0000-0003-4671-2225","institution":"University of Reading","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Rebecca","middleName":"","lastName":"Spake","suffix":""},{"id":270290448,"identity":"12147e82-1937-4baa-9892-a29a06e3307c","order_by":2,"name":"James M. Bullock","email":"","orcid":"https://orcid.org/0000-0003-0529-4020","institution":"UK Centre for Ecology \u0026 Hydrology","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"James","middleName":"M.","lastName":"Bullock","suffix":""},{"id":270290449,"identity":"62443e3d-559f-4000-a6a7-e3955a75e442","order_by":3,"name":"Booker Ogutu","email":"","orcid":"https://orcid.org/0000-0002-1804-6205","institution":"University of Southampton","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Booker","middleName":"","lastName":"Ogutu","suffix":""},{"id":270290450,"identity":"bec03e1c-4da5-4b16-bbd6-7840303ba7a5","order_by":4,"name":"Jadunandan Dash","email":"","orcid":"https://orcid.org/0000-0002-5444-2109","institution":"University of Southampton","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Jadunandan","middleName":"","lastName":"Dash","suffix":""},{"id":270290451,"identity":"78e742c0-a54f-4910-a224-a7f0f785c2f8","order_by":5,"name":"Felix Eigenbrod","email":"","orcid":"https://orcid.org/0000-0001-8982-824X","institution":"University of Southampton","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Felix","middleName":"","lastName":"Eigenbrod","suffix":""}],"badges":[],"createdAt":"2024-01-31 16:11:04","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false,"coiExplicitlySet":false},"doi":"10.21203/rs.3.rs-3914432/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3914432/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":50512412,"identity":"dd7f6740-ecd6-464e-8c63-6c263c5ec097","added_by":"auto","created_at":"2024-02-01 16:20:19","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":134460,"visible":true,"origin":"","legend":"\u003cp\u003eGrassland extent as obtained from the ESA CCI land cover map (classes 110, i.e., herbaceous cover \u0026gt; 50% and tree and shrub \u0026lt; 50%, and 130, i.e., grassland) and indicated by the One Earth framework (One Earth, 2020). The four bioregions are Sahel grasslands (780 pixels), Greater Karoo and Kalahari drylands (287 pixels), Southeast African subtropical grasslands (271 pixels), and Madagascar (319 pixels).\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/1e491ddb2eb8bc945e1b0c62.jpeg"},{"id":50512413,"identity":"dc191079-ae3f-493a-bb41-b97c5dbcf906","added_by":"auto","created_at":"2024-02-01 16:20:19","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":211602,"visible":true,"origin":"","legend":"\u003cp\u003eThe key elements in the grassland-forest transition as commonly reported in the literature (see Introduction) and their described overall effect (positive (+), negative (–), or both (+/–)) on woody vegetation and fire (in grey are the variables excluded from the analysis as not available for our study regions during 1997-2016) (a). The initial hypothesis (left), and an example of a simpler (middle) and more complex (right) hypothesis as indicated by the corresponding SEM causal structure. Vegetation optical depth (VOD) is the main response variable. Burned area (BA) is both predictor and response variable. Other variables are annual rainfall (Rain), temperature (T), dry season rainfall (DSR), population density (PD), soil moisture (SM), and previous year VOD (VOD\u003csub\u003eyr-1\u003c/sub\u003e). The full list of the 37 SEM causal structures is reported in Appendix S2 (b).\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/b8e644ad0058bcfb2b9486e1.jpeg"},{"id":50512414,"identity":"d920fba6-3171-48a7-831f-cdd9575b7bcc","added_by":"auto","created_at":"2024-02-01 16:20:19","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":181979,"visible":true,"origin":"","legend":"\u003cp\u003eThe four-step statistical modelling workflow. 37 hypotheses were tested by fitting spatial autoregressive error models within a structural equation modelling (SEM) framework in each year and each bioregion. Well-fitted SEMs were identified by means of satisfactory Fisher’s C and χ\u003csup\u003e2 \u003c/sup\u003egoodness of fit measures. From well-fitted SEMs, we selected final SEMs based on ΔAIC value. Final SEMs were further evaluated using Nagelkerke r\u003csup\u003e2\u003c/sup\u003e and residual plots.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/a9f8f27d43c3ffffcf10a16b.jpeg"},{"id":50512415,"identity":"e1028c5e-d862-4bbf-b952-7491f1b57bd0","added_by":"auto","created_at":"2024-02-01 16:20:19","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":652213,"visible":true,"origin":"","legend":"\u003cp\u003eFinal structural equation models (SEM) of the direct and indirect effects (i.e., mediated by burned area) of burned area (BA), annual rainfall (Rain), temperature (T), dry season rainfall (DSR), population density (PD), soil moisture (SM), and previous year VOD (VOD\u003csub\u003eyr-1\u003c/sub\u003e) on vegetation optical depth (VOD) (left) and standardized regression coefficient with 95% CI values during 1997-2016 (right) for Sahel grasslands (a), Greater Karoo and Kalahari drylands (b), Southeast African subtropical grasslands (c), and Madagascar (d). The thickness of the arrows is scaled to the standardized regression coefficients, which are calculated by natural model averaging the final SEMs in each year and later during 1997-2016. Black represents positive effect and red represents negative effect. Cyan crosses indicate years in which all available final SEM structures did not include the path between the two variables. No CIs for years with only one available standardized regression coefficient. Rainfall and temperature direct effects on VOD are similar but opposite in sign (positive and negative, respectively) in all bioregions. Overall, no significant relationships were observed between burned area and VOD. The indirect effects of any variable on VOD were negligible. Appendix S5 shows full results for each bioregion, including the list of final SEMs, Fisher’s C, Fisher’s C p-value, Chi-squared, Chi-squared p-value, degrees of freedom, ΔAIC, Nagelkerke r\u003csup\u003e2\u003c/sup\u003e, standardised path regression coefficient, and 95% CI, standard deviation, and coefficient of variation of standardized regression coefficients for each year.\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/c98daac579d92893a3cc6452.jpeg"},{"id":50513625,"identity":"e041223f-ab4d-4b02-ad74-29baca3b9d67","added_by":"auto","created_at":"2024-02-01 16:28:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":713016,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/dc9447fe-cbea-4dda-8433-9afd1c577387.pdf"},{"id":50512417,"identity":"f700cea3-e2e6-434b-bd07-5606a39eb719","added_by":"auto","created_at":"2024-02-01 16:20:20","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":15064122,"visible":true,"origin":"","legend":"","description":"","filename":"DAdamoetalSupplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-3914432/v1/085033b95975f4693581ab79.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003ePrecipitation and temperature drive woody dynamics in the grasslands of sub-Saharan Africa\u003c/p\u003e","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eUnderstanding how global environmental change will affect natural ecosystem dynamics is a central challenge in ecology. Such understanding is complicated by the complexity of natural systems, with changes that are often context-dependent across space and time (Spake et al. \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) due to cross-scale interactions among drivers (Spake et al. \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). A key biome for which context dependency effect on ecosystem dynamics is poorly understood are grasslands, which are relatively understudied compared to, e.g., savannas and forests (Bardgett et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Grasslands comprise ca. one-third of terrestrial ecosystems by area and are subject to widespread degradation (ca. half of global grasslands) that is leading to increasing concern for both biodiversity and human well-being (Bardgett et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Str\u0026ouml;mberg and Staver 2022). This is particularly true for the African continent, where grasslands (together with savannas) constitute the dominant vegetation cover, host diverse endemic faunas and flora, and provide a multitude of material and non-material benefits to humans (Osborne et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Here, dominant drivers of grassland loss and degradation include climate change, conversion to cropland, fire suppression, and overgrazing (Stevens et al. \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). In addition, cross-scale interactions among these drivers often trigger other forms of degradation. Woody plant encroachment, for instance, impoverishes grassland ecosystems as woody vegetation replacing the grass layer pushes the system toward a secondary state (e.g., shrubland, woodland) (Nerlekar and Veldman 2020) with lower herbaceous plant diversity and palatability (Wieczorkowski and Lehmann \u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Venter, Hawkins, and Cramer 2017).\u003c/p\u003e \u003cp\u003eCharacterising the drivers of grassland dynamics and the interactions over time and space is needed to improve predictions of grassland responses to global environmental change and ultimately limit degradation (Abdi et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). This is a complex task because of the many processes, both direct and indirect, linked to climate, soil, and disturbances (Pausas and Bond 2020) and given the functional differences between grasslands and savannas. Grasslands are defined as ecosystems dominated by indigenous or natural grasses and other herbaceous species, differing from savannas which are transitional woody-herbaceous systems between grassland and forest (Allen et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Grass-dominated states generally exist below 650\u0026ndash;1000 millimetres of annual rainfall (mm/yr) while forest ecosystems tend to dominate above 2500 mm/yr (Mayer and Khalyani 2011). Between these rainfall ranges, the grassland-forest transition is largely determined by feedback processes between rainfall, vegetation, and fire (Pausas and Bond 2020). Frequent fires and rainfall seasonality maintain an open system and promote the colonization of shade-intolerant flammable grasses, which in a positive feedback enhances fire and excludes woody plants (Wei, Wang, Brandt, et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). By contrast, shady, moist environments inhibit flammability and hamper grass growth, suppressing fire and promoting woody plant establishment (Charles-Dominique et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). Accordingly, the feedback between fire and woody vegetation is generally negative (i.e., each suppresses the other), with the likelihood of closed woody canopies increasing with rainfall. However, the interplay between rainfall, vegetation, and fire is more complex and often influenced by other drivers. Intensive rainfall before or after the main wet season, for instance, can promote woody over grassy vegetation (Brandt et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The rise in atmospheric CO\u003csub\u003e2\u003c/sub\u003e is favouring the C\u003csub\u003e3\u003c/sub\u003e woody component over C\u003csub\u003e4\u003c/sub\u003e grasses (Osborne et al. \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) but the concomitant increase in global temperature is increasing tree mortality by limiting photosynthesis via higher stomatal closure rates (McDowell et al. \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Shrub communities appear less constrained, and shrub encroachment may have increased because of a warming climate (D\u0026rsquo;Odorico, Okin, and Bestelmeyer 2012). At the same time, increasing temperature could also trigger fires either by facilitating seasonal fuel curing or as C\u003csub\u003e4\u003c/sub\u003e grasses have high-temperature photosynthesis and growth optima, which leads to rapid biomass build-up in warmer ecosystems (Lehmann et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Yet the opposite has also been reported across Africa, as a recent sequence of hot years caused a decrease in fuel loads and hence fire events (Wei, Wang, Fu, et al. \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Further, rainfall can also increase fire likelihood by promoting grass production and hence fuel availability, yet the amount of rain that maximises fuel availability may vary with different rainfall accumulation periods prior to the fire season (van der Werf et al. \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; Archibald et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Anthropogenic actions may also increase (e.g., tree planting schemes) or decrease (e.g., deforestation) regional woody vegetation density (Hansen et al. \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), and increase (via ignition) or decrease (e.g., fragmentation) fire activity (Alvarado et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). These examples highlight that the key drivers of woody dynamics on grasslands in Africa, and the degree to which their effects change across space and time, remain unclear. Part of this inconsistency may relate to the fact that most studies have assessed grasslands and savannas together, conflating any potentially different responses that these two ecosystems may have to changes in climate, disturbances, and their interactions.\u003c/p\u003e \u003cp\u003eHere we quantify the relative importance of environmental and anthropogenic disturbances on woody vegetation dynamics across all major grasslands in sub-Saharan Africa. Our goal was to increase understanding of the conditions determining the likelihood that grasslands transition towards a woody state. To this end, we used continental-scale datasets to explore the degree to which different ecologically reasonable specifications of pertinent variables could explain woody vegetation dynamics across sub-Saharan grasslands. More specifically, we used structural equation models (SEMs) to test a series of a priori formulated hypotheses of both direct and indirect (i.e., mediated by fire) effects of environmental and anthropogenic disturbances on woody dynamics. Analyses were run separately for four bioregions (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) to account for the ecological heterogeneity that exists across Africa and the distinct socio-ecological histories that may exist among regions. We expected a dominant direct role of rainfall on woody vegetation and fire, consistent with previous studies. Similarly, we anticipated negative feedback between fire and woody vegetation and dry season rainfall to promote woody vegetation. We did not have any specific expectations regarding direct or indirect effects of temperature and population density on woody vegetation. We did not try to anticipate to what extent support for each hypothesis varied across space and time.\u003c/p\u003e"},{"header":"2 Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n\u003ch2\u003e2.1 Study area\u003c/h2\u003e\n\u003cp\u003eWe defined grasslands by using the European Space Agency (ESA) Climate Change Initiative (CCI) land cover map (version 2.0.7), which adopts the United Nations Land Cover Classification System (UN-LCCS) and is available at 300 m spatial resolution (ESA \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). From this product, we selected classes 110 (mosaic herbaceous cover\u0026thinsp;\u0026gt;\u0026thinsp;50% and tree and shrub\u0026thinsp;\u0026lt;\u0026thinsp;50%) and 130 (grassland), and excluded areas dominated by croplands, trees, or where the herbaceous cover is lower than 50%. By doing so, we produced a grassland mask that matches the classification of grassland of the Intergovernmental Panel on Climate Change (ESA \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e) and fits our interest in grass-dominated ecosystems (as opposed to more mixed grass-tree savanna regions). Given the ecological and climatic heterogeneity that exists across Africa, we opted to subdivide grasslands by applying the One Earth Bioregions framework (One Earth \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). This framework aggregates similar Dinerstein et al. (\u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e) terrestrial ecoregions into larger-scale ecological systems that are therefore better suited for broad regional assessments (e.g., our study). The four One Earth Bioregions are Sahel Acacia savannas (hereafter Sahel grasslands for simplicity of terminology), Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n\u003ch2\u003e2.2 Data\u003c/h2\u003e\n\u003cp\u003eVegetation optical depth (VOD) describes the attenuation of the microwave signal by the vegetation layer (Meesters, DeJeu, and Owe 2005). It is proportional to the vegetation water content of aboveground vegetation, so that higher VOD values indicate high vegetation water content and more energy attenuation (e.g., dense vegetation), whereas lower VOD values refer to limited vegetation water content, little attenuation, and higher transmissivity (e.g., sparse vegetation). Compared to optical-based products, VOD is insensitive to atmospheric haze and dust, cloud cover, or sun illumination (Li et al. \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). We used the VOD Climate Archive (VODCA) Ku band (18.70-19.35 GHz, 1987\u0026ndash;2017, 25 km \u0026times; 25 km) (Moesinger et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e) to exploit the longest available time series, and took annual minimum values to reduce the effects of the green herbaceous layer and produce a VOD signal that is more representative of the general woody cover community (e.g., shrubs, small trees, large trees) (Brandt et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e; D\u0026rsquo;Adamo et al. \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). The VODCA dataset has been shown to agree well with other VOD, leaf area index, and vegetation continuous field global products (Moesinger et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e). Burned area data were obtained from the Global Fire Emissions Database (GFED), version 4s (1997\u0026ndash;2016, 25 km \u0026times; 25 km). This product provides burned area from GFED4 complemented with the contribution of small fires (s), among other data (e.g., fire carbon, dry matter emissions) (van der Werf et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). Burned area represents a direct estimate of fire impacts on ecosystems and has the advantage of persisting on the land surface, thus preventing potential fire data gaps due to cloud and smoke cover spells (Andela et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). We first converted monthly burned area fraction (dimensionless) to monthly burned area (ha) using the ancillary grid map (m\u003csup\u003e2\u003c/sup\u003e) that is embedded with the data and, second, we aggregated monthly burned area into annual sum composites. Daily rainfall data from Climate Hazards group Infrared Precipitation with Stations (CHIRPS v2.0) (1981\u0026ndash;2023, 5 km \u0026times; 5 km) (Funk et al. \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e) were used to produce annual sums (mm/yr) and to calculate the dry season rainfall following Liebmann et al. (2012) (Appendix S1). Quality assessments based on mean absolute error have shown how CHIRPS data perform better than many other products including or not including gauge stations as well as reanalysis data (Funk et al. \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e). Temperature data were obtained from the Climatologies at High resolution for the Earth\u0026rsquo;s Land Surface Areas (CHELSA v2.1) (1979\u0026ndash;2019, 1 km \u0026times; 1 km) (Karger et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e). Monthly data from daily means of synoptic hourly temperature at 2 metres were converted from Kelvin to Celsius and then averaged to produce annual mean composites. Soil moisture data were taken from the ESA CCI program (Dorigo et al. \u003cspan class=\"CitationRef\"\u003e2018\u003c/span\u003e). Produced as active, passive, and active-passive merged product, we used the merged product (version v04.2, 1978\u0026ndash;2016, 25 km \u0026times; 25 km) as it combines the advantages of active (better for averagely vegetated areas) and passive (preferable over sparse vegetation and at distinguishing between wet and dry soils) observations (Dorigo et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). We created annual soil moisture composites by summing only good quality daily data (i.e., pixels without issues) each year (m\u003csup\u003e3\u003c/sup\u003e m\u003csup\u003e\u0026minus;\u0026thinsp;3\u003c/sup\u003e). Soil moisture data were used to assess the role of soil moisture on woody vegetation and the effect of moisture availability on fire (Lehmann et al. \u003cspan class=\"CitationRef\"\u003e2014\u003c/span\u003e). The gridded Population of the World (GPWv4) dataset provides spatially explicit global distribution of the human population at ca. 1 km spatial resolution (CIESIN 2018). GPWv4 data do not rely on ancillary data sources (e.g., land cover, vegetation indices), thus precluding potential problems of collinearity with VOD (Brandt et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). We used population density data (persons/km\u003csup\u003e2\u003c/sup\u003e) adjusted to the 2015 revision of the United Nations World Population Prospects to investigate the effect of people on woody vegetation and fire dynamics (Archibald et al. \u003cspan class=\"CitationRef\"\u003e2010\u003c/span\u003e; Brandt et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). A continuous population density time series was produced using the available data (i.e., 2000, 2005, 2010, 2015, and 2020) to interpolate missing years (Abel et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n\u003ch2\u003e2.3 Statistical analysis\u003c/h2\u003e\n\u003cp\u003eStructural equation modelling (SEM) is a probabilistic tool that allows the inclusion of multiple dependent and independent variables with different distributions in a single framework (Lefcheck \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e). Unlike standard regression, SEM is capable of evaluating both direct and indirect effects among variables, which is typical within complex natural ecosystems (Fan et al. \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e). Here we used a SEM approach as it is a useful tool for testing causal relationships hypothesised from theory and knowledge (Lehmann et al. \u003cspan class=\"CitationRef\"\u003e2014\u003c/span\u003e). Our statistical modelling workflow consisted of four steps:\u003c/p\u003e\n\u003cp\u003e\u003cspan class=\"Underline\"\u003e1 - Hypotheses.\u003c/span\u003e We reviewed the literature on direct and indirect relationships among variables in the grassland-forest transition (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ea) (see Introduction). Based on this, we created an initial most plausible hypothesis (represented by a causal structure), which we then modified to both simpler and more complex hypotheses to account for all ecologically reasonable combinations among variables (corresponding to alternative causal structures) (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb). This exercise also allowed us to assess whether simpler SEMs are able to explain grassland dynamics than more complex ones. We ended up with a total of 37 plausible hypotheses, each of which was formalized as an individual causal SEM structure (Appendix S2). All SEMs have VOD as the main response variable, with the effect of the other variables quantified directly and/or indirectly via burned area. Two exceptions were dry season rainfall and previous year VOD, which we deemed as reasonable predictors of VOD only (e.g., Brandt et al. \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e) and were not included as indirect predictors of VOD to avoid any circular causality problems (Bowman, Perry, and Marston 2015). Annual rainfall was specified as a predictor of both VOD and burned area. We acknowledge that fire regimes may be better explained by rainfall preceding the fire season, yet we only used annual rainfall as it was strongly collinear (Pearson\u0026rsquo;s r\u0026thinsp;\u0026gt;\u0026thinsp;0.774) with rainfall metrics accumulated over 6, 12, 18, and 24 months before the fire season (Appendix S3).\u003c/p\u003e\n\u003cp\u003e\u003cspan class=\"Underline\"\u003e2 - Model runs.\u003c/span\u003e All SEMs were run using the psem function in the R package \u0026lsquo;piecewiseSEM\u0026rsquo; (Lefcheck, Byrnes, and James 2020). Rather than fitting continent-wide models with all years, separate models were fit for each bioregion (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) and year (i.e., 1997\u0026ndash;2016) to be able to account for both the spatial and temporal dependency structures of the data (e.g., Walker et al. \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e) and, importantly, to assess to what extent support for each hypothesis varied across space and time. Preliminary model runs revealed strong spatial autocorrelation (SAC) in the residuals, indicated by statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) Moran\u0026rsquo;s I values and correlograms (Dormann et al. \u003cspan class=\"CitationRef\"\u003e2007\u003c/span\u003e). To account for SAC, SEMs were embedded with spatial autoregressive error models produced with the errorsarlm function in the R package \u0026lsquo;spatialreg\u0026rsquo; (Bivand, Piras, et al. \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e). Errorsarlm models handle SAC through spatial weighting matrices computed as the Euclidean distance between neighbouring sites (i.e., pixels) (Bivand and Wong 2018). We calculated spatial weighting matrices using the dnearneigh function from the R package \u0026lsquo;spdep\u0026rsquo; (Bivand, Altman, et al. \u003cspan class=\"CitationRef\"\u003e2022\u003c/span\u003e), setting the lower and upper distance bounds at 0 km and 39 km, respectively (39 km is the distance that allows all pixels to be linked to at least another pixel, including some farther pixels at the edge of the bioregions). The upper distance bound of 39 km did not vary significantly across bioregions as the data are regularly gridded (Appendix S4). SEM path coefficients were calculated as standardised regression coefficients (\u0026beta;) to enhance comparability across responses of different units. The indirect effect of a variable on VOD was calculated by multiplying the standardized regression coefficients of the two respective paths.\u003c/p\u003e\n\u003cp\u003e\u003cspan class=\"Underline\"\u003e3 - Model selection.\u003c/span\u003e We used Fisher\u0026rsquo;s C and Chi-squared (\u0026chi;\u003csup\u003e2\u003c/sup\u003e) goodness of fit measures to identify SEM specifications that best reproduce the relationships among the variables in the sample data (hereafter well-fitted SEMs). Fisher\u0026rsquo;s C is calculated by summing the p-values of all unspecified paths, and p\u0026thinsp;\u0026gt;\u0026thinsp;0.05 suggests that the model is well structured (i.e., no paths are missing) (Haynes et al. 2022; Shipley \u003cspan class=\"CitationRef\"\u003e2009\u003c/span\u003e). Similarly, \u0026chi;\u003csup\u003e2\u003c/sup\u003e tests whether there is a discrepancy between the model-implied and observed covariance matrices, and also in this case p\u0026thinsp;\u0026gt;\u0026thinsp;0.05 is recommended (Fan et al. \u003cspan class=\"CitationRef\"\u003e2016\u003c/span\u003e). In each year and each bioregion, we therefore selected only SEMs showing p\u0026thinsp;\u0026gt;\u0026thinsp;0.05 for Fisher\u0026rsquo;s C and \u0026chi;\u003csup\u003e2\u003c/sup\u003e (complete Fisher\u0026rsquo;s C and \u0026chi;\u003csup\u003e2\u003c/sup\u003e statistics, p-value, and degrees of freedom for each statistical test are provided in Appendix S5). The number of well-fitted SEMs was then refined by calculating the difference in the Akaike Information Criterion (\u0026Delta;AIC) between each SEM and the SEM with the lowest AIC (Cade \u003cspan class=\"CitationRef\"\u003e2015\u003c/span\u003e), and selecting only SEMs with \u0026Delta;AIC\u0026thinsp;\u0026lt;\u0026thinsp;3 (hereafter final SEMs) (Burnham, Anderson, and Huyvaert 2011). We present the results for each bioregion by natural model averaging the final SEMs \u0026beta; path coefficients each year, and then averaging over 1997\u0026ndash;2016. We took this approach as the values of the statistically significant standardised regression coefficients did not change appreciably over time (coefficient of variations\u0026thinsp;\u0026lt;\u0026thinsp;5%). Full result tables are provided in Appendix S5.\u003c/p\u003e\n\u003cp\u003e\u003cspan class=\"Underline\"\u003e4 - Model evaluation.\u003c/span\u003e Although final SEMs were selected via satisfactory Fisher\u0026rsquo;s C, \u0026chi;\u003csup\u003e2\u003c/sup\u003e, and \u0026Delta;AIC values, we further evaluated them by calculating Nagelkerke r\u003csup\u003e2\u003c/sup\u003e and plotting residuals against fitted values for each variable included and not included in the model, latitude and longitude, and time (Zuur and Ieno 2016) (Appendix S6). This step is important to assess the goodness of fit of any saturated SEMs (i.e., all variables are linked), as these have no degrees of freedom (Cortina et al. \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e). The four-step statistical modelling workflow is shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. The preprocessing we computed before the statistical analysis is detailed in Appendix S7. All analyses were performed within the R environment (R Core Team \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"3 Results","content":"\u003cp\u003eFrom the initial 37 SEM structures, the mean number of final SEMs (i.e., models with non-significant χ\u003csup\u003e2\u003c/sup\u003e and Fisher\u0026rsquo;s C statistics and ΔAIC\u0026thinsp;\u0026lt;\u0026thinsp;3) per year varied from 5.3 in Sahel grasslands, 5.7 in Greater Karoo and Kalahari drylands, 4.7 in Southeast African subtropical grasslands, to 2.4 in Madagascar, with a maximum of eleven (2009 in Greater Karoo and Kalahari drylands and 2005 in Southeast African subtropical grasslands) and a minimum of one (nine different years across all bioregions) (Appendix S5: Tables S1, S3, S5, S7). In other words, only ca. 12.2% of our hypothesized SEM causal structures could satisfy the criteria determining final SEMs each year. Further, our simple SEM structures were systematically excluded from the final SEMs, meaning that the observed direct and indirect relationships resulted from more complex structures (except for SEM 3 and SEM 8) (Appendixes S2).\u003c/p\u003e \u003cp\u003eIn Sahel grasslands, rainfall showed a positive direct effect on both VOD (average β\u0026thinsp;=\u0026thinsp;0.216 [95% confidence interval range during 1997\u0026ndash;2016 {CI}: 0.100, 0.398]) and burned area (β\u0026thinsp;=\u0026thinsp;0.256 [0.024, 0.537]) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). VOD showed a negative direct relationship with temperature (β = -0.216 [-0.387, -0.050]) while being largely unaffected by other variables. As expected, previous year VOD was a strong predictor of VOD (β\u0026thinsp;=\u0026thinsp;0.511 [0.385, 0.758]) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea and Appendix S5: Table S2). The Greater Karoo and Kalahari drylands and Southeast African subtropical grasslands showed overall similar relationships regarding the effects of rainfall on VOD and burned area, temperature on VOD, and previous year VOD on VOD (path coefficients of the remaining relationships were largely negligible) (Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eb and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ec and Appendix S5: Tables S4 and S6). In addition, in the Greater Karoo and Kalahari drylands, we also found a weak negative direct effect (β = -0.031 [-0.069, 0.021]) of burned area on VOD. This is likely because the Greater Karoo and Kalahari drylands is the only bioregion where VOD significantly increased during 1997\u0026ndash;2016, which indicates a potential increase in woody vegetation and, therefore, stronger feedback with burned area (Appendix S8: Figure S2). Madagascar also featured a positive relationship between rainfall and both VOD (β\u0026thinsp;=\u0026thinsp;0.126 [-0.069, 0.267]) and burned area (β\u0026thinsp;=\u0026thinsp;0.208 [-0.063, 0.599]) and a negative effect of temperature on VOD (β = -0.037 [-0.137, 0.133]), yet here, we observed a strong negative effect of both temperature (β = -0.429 [-1.215, -0.250]) and population density (β = -0.354 [-0.651, -0.015]) on burned area (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed and Appendix S5: Table\u0026nbsp;8). Noticeably, both these variables showed a sharp change during 1997\u0026ndash;2016 (Appendix S8: Figure S4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThree key features were shared by all bioregions. First, we did not observe any significant indirect effects (i.e., mediated by burned area) of the variables on VOD. In Sahel grasslands, for instance, the indirect effect of rainfall on VOD was much weaker (β\u0026thinsp;=\u0026thinsp;0.256 \u0026times; 0.012\u0026thinsp;=\u0026thinsp;0.0031) than its direct effect (β\u0026thinsp;=\u0026thinsp;0.216 [0.100, 0.398]) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea). This finding applied also in case of strong direct effects of variables on burned area (e.g., temperature and population density in Madagascar) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ed). Second, we noticed that the direct effects of precipitation and temperature on VOD had similar magnitudes but opposite signs. Given the marginal role of other variables, these diametrically opposite effects might have constrained changes in VOD, except for the Greater Karoo and Kalahari drylands (Appendix S8). Finally, all final SEMs showed both satisfactory residual plots (Appendix S6) and explained a high degree of variation in VOD as indicated by high Nagelkerke r\u003csup\u003e2\u003c/sup\u003e values, i.e., r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.88 in Sahel grasslands, r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.96 in Greater Karoo and Kalahari drylands, r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.84 in Southeast African subtropical grasslands, and r\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.94 in Madagascar (Figs.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003ea-d and Appendix S5: Tables S1, S3, S5, S7). The high Nagelkerke r\u003csup\u003e2\u003c/sup\u003e values are likely related to the strong spatial autocorrelation component observed in the gridded data and, hence, to the high explanatory power of the spatial weighting distances in our models.\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eEcological systems are complex because of the many interactions among biotic and abiotic components that occur at different spatiotemporal scales. As such, scientists tend to characterize ecological relationships as context-dependent (Spake et al. \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) or with narrative descriptions of relationships (Zellmer, Allen, and Kesseboehmer 2006). Our results, however, help disentangle this complexity for grassland systems by clarifying the role of temperature, rainfall, fire, and population density on woody dynamics at broad spatiotemporal scales. In addition, they also provide empirical evidence of complexity. The low number of final SEMs each year, for instance, indicates that only specific combinations of variables yield meaningful explanations of woody dynamics. Further, final SEMs are systematically described by more complex structures, suggesting that simple models including only woody vegetation, burned area, and one or two other variables are not able to characterize the ecological relationships in time and space.\u003c/p\u003e \u003cp\u003eYet complexity does not mean that ecosystems have no common features. Rainfall controls both woody vegetation and burned area, which is expected as water availability is known to be the key resource for both plant growth (Ogutu, D\u0026rsquo;Adamo, and Dash 2021; Sankaran et al. \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2005\u003c/span\u003e) and fuel availability (Lehmann et al. \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Staver, Botha, and Hedin 2017) in arid and semi-arid regions of Africa. However, and contradicting our initial hypothesis, the amount of dry season rainfall consistently did not predict woody vegetation dynamics, suggesting that the establishment of woody plants in these bioregions is more related to rainfall totals (Garc\u0026iacute;a Criado et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) than rainfall timing (Brandt et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). In contrast to rainfall, temperature showed a consistent negative effect on woody vegetation. While this is unsurprising as rainy days are generally colder, climate change-driven rising aridity may cause woody plants to be more prone to hydraulic failure (Abel et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Importantly, precipitation and temperature showed overall comparable but diametrically opposite effects on woody vegetation, which might have contributed to the prevention of encroachment during 1997\u0026ndash;2016 (i.e., no significant changes in VOD during 1997\u0026ndash;2016) (see Appendix 8). In turn, these results indicate that any future imbalances between these two variables might push the system toward a woody state. There is evidence that this may already be the case for the Greater Karoo and Kalahari drylands, which is the sole bioregion with significant positive trends in VOD (i.e., woody encroachment), and where weaker effects of temperature might have favoured rainfall-driven plant proliferation (see Appendix S8: Figure S2). This increase in VOD likely explains why the Greater Karoo and Kalahari drylands is the only bioregion where we observed the expected negative feedback between woody vegetation and fire, i.e., an increase in woody plants suppressed fire activity (Andela et al. \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). As such, it seems plausible that within the grasslands of sub-Saharan Africa (as opposed to grasslands and savanna together) the negative feedback between woody vegetation and fire is not pronounced and can only be detected at higher VOD. Madagascar also displayed a few distinct features, i.e., the effect of temperature and human population density on fire (see Appendix S8: Figure S4). This may relate not only to the different geographical setting (i.e., island vs. continental), but also to widespread human activities in the country (Ralimanana et al. \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Unsustainable agriculture and overexploitation, for instance, are likely explanations for our results, as fragmented landscapes reduce fuel connectivity and, therefore, fire spread (Bowman et al. \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Further, these anthropogenic activities (e.g., clearing woody vegetation) may explain why fire decline did not trigger woody encroachment (Phelps et al. \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe similarities in some relationships suggest that general rules of drivers of woody vegetation dynamics exist across the four bioregions investigated here. However, local changes or interactions for one or more variables may lead to context-dependent patterns that need to be accounted for ahead of effective grassland management (Bullock et al. 2021). However, it is also possible that some unexplained patterns may relate to important variables not investigated in this study (e.g., atmospheric CO\u003csub\u003e2\u003c/sub\u003e, grazing pressure, etc.) (Stevens et al. 2017). Another caveat to our findings is that the coarse spatial resolution of VOD data did not allow us to capture finer-grain processes and changes. For instance, a surprisingly large number of non-forest trees was observed in African drylands using 10 m to 0.5 m spatial resolution data (Reiner et al. \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Similarly, the contribution of small fires undetectable from coarse spatial resolution products has been shown to be decisive for appropriate burned area and fire emission estimations in Africa (e.g., up to +\u0026thinsp;80% of burned area detected with 20 m spatial resolution Sentinel 2 data as compared to MODIS data) (Ramo et al. \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). These examples suggest that the consistent availability of long-term high-resolution data will be essential to better understand finer-grain woody and fire dynamics and, ultimately, grassland distribution across sub-Saharan Africa.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eWe thank the groups that produced and shared the datasets used here. We thank J Lefcheck for his resources and guidance with the CRAN package \u0026lsquo;piecewiseSEM\u0026rsquo;, R Bivand for his help with the CRAN package \u0026lsquo;spdep\u0026rsquo;, and P Fibich for the fruitful discussions on errorsarlm models. This work was funded by the European Research Council Starting Grant SCALEFORES (515825101) granted to F Eigenbrod and the UKCEH National Capability project (06895) to J M Bullock.\u003c/p\u003e \u003ch2\u003eConflict of interest statement\u003c/h2\u003e \u003cp\u003eThe authors declare no conflicting interests.\u003c/p\u003e "},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbdi, Abdulhakim M, Martin Brandt, Christin Abel, and Rasmus Fensholt. 2022. \u0026ldquo;Review Article Satellite Remote Sensing of Savannas: Current Status and Emerging Opportunities.\u0026rdquo; Journal of Remote Sensing, 2022:2022. https://doi.org/DOI: 10.34133/2022/9835284.\u003c/li\u003e\n\u003cli\u003eAbel, Christin, Abdulhakim M Abdi, Torbern Tagesson, Stephanie Horion, and Rasmus Fensholt. 2023. \u0026ldquo;Contrasting Ecosystem Vegetation Response in Global Drylands under Drying and Wetting Conditions.\u0026rdquo; Global Change Biology, 29: 3954\u0026ndash;69. https://doi.org/10.1111/gcb.16745.\u003c/li\u003e\n\u003cli\u003eAbel, Christin, St\u0026eacute;phanie Horion, Torbern Tagesson, Wanda De Keersmaecker, Alistair W R Seddon, Abdulhakim M Abdi, and Rasmus Fensholt. 2020. \u0026ldquo;The Human\u0026ndash;Environment Nexus and Vegetation\u0026ndash;Rainfall Sensitivity in Tropical Drylands.\u0026rdquo; Nature Sustainability, 4: 25\u0026ndash;32. https://doi.org/10.1038/s41893-020-00597-z.\u003c/li\u003e\n\u003cli\u003eAllen, Vivien G, Caterina Batello, Elbio J Berretta, John Hodgson, Mort Kothmann, Xianglin Li, John McIvor, et al. 2011. \u0026ldquo;An International Terminology for Grazing Lands and Grazing Animals.\u0026rdquo; Grass and Forage Science 66: 2\u0026ndash;28. https://doi.org/10.1111/j.1365-2494.2010.00780.x.\u003c/li\u003e\n\u003cli\u003eAlvarado, Swanni T, Niels Andela, Thiago S F Silva, and Sally Archibald. 2020. \u0026ldquo;Thresholds of Fire Response to Moisture and Fuel Load Differ between Tropical Savannas and Grasslands across Continents.\u0026rdquo; Global Ecology and Biogeography 29: 331\u0026ndash;44. https://doi.org/10.1111/geb.13034.\u003c/li\u003e\n\u003cli\u003eAndela, Niels, Douglas C Morton, Luis Giglio, Yang Chen, Guido R van der Werf, Prasad S Kasibhatla, Rurth S DeFries, et al. 2017. \u0026ldquo;A Human-Driven Decline in Global Burned Area.\u0026rdquo; Science 356: 1356\u0026ndash;62. https://doi.org/10.1126/science.aal4108.\u003c/li\u003e\n\u003cli\u003eArchibald, Sally, Alecia Nickless, Nicolin Govender, Robert J Scholes, and Veiko Lehsten. 2010. \u0026ldquo;Climate and the Inter-Annual Variability of Fire in Southern Africa: A Meta-Analysis Using Long-Term Field Data and Satellite-Derived Burnt Area Data.\u0026rdquo; Global Ecology and Biogeography 19: 794\u0026ndash;809. https://doi.org/10.1111/j.1466-8238.2010.00568.x.\u003c/li\u003e\n\u003cli\u003eArchibald, Sally, David P Roy, Brian W van Wilgen, and Robert J Scholes. 2009. \u0026ldquo;What Limits Fire? An Examination of Drivers of Burnt Area in Southern Africa.\u0026rdquo; Global Change Biology 15: 613\u0026ndash;30. https://doi.org/10.1111/j.1365-2486.2008.01754.x.\u003c/li\u003e\n\u003cli\u003eBardgett, Richard D, James M Bullock, Sandra Lavorel, Peter Manning, Urs Schaffner, Nicholas Ostle, Mathilde Chomel, et al. 2021. \u0026ldquo;Combatting Global Grassland Degradation.\u0026rdquo; Nature Reviews Earth and Environment 2: 720\u0026ndash;35. https://doi.org/10.1038/s43017-021-00207-2.\u003c/li\u003e\n\u003cli\u003eBivand, Roger, Micah Altman, Luc Anselin, Renato Assun\u0026ccedil;\u0026atilde;o, Olaf Berke, Guillaume Blanchet, Marilia Carvalho, et al. 2022. \u0026ldquo;Package \u0026lsquo;Spdep.\u0026rsquo;\u0026rdquo; CRAN.\u003c/li\u003e\n\u003cli\u003eBivand, Roger, Gianfranco Piras, Luc Anselin, Andrew Bernat, Eric Blankmeyer, Yongwan Chun, Virgilio G\u0026oacute;mez-Rubio, et al. 2022. \u0026ldquo;Package \u0026lsquo;Spatialreg.\u0026rsquo;\u0026rdquo; CRAN.\u003c/li\u003e\n\u003cli\u003eBivand, Roger S, and David W Wong. 2018. \u0026ldquo;Comparing Implementations of Global and Local Indicators of Spatial Association.\u0026rdquo; Test 27: 716\u0026ndash;48. https://doi.org/10.1007/s11749-018-0599-x.\u003c/li\u003e\n\u003cli\u003eBowman, David M J S, Crystal A Kolden, John T Abatzoglou, Fay H Johnston, Guido R van der Werf, and Mike Flannigan. 2020. \u0026ldquo;Vegetation Fires in the Anthropocene.\u0026rdquo; Nature Reviews Earth and Environment 1: 500\u0026ndash;15. https://doi.org/10.1038/s43017-020-0085-3.\u003c/li\u003e\n\u003cli\u003eBowman, David M J S, George L W Perry, and J B Marston. 2015. \u0026ldquo;Feedbacks and Landscape-Level Vegetation Dynamics.\u0026rdquo; Trends in Ecology and Evolution 30: 255\u0026ndash;60. https://doi.org/10.1016/j.tree.2015.03.005.\u003c/li\u003e\n\u003cli\u003eBrandt, Martin, Pierre Hiernaux, Kjeld Rasmussen, Compton J Tucker, Jean-Pierre Wigneron, Abdoul Aziz Diouf, Stefanie M Herrmann, et al. 2019. \u0026ldquo;Changes in Rainfall Distribution Promote Woody Foliage Production in the Sahel.\u0026rdquo; Communications Biology 2: 133. https://doi.org/10.1038/s42003-019-0383-9.\u003c/li\u003e\n\u003cli\u003eBrandt, Martin, Kjeld Rasmussen, Josep Pe\u0026ntilde;uelas, Feng Tian, Guy Schurgers, Aleixandre Verger, Ole Mertz, John R B Palmer, and Rasmus Fensholt. 2017. \u0026ldquo;Human Population Growth Offsets Climate-Driven Increase in Woody Vegetation in Sub-Saharan Africa.\u0026rdquo; Nature Ecology \u0026amp; Evolution 1: 0081. https://doi.org/10.1038/s41559-017-0081.\u003c/li\u003e\n\u003cli\u003eBurnham, Kenneth P, David R Anderson, and Kathryn P Huyvaert. 2011. \u0026ldquo;AIC Model Selection and Multimodel Inference in Behavioral Ecology: Some Background, Observations, and Comparisons.\u0026rdquo; Behavioral Ecology and Sociobiology 65: 23\u0026ndash;35. https://doi.org/10.1007/s00265-010-1029-6.\u003c/li\u003e\n\u003cli\u003eCade, Brian S. 2015. \u0026ldquo;Model Averaging and Muddled Multimodel Inferences.\u0026rdquo; Ecology 96: 2370\u0026ndash;82. https://doi.org/10.1890/14-1639.1.\u003c/li\u003e\n\u003cli\u003eCharles-Dominique, Tristan, Guy F Midgley, Kyle W Tomlinson, and William J Bond. 2018. \u0026ldquo;Steal the Light: Shade vs Fire Adapted Vegetation in Forest\u0026ndash;Savanna Mosaics.\u0026rdquo; New Phytologist 218: 1419\u0026ndash;29. https://doi.org/10.1111/nph.15117.\u003c/li\u003e\n\u003cli\u003eCIESIN. 2018. \u0026quot;Gridded Population of the World, Version 4 (GPWv4): Population Count Adjusted to Match 2015 Revision of UN WPP Country Totals, Revision 11\u0026quot;. Palisades, New York: NASA Socioeconomic Data and Applications Center (SEDAC). https://doi.org/10.7927/H4PN93PB.\u003c/li\u003e\n\u003cli\u003eCortina, Jose M, Jennifer P Green, Kathleen R Keeler, and Robert J Vandenberg. 2017. \u0026ldquo;Degrees of Freedom in SEM: Are We Testing the Models That We Claim to Test?\u0026rdquo; Organizational Research Methods 20: 350\u0026ndash;78. https://doi.org/10.1177/1094428116676345.\u003c/li\u003e\n\u003cli\u003eD\u0026rsquo;Adamo, Francesco, Booker Ogutu, Martin Brandt, Guy Schurgers, and Jadunandan Dash. 2021. \u0026ldquo;Climatic and Non-Climatic Vegetation Cover Changes in the Rangelands of Africa.\u0026rdquo; Global and Planetary Change 202: 103516. https://doi.org/10.1016/j.gloplacha.2021.103516.\u003c/li\u003e\n\u003cli\u003eDinerstein, Eric, David Olson, Anup Joshi, Carly Vynne, Neil D Burgess, Eric Wikramanayake, Nathan Hahn, et al. 2017. \u0026ldquo;An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm.\u0026rdquo; BioScience 67: 534\u0026ndash;45. https://doi.org/10.1093/biosci/bix014.\u003c/li\u003e\n\u003cli\u003eD\u0026rsquo;Odorico, Paolo, Gregory S Okin, and Brandon T Bestelmeyer. 2012. \u0026ldquo;A Synthetic Review of Feedbacks and Drivers of Shrub Encroachment in Arid Grasslands.\u0026rdquo; Ecohydrology 5: 520\u0026ndash;30. https://doi.org/10.1002/eco.259.\u003c/li\u003e\n\u003cli\u003eDorigo, Wouter, Wolfgang Wagner, Clement Albergel, Franziska Albrecht, Gianpaolo Balsamo, Luca Brocca, Daniel Chung, et al. 2017. \u0026ldquo;ESA CCI Soil Moisture for Improved Earth System Understanding: State-of-the-Art and Future Directions.\u0026rdquo; Remote Sensing of Environment 203: 185\u0026ndash;215. https://doi.org/10.1016/j.rse.2017.07.001.\u003c/li\u003e\n\u003cli\u003eDorigo, Wouter, Wolfgang Wagner, Gruber, Alexander, Scanlon, Tracy; Hahn, Sebastian; Kidd, Richard, Paulik, C, Reimer, Christoph; Van der Schalie, Robin; De Jeu, Richard (2018): \u0026quot;ESA Soil Moisture Climate Change Initiative: Combined Product, Version 04.2.\u0026quot; Centre for Environmental Data Analysis. https://doi.org/10.5285/3a8a94c3fa464d68b6d70df291afd457.\u003c/li\u003e\n\u003cli\u003eDormann, Carsten F, Jana M. McPherson, Miguel B Ara\u0026uacute;jo, Roger Bivand, Janine Bolliger, Gudrun Carl, Richard G Davies, et al. 2007. \u0026ldquo;Methods to Account for Spatial Autocorrelation in the Analysis of Species Distributional Data: A Review.\u0026rdquo; Ecography 30: 609\u0026ndash;28. https://doi.org/10.1111/j.2007.0906-7590.05171.x.\u003c/li\u003e\n\u003cli\u003eESA. 2017. \u0026ldquo;Land Cover CCI: Product User Guide Version 2.0.\u0026rdquo; UCL-Geomatics (Belgium).\u003c/li\u003e\n\u003cli\u003eFan, Yi, Jiquan Chen, Gabriela Shirkey, Ranjeet John, Susie R Wu, Hogeun Park, and Changliang Shao. 2016. \u0026ldquo;Applications of Structural Equation Modeling (SEM) in Ecological Studies: An Updated Review.\u0026rdquo; Ecological Processes 5. https://doi.org/10.1186/s13717-016-0063-3.\u003c/li\u003e\n\u003cli\u003eFunk, Chris, Pete Peterson, Martin Landsfeld, Diego Pedreros, James Verdin, Shraddhanand Shukla, Gregory Husak, et al. 2015. \u0026ldquo;The Climate Hazards Infrared Precipitation with Stations\u0026mdash;a New Environmental Record for Monitoring Extremes.\u0026rdquo; Scientific Data 2: 150066. https://doi.org/10.1038/sdata.2015.66.\u003c/li\u003e\n\u003cli\u003eGarc\u0026iacute;a Criado, Mariana, Isla H Myers-Smith, Anne D Bjorkman, Caroline E R Lehmann, and Nicola Stevens. 2020. \u0026ldquo;Woody Plant Encroachment Intensifies under Climate Change across Tundra and Savanna Biomes.\u0026rdquo; Global Ecology and Biogeography 29: 925\u0026ndash;43. https://doi.org/10.1111/geb.13072.\u003c/li\u003e\n\u003cli\u003eHansen, Matthew C, Peter V Potapov, Rebecca Moore, Matt Hancher, Svetlana A Turubanova, Alexandra Tyukavina, et al. 2013. \u0026ldquo;High-Resolution Global Maps of 21st-Century Forest Cover Change.\u0026rdquo; Science 342: 850\u0026ndash;53. https://doi.org/10.1126/science.1244693.\u003c/li\u003e\n\u003cli\u003eHaynes, Kyle J, Andrew M Liebhold, Jonathan S Lefcheck, Randall S Morin, and Guiming Wang. 2022. \u0026ldquo;Climate Affects the Outbreaks of a Forest Defoliator Indirectly through Its Tree Hosts.\u0026rdquo; Oecologia. https://doi.org/10.1007/s00442-022-05123-w.\u003c/li\u003e\n\u003cli\u003eJames M Bullock, Elisa Fuentes-Montemayor, Ben McCarthy, Kirsty Park, Rosie S Hails, Ben A Woodcock, Kevin Watts, Ron Corstanje, and Jim Harris. 2021. \u0026ldquo;Future Restoration Should Enhance Ecological Complexity and Emergent Properties at Multiple Scales.\u0026rdquo; Ecography, 1\u0026ndash;11. https://doi.org/10.1111/ecog.05780.\u003c/li\u003e\n\u003cli\u003eKarger, Dirk N, Olaf Conrad, Jurgen B\u0026ouml;hner, Tobias Kawohl, Holger Kreft, Rodrigo W Soria-Auza, Niklaus Zimmermann, Peter Linder, and Michael Kessler. 2017. Climatologies at high resolution for the earth\u0026rsquo;s land surface areas. Scientific Data 4, 170122. https://doi.org/10.1038/sdata.2017.122.\u003c/li\u003e\n\u003cli\u003eKarger, Dirk N, Olaf Conrad, Jurgen B\u0026ouml;hner, Tobias Kawohl, Holger Kreft, Rodrigo W Soria-Auza, Niklaus Zimmermann, Peter Linder, and Michael Kessler. 2021. \u0026ldquo;Climatologies at high resolution for the earth land surface areas.\u0026rdquo; EnviDat. https://doi.org/10.16904/envidat.228.v2.1.\u003c/li\u003e\n\u003cli\u003eLefcheck, Jonathan S, Jarrett Byrnes, and Grace James. 2020. \u0026ldquo;Package \u0026lsquo;PiecewiseSEM.\u0026rsquo;\u0026rdquo; CRAN.\u003c/li\u003e\n\u003cli\u003eLefcheck, Jonathan S. 2016. \u0026ldquo;PiecewiseSEM: Piecewise Structural Equation Modelling in r for Ecology, Evolution, and Systematics.\u0026rdquo; Methods in Ecology and Evolution 7: 573\u0026ndash;79. https://doi.org/10.1111/2041-210X.12512.\u003c/li\u003e\n\u003cli\u003eLehmann, Caroline E R, T M Anderson, M Sankaran, S I Higgins, Sally Archibald, W A Hoffmann, N P Hanan, et al. 2014. \u0026ldquo;Savanna Vegetation-Fire-Climate Relationships Differ Among Continents.\u0026rdquo; Science 343: 548\u0026ndash;52. https://doi.org/10.1126/science.1247355.\u003c/li\u003e\n\u003cli\u003eLi, Xiaojun, Jean Pierre Wigneron, Fr\u0026eacute;d\u0026eacute;ric Frappart, Lei Fan, Philippe Ciais, Rasmus Fensholt, Dara Entekhabi, et al. 2021. \u0026ldquo;Global-Scale Assessment and Inter-Comparison of Recently Developed/Reprocessed Microwave Satellite Vegetation Optical Depth Products.\u0026rdquo; Remote Sensing of Environment 253:112208. https://doi.org/10.1016/j.rse.2020.112208.\u003c/li\u003e\n\u003cli\u003eMayer, Audrey L, and Azad H Khalyani. 2011. \u0026ldquo;Grass Trumps Trees with Fire.\u0026rdquo; Science 334: 188\u0026ndash;89. https://doi.org/10.1126/science.1213908.\u003c/li\u003e\n\u003cli\u003eMcDowell, Nate G, Gerard Sapes, Alexandria Pivovaroff, Henry D Adams, Craig D Allen, William R L Anderegg, Matthias Arend, et al. 2022. \u0026ldquo;Mechanisms of Woody-Plant Mortality under Rising Drought, CO2 and Vapour Pressure Deficit.\u0026rdquo; Nature Reviews Earth \u0026amp; Environment 3: 294\u0026ndash;308. https://doi.org/10.1038/s43017-022-00272-1.\u003c/li\u003e\n\u003cli\u003eMeesters, Antoon G C, Richard A M DeJeu, and Manfred Owe. 2005. \u0026ldquo;Analytical Derivation of the Vegetation Optical Depth From the Microwave Polarization Difference Index.\u0026rdquo; IEEE Geoscience and Remote Sensing Letters 2: 121\u0026ndash;23. https://doi.org/10.1109/LGRS.2005.843983.\u003c/li\u003e\n\u003cli\u003eMoesinger, Leander, Wouter Dorigo, Richard De Jeu, Robin Van Der Schalie, Tracy Scanlon, Irene Teubner, and Matthias Forkel. 2020. \u0026ldquo;The Global Long-Term Microwave Vegetation Optical Depth Climate Archive (VODCA).\u0026rdquo; Earth System Science Data 12: 177\u0026ndash;96. https://doi.org/10.5194/essd-12-177-2020.\u003c/li\u003e\n\u003cli\u003eNerlekar, Ashish N, and Joseph W Veldman. 2020. \u0026ldquo;High Plant Diversity and Slow Assembly of Old-Growth Grasslands.\u0026rdquo; Proceedings of the National Academy of Sciences of the United States of America 117: 18550\u0026ndash;56. https://doi.org/10.1073/pnas.1922266117.\u003c/li\u003e\n\u003cli\u003eOgutu, Booker O, Francesco D\u0026rsquo;Adamo, and Jadunandan Dash. 2021. \u0026ldquo;Impact of Vegetation Greening on Carbon and Water Cycle in the African Sahel-Sudano-Guinean Region.\u0026rdquo; Global and Planetary Change 202: 103524. https://doi.org/10.1016/j.gloplacha.2021.103524.\u003c/li\u003e\n\u003cli\u003eOne Earth. 2020. \u0026ldquo;Bioregions.\u0026rdquo; 2020. URL https://www.oneearth.org/.\u003c/li\u003e\n\u003cli\u003eOsborne, Colin P, Tristan Charles-Dominique, Nicola Stevens, William J Bond, Guy Midgley, and Caroline E R Lehmann. 2018. \u0026ldquo;Human Impacts in African Savannas Are Mediated by Plant Functional Traits.\u0026rdquo; New Phytologist 220: 10\u0026ndash;24. https://doi.org/10.1111/nph.15236.\u003c/li\u003e\n\u003cli\u003ePausas, Juli G, and William J Bond. 2020. \u0026ldquo;Alternative Biome States in Terrestrial Ecosystems.\u0026rdquo; Trends in Plant Science 25: 250\u0026ndash;63. https://doi.org/10.1016/j.tplants.2019.11.003.\u003c/li\u003e\n\u003cli\u003ePhelps, Leanne N, Niels Andela, Mathieu Gravey, Dylan S Davis, Christian A Kull, Kristina Douglass, and Caroline E R Lehmann. 2022. \u0026ldquo;Madagascar\u0026rsquo;s Fire Regimes Challenge Global Assumptions about Landscape Degradation.\u0026rdquo; Global Change Biology, 1\u0026ndash;17. https://doi.org/10.1111/gcb.16206.\u003c/li\u003e\n\u003cli\u003eR Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.\u003c/li\u003e\n\u003cli\u003eRalimanana, H\u0026eacute;l\u0026egrave;ne, Allison L Perrigo, Rhian J Smith, James S Borrell, S\u0026oslash;ren Faurby, Mamy Tiana Rajaonah, Tianjanahary Randriamboavonjy, et al. 2022. \u0026ldquo;Madagascar\u0026rsquo;s Extraordinary Biodiversity: Threats and Opportunities.\u0026rdquo; Science 378: eadf1466. https://doi.org/10.1126/science.adf1466.\u003c/li\u003e\n\u003cli\u003eRamo, Ruben, Ekhi Roteta, Ioannis Bistinas, Dave van Wees, Aitor Bastarrika, Emilio Chuvieco, and Guido R van der Werf. 2021. \u0026ldquo;African Burned Area and Fire Carbon Emissions Are Strongly Impacted by Small Fires Undetected by Coarse Resolution Satellite Data.\u0026rdquo; Proceedings of the National Academy of Sciences of the United States of America 118: 1\u0026ndash;7. https://doi.org/10.1073/pnas.2011160118.\u003c/li\u003e\n\u003cli\u003eReiner, Florian, Martin Brandt, Xiaoye Tong, David Skole, Ankit Kariryaa, Philippe Ciais, Andrew Davies, et al. 2023. \u0026ldquo;More than One Quarter of Africa\u0026rsquo;s Tree Cover Is Found Outside Areas Previously Classified as Forest.\u0026rdquo; Nature Communications 14: 2258. https://doi.org/10.1038/s41467-023-37880-4.\u003c/li\u003e\n\u003cli\u003eSankaran, Mahesh, Niall P Hanan, Robert J Scholes, Jayashree Ratnam, David J Augustine, Brian S Cade, Jacques Gignoux, et al. 2005. \u0026ldquo;Determinants of Woody Cover in African Savannas.\u0026rdquo; Nature 438: 846\u0026ndash;49. https://doi.org/10.1038/nature04070.\u003c/li\u003e\n\u003cli\u003eShipley, Bill. 2009. \u0026ldquo;Confirmatory Path Analysis in a Generalized Multilevel Context.\u0026rdquo; Ecology 90: 363\u0026ndash;68. https://doi.org/10.1890/08-1034.1.\u003c/li\u003e\n\u003cli\u003eSpake, Rebecca, Chloe Bellamy, Laura J Graham, Kevin Watts, Tom Wilson, Lisa R Norton, Claire M Wood, Reto Schmucki, James M Bullock, and Felix Eigenbrod. 2019. \u0026ldquo;An Analytical Framework for Spatially Targeted Management of Natural Capital\u0026rdquo; Nature Sustainability 2: 90\u0026ndash;97. https://doi.org/10.1038/s41893-019-0223-4.\u003c/li\u003e\n\u003cli\u003eSpake, Rebecca, Diana E Bowler, Corey T Callaghan, Shane A Blowes, Patrick C Doncaster, Laura H Ant\u0026atilde;o, Shinichi Nakagawa, Richard McElreath, and Jonathan M Chase. 2023. \u0026ldquo;Understanding \u0026lsquo;It Depends\u0026rsquo; in Ecology: A Guide to Hypothesising, Visualising and Interpreting Statistical Interactions.\u0026rdquo; Biological Reviews, brv.12939. https://doi.org/10.1111/brv.12939.\u003c/li\u003e\n\u003cli\u003eSpake, Rebecca, Rose E O\u0026rsquo;Dea, Shinichi Nakagawa, Patrick C Doncaster, Masahiro Ryo, Corey T Callaghan, and James M Bullock. 2022. \u0026ldquo;Improving Quantitative Synthesis to Achieve Generality in Ecology.\u0026rdquo; Nature Ecology \u0026amp; Evolution 6: 1818\u0026ndash;28. https://doi.org/10.1038/s41559-022-01891-z.\u003c/li\u003e\n\u003cli\u003eStaver, Carla A, Judith Botha, and Lars Hedin. 2017. \u0026ldquo;Soils and Fire Jointly Determine Vegetation Structure in an African Savanna.\u0026rdquo; New Phytologist 216: 1151\u0026ndash;60. https://doi.org/10.1111/nph.14738.\u003c/li\u003e\n\u003cli\u003eStevens, Nicola, William Bond, Angelica Feurdean, and Caroline E R Lehmann. 2022. \u0026ldquo;Grassy Ecosystems in the Anthropocene.\u0026rdquo; Annual Review of Environment and Resources 47, 261\u0026ndash;289. https://doi.org/10.1146/annurev-environ-112420-015211.\u003c/li\u003e\n\u003cli\u003eStr\u0026ouml;mberg, Caroline A E, and Carla A Staver. 2022. \u0026ldquo;The History and Challenge of Grassy Biomes.\u0026rdquo; Science 377: 592\u0026ndash;93. https://doi.org/10.1126/science.add1347.\u003c/li\u003e\n\u003cli\u003eVenter, Zander S, Heidi Jayne Hawkins, and Michael D Cramer. 2017. \u0026ldquo;Implications of Historical Interactions between Herbivory and Fire for Rangeland Management in African Savannas.\u0026rdquo; Ecosphere 8: 1\u0026ndash;14. https://doi.org/10.1002/ecs2.1946.\u003c/li\u003e\n\u003cli\u003eWalker, Xanthe J, Brendan M Rogers, Sander Veraverbeke, Jill F Johnstone, Jennifer L Baltzer, Kirsten Barrett, Laura Bourgeau-Chavez, et al. 2020. \u0026ldquo;Fuel Availability Not Fire Weather Controls Boreal Wildfire Severity and Carbon Emissions.\u0026rdquo; Nature Climate Change 10: 1130\u0026ndash;36. https://doi.org/10.1038/s41558-020-00920-8.\u003c/li\u003e\n\u003cli\u003eWei, Fangli, Shuai Wang, Martin Brandt, Bojie Fu, Michael E Meadows, Lixin Wang, Lanhui Wang, Xiaowei Tong, and Rasmus Fensholt. 2020. \u0026ldquo;Responses and Feedbacks of African Dryland Ecosystems to Environmental Changes.\u0026rdquo; Current Opinion in Environmental Sustainability 48: 29\u0026ndash;35. https://doi.org/10.1016/j.cosust.2020.09.004.\u003c/li\u003e\n\u003cli\u003eWei, Fangli, Shuai Wang, Bojie Fu, Martin Brandt, Naiqing Pan, Cong Wang, and Rasmus Fensholt. 2020. \u0026ldquo;Nonlinear Dynamics of Fires in Africa over Recent Decades Controlled by Precipitation.\u0026rdquo; Global Change Biology 26: 4495\u0026ndash;4505. https://doi.org/10.1111/gcb.15190.\u003c/li\u003e\n\u003cli\u003eWerf, Guido R van der, James T Randerson, Louis Giglio, Nadine Gobron, and A J Dolman. 2008. \u0026ldquo;Climate Controls on the Variability of Fires in the Tropics and Subtropics.\u0026rdquo; Global Biogeochemical Cycles 22: 1\u0026ndash;13. https://doi.org/10.1029/2007GB003122.\u003c/li\u003e\n\u003cli\u003eWerf, Guido R van der, James T Randerson, Louis Giglio, Thijs T Van Leeuwen, Yang Chen, Brendan M Rogers, Mingquan Mu, et al. 2017. \u0026ldquo;Global Fire Emissions Estimates during 1997-2016.\u0026rdquo; Earth System Science Data 9: 697\u0026ndash;720. https://doi.org/10.5194/essd-9-697-2017.\u003c/li\u003e\n\u003cli\u003eWieczorkowski, Jakub D, and Caroline E R Lehmann. 2022. \u0026ldquo;Encroachment Diminishes Herbaceous Plant Diversity in Grassy Ecosystems Worldwide.\u0026rdquo; Global Change Biology, 1\u0026ndash;15. https://doi.org/10.1111/gcb.16300.\u003c/li\u003e\n\u003cli\u003eZellmer, Amanda J, Timothy F H Allen, and Kirsten Kesseboehmer. 2006. \u0026ldquo;The Nature of Ecological Complexity: A Protocol for Building the Narrative.\u0026rdquo; Ecological Complexity 3: 171\u0026ndash;82. https://doi.org/10.1016/j.ecocom.2006.06.002.\u003c/li\u003e\n\u003cli\u003eZuur, Alain F, and Elena N Ieno. 2016. \u0026ldquo;A Protocol for Conducting and Presenting Results of Regression-Type Analyses.\u0026rdquo; Methods in Ecology and Evolution 7: 636\u0026ndash;45. https://doi.org/10.1111/2041-210X.12577.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[{"identity":"88fa172e-7c12-40c6-8d85-ab8e9ae48acd","identifier":"10.13039/501100000781","name":"European Research Council","awardNumber":"Starting Grant SCALEFORES (515825101)","order_by":0},{"identity":"bd2b5d96-c1bf-4056-94d1-93997defd235","identifier":"10.13039/501100011027","name":"Centre for Ecology and Hydrology","awardNumber":"National Capability project (06895)","order_by":1}],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"University of Southampton","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"grassland ecosystems, grassland-forest transition, rainfall-vegetation-fire feedback, disturbances, structural equation modelling, sub-Saharan Africa, woody dynamics","lastPublishedDoi":"10.21203/rs.3.rs-3914432/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3914432/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eUnderstanding the drivers of ecosystem dynamics, and how responses vary spatially and temporally, is a critical challenge in the face of global change. Here we used structural equation models and remote sensing datasets to understand the direct and indirect effects of climatic, environmental, and anthropogenic variables on woody vegetation dynamics across four grasslands regions (i.e., Sahel grasslands, Greater Karoo and Kalahari drylands, Southeast African subtropical grasslands, and Madagascar) of sub-Saharan Africa. We focus on African grasslands given their importance for biodiversity and ecosystem services, the lack of clarity on how they are likely to respond to changes in disturbances, and how such responses vary geographically. This is particularly the case of grass-dominated ecosystems \u0026ndash; the focus of our study \u0026ndash; rather than more mixed grass-tree regions (e.g., savannas). Rainfall (β\u0026thinsp;=\u0026thinsp;0.148 [-0.111, 0.398]) and temperature (β = -0.109 [-0.387, 0.133]) showed consistently opposing effects on woody vegetation (average standardised regression coefficients and 95% confidence interval range during 1997\u0026ndash;2016) across the four bioregions. Other variables showed overall negligible effects including, for instance, dry season rainfall, soil moisture and, notably, fire. Other relationships were more context-dependent. Only Greater Karoo and Kalahari drylands showed a negative relationship between woody vegetation and fire (β = -0.031 [-0.069, 0.021]). Similarly, in Madagascar we observed strong negative effects of temperature (β = -0.429 [-1.215, -0.259]) and population density (β = -0.354 [-0.651, -0.015]) on burned area, yet these did not result in any significant indirect effects on woody vegetation. Our results clarify the contribution of environmental and anthropogenic variables in controlling woody dynamics at broad spatiotemporal scales and reveal that the widely documented negative feedback between fire and woody vegetation does not necessarily apply across all African grasslands.\u003c/p\u003e","manuscriptTitle":"Precipitation and temperature drive woody dynamics in the grasslands of sub-Saharan Africa","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-02-01 16:20:14","doi":"10.21203/rs.3.rs-3914432/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"033ee579-2915-4e09-980f-51c393ef3102","owner":[],"postedDate":"February 1st, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":28489942,"name":"Terrestrial Ecology"}],"tags":[],"updatedAt":"2024-02-01T16:20:14+00:00","versionOfRecord":[],"versionCreatedAt":"2024-02-01 16:20:14","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3914432","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3914432","identity":"rs-3914432","version":["v1"]},"buildId":"ApUGefWb6u5IBVtyqm6d5","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}