Plant dispersal mechanisms
Bush tomatoes are a highly desirable cultural resource for Aboriginal people living in remote desert communities. Not only is the fruit a high quality source of nutrients and calories that can be obtained with relatively little effort, it also plays an important role in people’s ritual lives as expressed through the Dreaming36. For Martu, there are two species of primary importance: (a) Solanum diversiflorum (known locally as wamula or kumpulpaja), a sweet, large-fruited species (25 g) with a hollow interior holding a mass of inedible, bitter, black seeds that must be removed before consumption (Fig. 1); and (b) Solanum centrale (jinjiwirri) a small-fruited species (5 g) with edible seeds embedded in a fleshy, sweet berry. Both species are more abundant following recent fire19, and in our study, more than 90% of observations of plant presence occur within 5 years following a fire.
Because S. diversiflorum has fairly large and inedible seeds encased within heavy fruit, it is most likely to be dispersed via epizoochory (active or passive transport of seed), in which people carry fruit to a location for processing. Seed dispersal is thus likely to be associated with places on the landscape where people have camped, in both past and present. People are likely to be the most important seed dispersers, as it is unclear what animals beyond humans regularly consume S.diversiflorum fruit, or act as its dispersal agents58,59,60, and it appears there is little competition with people for the fruit. Most fruits that go unpicked by people simply dry up on the stem and the seeds drop to the base of the plant within the desiccated fruit case. On rare occasions, Martu have pointed out S.diversiflorum fruit that hill kangaroo have nibbled, but the kangaroo do so in a way that leaves the seeds attached to the calyx hanging from the stem—they do not transport seed beyond the parent plant.
In comparison, Solanum centrale is less likely to be dispersed primarily by people. The seeds are small, edible and consumed rather than discarded; the fruit is small, soft and is always transported whole to residential sites or eaten on the spot; historically, it was often processed into large cakes and dried for storage and transport61. Some epizoochory through seed loss might be expected based on the occasional escape of whole or dried fruit from dishes and seedcakes as it is transported cross country, but we believe this occurred on a scale that was much more restricted than S.diversiflorum. Depending on the efficiency of digestion, human endozoochory (consumption and excretion of undigested seed) might be possible for this species, although Martu traditional knowledge does not recognize this as a potential dispersal mechanism. We do note that Solanaceae taxa with similar morphology (such as Physalis, which also possesses small fleshy fruits with embedded seeds) have been found at high densities in human coprolites62. People are not the only potential disperser of this plant, however, as it is often consumed by many other organisms, especially hare wallabies63 (Lagorchestes hirsutus, now locally extinct) emus (Dromaius novaehollanidae) and bustards (Ardeotis australis)64,65. Aboriginal Dreaming narratives relate how birds often fight over the fruit. It rarely remains on the plant for long after ripening. Unlike S.diversiflorum, it is clonal; thus vegetative propagation may be a more significant influence on its distribution than seed dispersal40,58, and past anthropogenic contributions to its dispersal may be difficult to detect. In addition, because people now typically bring most of the fruit back to the community for consumption, there are likely to be fewer opportunities for endozoochory at a landscape scale and thus we would not expect to detect strong effects of contemporary landscape use.
Our research design makes use of a controlled comparison not only between two Solanaceae that are more or less likely to be dispersed by people but also between two additional taxa: Eragrostis setifolia/eriopoda (both species of Eragrostis are utilized interchangably and not distinguished by Martu foragers), and Scaevola parvifolia. Both plants are common and co-occur with bush tomatoes in early-mid successional sandplain habitats (Fig. 1C), and both are edible but much lower ranked (yielding lower rates of energetic return due to high processing costs) compared with both bush tomatoes. Of the two, Eragrostis is more likely to be human-dispersed. It is a tussock forming grass locally known as woolybutt (kunuruntu), which produces a tiny edible seed nearly identical to the Ethiopian domesticate teff; it was historically significant in the diet but very low ranked (low net caloric gain relative to its processing cost66) and is not currently collected except on rare occasions. This species could respond to contemporary human fire use and past foraging activities, as seed may have been dispersed near residential sites due to losses during transport and threshing/grinding activities, but such signals will be difficult to detect as the seed could also be dispersed by a wide range of other seed consuming organisms (ants), wind and water.
Scaevola parvifolia is least likely to be dispersed by people and as such, serves as our control species. It is a small herbaceous plant with an edible but very low-ranked tiny fruit about 0.5 cm in length and 1 g in weight, with a fleshy, gooseberry-like portion that comprises only a thin skin over a large, inedible internal seed. Often dismissed as kipara mayi by Martu people (bustard food), it is not eaten in any great numbers by adults, although children occasionally consume small amounts of fruit on the spot28. Formerly, it may have been consumed by the burrowing bettong, extinct in the region since the 1960s67. Of all four species, it is the least likely to be influenced by anthropogenic dispersal.
Hypotheses and predictions
If S. diversiflorum is dispersed by people, the intensity of landscape use in either the past, the present, or both should predict presence and/or abundance (see Table 1). During the pre-contact era for Martu peoples, residential camps were situated near water sources of variable permanency depending on season68,69. If past landscape use has had a significant effect on plant presence, plants should thus be more likely to be found close to occupation sites and where the nearest water source is more reliable. However, when patches of fruit were more distant than a kilometer or two from camp, people would often make a temporary logistical or foraging camp near the patch to remove seed in bulk68,70 before carrying the processed fruit back to the main residential site. Thus, for this species, the type of occupation site may also matter: this region is well surveyed, and hosts a wide range of site types covering the last several milennia68,71. Whether or not a site is used for short term occupation vs long term occupation could influence spatial variation in S. diversiflorum seed due to the way that people remove seed to increase the utility (proportion of load comprised of useful parts) of a transported load48. This is because processing of resources with inedible parts is subject to threshold effects, such that the distance one must move the item and the time to remove its inedible parts influences whether one will field-process the item or carry it home to be processed at camp. Because the processing time of diversiflorum is fairly quick (30 s per fruit) and the load utility (percent edible) increases dramatically with the removal of the inedible central portion (allowing the hollow fruit to be nested together), the transport threshold should be fairly short, meaning that people will be more likely to make a temporary camp near a fruit patch to clean and discard seed if the patch is farther than a kilometer or two from a water source. Patches should thus be more likely to be found near ephemeral or task-specific sites used mainly over the short term by smaller groups of people (minor site), compared to those sites with evidence of longer-term use by larger groups of people (major site).
Present landscape use should also predict presence/abundance. Plants should be more likely to be found or be at higher density in landscapes that are more intensively used for present day foraging activities. Our previous, systematic work on anthropogenic fire regimes has shown that the number of winter season fire footprints in a region is closely correlated with foraging effort, particularly for hunting monitor lizards (Varanus spp)49. If people are frequently hunting in a region, companion foods such as fruit may be more likely to be collected. Here hunting intensity as measured by fire density, is treated as a proxy for the general intensity of landscape use, which could be more accurate than travel distance from a central community, which we have previously used as a measure of the intensity of landscape use. Ethnographic observations of plant collection and processing should also provide evidence for the potential for seed dispersal, both past and present: places where S. diversiflorum was known to be processed in the past should have a higher probability of plant presence than nearby locations without such histories, and observations of foraging activities should reveal systematic discard of seed around camping locations.
Our predictions concerning past and present landscape use for S. centrale and Eragrostis differ from that of S. diversiflorum since the dispersal mechanisms are different. For both of these plants, there should be no effect of site type, as there is no field processing of either species. Second, the effects of past land use as inferred by site distance and water permanence rank (as proxies) will be weaker and therefore more difficult to detect. Ethnographic observations should show the potential for seed dispersal for both species, however our measures of contemporary landscape use may have little, if any, effect on either. Our control species, Scaevola, should show no effect of past or present landscape use, and ethnographic observations should reveal little if any potential for direct seed dispersal.
While seed dispersal via epi- and endozoochory are potential mechanisms for influencing plant distribution, changing disturbance regimes, most obviously through fire, may also play a role for some species. By burning small grassland patches more frequently, anthropogenic fire regimes constructed and maintained by Martu people in the Western Desert shape habitat structure by increasing local pyrodiversity (the diversity of patches at different time since fire). This reduces the scale of habitat patchiness compared to landscapes where lightning dominates, and promotes the abundance of many native animal species72. This reshaping of landscape vegetative diversity is a type of ecosystem engineering for many plant species, which may reduce the cost to people of acquiring early-mid successional plant resources73 and may also increase the availability of some culturally significant plant species17,18. Martu use of fire at a landscape scale creates an intermediate disturbance that could favor plant presence for all four species independently of seed dispersal mechanisms. Potential measures of anthropogenic landscape fire that distinguish human fire regimes at the landscape scale are time-since-fire diversity74,75 and at the patch scale, the seasonality of firing events (most anthropogenic fires occur in the cool dry season, while most lightning fires are in the hot summer months). Fire frequency increases under anthropogenic influence in some ecosystems (e.g., SE Australia76, but it is not clear whether it does in this case: prior work suggests that people increase the diversity of frequencies, e.g., increase both low and high fire frequencies at a landscape scale75,77. Here, we have no specific expectations for each species, but test against the null hypothesis of no effect of fire. While all our plants are clearly fire-dependent, lightning fires are a frequent source of landscape fire in this region, thus dependence on fire itself is not a clear indication of dependence on human fire regimes.
Ethics and inclusion
Research approvals from the Martu Prescribed Body Corporate (now Jamukurnu Yapalikurnu Aboriginal Corporation, the trustee for Martu native title rights) were first obtained in June 2000, renewed in July 2004, and again in 2023. All ethnographic work was conducted under human subjects approvals from Stanford University (#1915, 2005-2015) and Penn State University (#00011455, 2015–present). Informed consent was obtained from all participants.
While intermittent interaction between some Martu and Europeans began early in the 20th century, others avoided contact with settler-colonial invasion until the 1960s. Many remote-living Martu were first contacted in the 1960s during government welfare patrols tasked with evaluating the drop zone for inter-ballistic missiles that were eventually launched in the International Weapons Research Establishment, Australia’s principal contribution to Cold War efforts. The last groups left the desert in 1967, but reoccupation of Martu lands began in the 1980s with the establishment of three remote communities at Punmu, Parnngurr, and Kunawarritji. One of the authors who is Martu (DT) was born in Martu country before the hiatus and was involved in all aspects of the homelands movement; while another (PV) began archeological programs in collaboration with remote-living Martu families in the mid-1980s. Many of the archeological sites recorded by PV with Martu custodians in the 1980s have been re-occupied with the advent of the outstation movement. Both pre-contact archeological assemblages and contemporary plant use around the sites, were recorded with ethnobotanists28,68,69,78. RBB and DWB began work here in 2000 as Martu were preparing their Native Title claim, which was determined in 200231.
On return to their homelands, Martu reinvigorated their customary subsistence and landscape burning practices32,33, and many remote-living families continue to maintain fundamental aspects of their foraging livelihoods, typically spending about 25% of their days in hunting or gathering activities28,36,79,80,81,82,83,84. People often go out foraging on day trips to various locales, which includes establishing a temporary ‘dinner camp’ for the purpose of bush food preparation, cooking and consumption. We have previously reported quantitative details of foraging time allocation, labor decisions, bush food yields, return rates from regular hunting and gathering practices, the role of cultural burning in facilitating hunting strategies, the ecological role of these practices, and significant changes in subsistence over the last half-century32,33,36,3781,82,83,84,85,86.
While the three remote communities in the Martu Native Title lands have semi-permanent infrastructure—including housing, a primary school, a small shop, and a health clinic—it would be misleading to describe Martu as “settled” or even “semi-settled”. Most exercise extremely high mobility, shifting residence many times over the course of a year across the remote communities and throughout settlements and towns beyond the Native Title boundaries. The composition of Martu residential communities and foraging groups is thus constantly fluctuating87. Their mobility and foraging activities are vital in maintaining the landesque capital of their estates, fundamentally entangled in a web of both social and ecological interactions between people, fire, plants, and animals that make the landscape of country “ngurra-ra”, or homeland33. These interactions are material manifestations of the creative epoch (The Dreaming), in which people re-enact the process of creation through the maintenance and support of ecological networks32,52,,86,88.
The ethnographic and ecological work reported in this study began in June of 2000 by DWB and RBB as part of a long-term project centered on understanding the socio-ecological dynamics of human–environment interactions. As part of this project, we quantitatively recorded observations of time allocation, production, distribution, and cultural burning associated with a broad set of foraging activities and land use by Martu men and women. The foraging data used in this study were obtained by accompanying a foraging party leaving the community over a cumulative 800-day observation period between June 2000 and August 2017. Martu typically use vehicles to access the day’s foraging region. On arrival they establish a temporary foraging camp (referred to as mirrka ngurra) from which people usually depart on foot, solo or in small cooperative groups. Depending on season and habitat, women most often engage in burning patches of older growth spinifex grass to facilitate their monitor lizard hunting, while men focus much of their efforts in searching for bustards or hill kangaroo. After the day’s foraging, people gather back at the hearth of the foraging camp to process, cook, share, and consume the day’s catch before returning to the community in the evening32,33,34,36,,81,82,84,85,89,90. During foraging trips, we recorded all foraging time allocation, yields, and distribution of bush foods among all participants, along with focal follows of individual foragers recording all search, pursuit, capture, transport, and processing. These data comprise a total of 385 trip-days averaging 8 people per trip and 160 unique individuals (80F, 80M). We recorded Solanum harvesting time allocation, yields, and sharing on a total of 28 of these trips, although we participated in many more Solanum harvests than were recorded quantitatively.
Plant presence surveys
Occurrence records91,92,93,94 for all four species place the study area well within each species’ known range (Fig. 1C), with the caveat that our study area is very remote, roadless, and poorly sampled. Our own surveys show all four species to be common and widely distributed in spinifex sandplain habitat across the entire study area. Plant presence was assessed using pedestrian surveys of plant distributions conducted in June and July of 2003. These surveys employed ten 10 km × 10-m belt transects to assess presence, absence and abundance via stem counts (see Table 2). Transects radiated outward in two directions from locations centered at ten past and present camping places so that we could more accurately capture the effect of distance from site on plant presence and abundance. Transect location was also stratified according to landscape use intensity, with two transects near ethnohistoric sites in little used landscapes, two near sites in heavily used landscapes, and the remainder near sites in landscapes with moderate use. Stem counts (counting main stem only) were aggregated to 300 m2 in the analysis to more closely match the 30 m pixel resolution of the remotely sensed imagery, producing a total of 3130 plots of which 2924 were located in early-mid successional habitat.
We also assessed the potential for contemporary Martu fruit processing to result in newly established patches of S. diversiflorum fruit by surveying previously observed fruit consumption sites for the presence and abundance of S.diversiflorum (Fig. 1G). In 2018, we selected 12 foraging locations with central hearths used between 2002 and 2017, at which bush foods were consumed, and for which we had complete quantitative foraging records. Half of these sites (controls, n = 6) had no S.diversiflorum consumption recorded during use. At the 6 remaining sites we had detailed records of S.diversiflorum harvesting, consumption, and seed dispersal, and the hearth was located in an area where no plants were growing at the time of use. Each foraging camp dispersal survey consisted of one 50 × 50 m plot centered at the cooking hearth around which people were sitting as they ate, and one randomly placed comparison plot located at least 500 m from the hearth. To avoid conflating present and past land use effects, none of these survey sites was located near archeological or ethnohistoric sites as described below.
Measures of past landscape use: distance to ethnohistoric site, water permanence, and site type
To operationalize questions about how past legacies of land use may shape landscape-level distribution of key plants, we ask whether the locations of contemporary plant patches are affected by distance to both ethnohistoric and archeological sites7. Ethnohistoric sites are sites that are known in living memory as old residential sites; archeological sites are those that have a known archeological component whether or not people remember using them as residential sites in the past. We do not distinguish in our analysis between the two types of sites. Before their homeland exodus, Martu utilized hundreds of residential campsites and foraging camps within the study area. While visits to more remote sites declined during the mid-20th century hiatus, Martu resumed frequenting many of these sites upon returning to their homelands. Most of the sites in the study area are within daily travel range of the remote communities and are still utilized for both short-term foraging camps and longer multi-day camping trips. While the earliest use of most of these sites (ethnohistoric and archeological) is not known, the depth of Martu ancestral ties in this region is profound, with occupation of these landscapes extending well into the Late Pleistocene as far back as 48k years ago30,95. The sample of sites in this analysis were all in use prior to Martu exodus from their homelands in the 1960s, meaning that all have had at least some use within the last 70 years, and many still continue to be used in the course of regular foraging activities.
To evaluate the legacy effect of anthropogenic land use on contemporary plant distribution, archeological site locations from publicly available datasets (n = 192, see refs. 70,72,96) were supplemented with informant recall of residential and foraging camp locations (n = 179 sites), and water sources known on topographic maps likely to be used as camping places (n = 21 sites). To capture any missing and potentially significant residential sites, we also surveyed high-resolution satellite imagery within a 20 km radius of the survey location for any significant water sources not already recorded, mainly soaks, springs, pools and larger rockholes. Using expert knowledge, including our own on the ground observations, each water source was associated with a relative permanence value, 1: ephemeral, to 4: nearly year-round availability. Camping sites and potable water sources are highly correlated: out of 192 sites with known archeology, the median distance to the nearest water source was 296 m, and 90% of all sites in our database are within 3 km of water. We are confident in assuming that water sources for which little information is recorded are likely to be potential camping places. To cross-check the campsite surveys, we used an aerial photo mosaic from 1953 which provided coverage for 366 campsite locations. 86% of sites showed signs of use in the form of nearby anthropogenic fire mosaics. We employ those historic photo mosaics and interviews with elders to assist in classifying each site according to its type of use. Major sites (n = 74) were defined as such based on ethnographic knowledge of the location as a major meeting place, ceremonial site, or aggregation site and/or visible signs of extensive fire mosaics in 1953. Where detailed archeological mapping, sampling and dating (both absolute and relative) had occurred68 this was cross-checked with information from Martu elders. The remaining sites were assigned as minor sites (n = 376). Informant knowledge of such sites is likely to be biased to those locations frequented by the few people to remain in the desert post 1950. The only potential sites missing from this analysis would be those which are associated with water sources invisible on satellite imagery that were abandoned prior to the 1950s and for which there is no ethnographic or archeological knowledge. However, given that the plant sampling transects were solely within regions well known to informants and to many of the authors, it is unlikely that this bias affects our results. To obtain the distance to nearest ethnohistoric site from each survey location, we performed a nearest neighbor analysis in ArcGIS Pro v. 2.597) on the center points layer of each 30 m transect interval and the center points of the ethnohistoric site layer.
Measure of present landscape use: winter fire density
We measured hunting intensity by summing the density of remotely sensed winter season fire footprints as visible in Landsat imagery over a 30-year period prior to the survey date (1973–2003). We reconstructed the fire history of the study region using a time series of 75 30-m resolution Landsat 2–7 image mosaics taken at roughly 6 month intervals (October and April) between 1980 and 2003. Two additional time steps were available from 1973 and 1979, which were used to attach an approximate age to the fires visible in the 1980 base image. Between 1987 and 2003, surface reflectance normalized burn ratio imagery was available from USGS-ESPA, from which we constructed difference images to highlight only those fires burning in each time step. Between 1973 and 1987, fires were detected using the difference of band 4 between time steps. Fires were hand-drawn using the pixel-based region-growing algorithm in QGIS 3.1498 and each time step was converted to a raster image. For our landscape use proxy, only winter fires were counted as there are no non-anthropogenic sources of fire during that season. Winter fire footprints are small and easily represented by a single center-point; we converted the point layer to a raster fire density layer using a kernel density estimation with a 3 km radius. We then extracted fire density for each presence point by using a point sampling function to query the raster layer.
Fire covariates: time-since-fire, time-since-fire diversity, fire frequency, proportion of ground cover
Fire regimes were assessed using the Shannon diversity index of TSF age classes, which are calculating using remotely sensed 20-year fire history maps74. To construct the fire history maps in QGIS 3.14, we stacked each raster fire footprint between 1983 and 2003 with the most recent fires on the top layer. Each layer was given a value corresponding to the elapsed months since time zero in 6-month intervals (with Oct–April 1983 being time zero), such that the resulting image maps out the time since last fire of each landscape patch. Martu landscape use has a direct impact on Shannon diversity measures, increasing diversity close to communities where people forage and burn more actively74. Shannon diversity was measured at the 3 km scale in a circular region centered on each 30 m plot centroid as this scale best differentiates anthropogenic fire regimes49.
All four plant species also exhibit declining density with time-since-fire (see Table 2). TSF for each transect plot was assessed in two ways: on the ground, by experienced researchers estimating months since last fire, and secondly, by utilizing this ground classification in conjunction with fire history maps to convert the categorical classification and estimations into months since last fire. As the resolution of the satellite imagery used to derive the fire maps was a maximum of 30 m, and the classification maps were at best only 90% accurate, there were some discrepancies between ground and satellite-based estimation of TSF, especially at the fire boundary; these were resolved by using the ground estimation to adjust the boundary of each TSF age class.
Plant presence may also be affected by the frequency of fire, which is influenced by the interval between fires. Martu burning increases the diversity of fire frequencies, creating patches close to the community where fire is both more frequent and less frequent75. Fire frequency at the time of survey was estimated using the fire history maps, as the number of times burnt in the 30 years prior to the survey.
Time since fire is only a rough proxy of successional vegetation structure, as rainfall controls the rate at which plants grow after fire99,100. To account for this, we also added a covariate describing the relative proportion of green, brown and total vegetation sampled at each plot centroid. Green vegetation increases with rainfall and is dominant soon after fire, while brown vegetation consists primarily of senescent Triodia grass hummocks and dominates 5–10 years after fire. These raster layers were created in QGIS 3.14 from Digital Earth Australia 30 m Landsat-derived fractional cover datasets available through the Terrestrial Ecosystem Research Network101.
Soil covariates: soil moisture, rainfall, soil components, slope, and aspect
Soil water limitations combined with microgeological variation in nutrient availability are two fundamental drivers in the distribution of endemic plant species across Australia102. To help control for this variation in water and nutrient availability, we included several topographic elements (e.g., slope and aspect) and soil attributes (e.g., water-holding capacity, clay content and sand content) within our analysis. For the terrain features, we derived slope (degrees) and aspect (the compass direction of the slope) values at each sample location in ArcGIS Pro 2.5 from a 1 s SRTM digital elevation model (DEM, resampled to 30 m) provided through the USGS. Both slope and aspect have been shown to influence the distribution of plants in semi-arid regions through the creation of microclimates (e.g., soil temperature, evapotranspiration, wind speed, etc.), the alteration of soil properties (e.g., organic matter context, soil depth, texture, etc.), and the control of hydrological processes (e.g., runoff dynamics, soil water retention, etc.)103,104. Soil properties were available as a set of 30 m raster layers available through the Soil and Landscape Grid of Australia105. Those that we included in the initial parameter selection were organic carbon, available water capacity at 0–5 cm, percentage clay, silt, phosphorus, and sand.
To capture variability due to persistent soil moisture, we calculated NDMI (the normalized difference water index, a measure of vegetation moisture) as a long-term average measure (since 1987), using 30 m Landsat imagery available as a derived product from the EROS data archive. We also used cumulative rainfall for the 12 months prior to the survey date, drawing on 5 km scale gridded (raster) monthly rainfall datasets available from the Australian Bureau of Meteorology. The grids were resampled in QGIS 3.14 to 30 m, summed over the relevant 12-month period for each survey date and the total rainfall amount at each survey center point was extracted using the raster point sampling tool.
Analytical methods
We explore the anthropogenic and environmental factors influencing species distributions with an information theoretic model selection approach106 in R (v 4.2.2)107, predicting plant presence/absence using generalized linear mixed models (glmer) for the binomial distribution in the lme4 (v 1.1) package108. Because we had many, often collinear, non-anthropogenic climatic and geophysical covariates that could obscure the relationship with our anthropogenic predictors, our variable selection approach followed best practices outlined in ref. 109. We first screened the potential set of non-anthropogenic covariates for associations with the presence of each species, and then checked for any significant collinearities. This initial process identified a total of nine of the most informative, non-collinear soil/climate and fire-related covariates: soil carbon, percentage green ground cover, percent sand, long term average NDMI, elevation, percent total green+brown vegetation, time-since-fiire, fire frequency since 1973, and season of most recent fire (winter vs summer). These were retained, along with our anthropogenic predictors, in the model selection approach. All binomial presence models included a random effect (transect, n = 10) to control for spatial covariance in the transect plots; poisson models for abundance included individual-level random effects as recommended to reduce overdispersion110. After running each model and checking model diagnostics for spatial autocorrelation and fit, we used the MuMIn package (v 1.47.1) in R to calculate AIC, BIC, model weights and Nakagawa’s pseudo R2 values111.
The model selection procedure to identify the best models for both presence and abundance involved four steps. First, we ran models predicting species presence with only the random effect, then with each single covariate + random effect. Secondly, we combined all variables of each type (anthropogenic, fire, and soil) into three separate global models (considering only first order interactions), removing the poorest performing covariate until we reached a minimum AIC value. Third, we explored combinations of anthropogenic and fire variables, fire and soil variables, and anthropogenic and soil variables. Fourth, we combined all covariates and predictors into a single model, again removing the poorest performing until we reached a minimum AIC value. Details of all model parameters, AIC values and Akaike weights can be found in Supplementary Data 1.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Discover more from PressNewsAgency
Subscribe to get the latest posts sent to your email.