Seed dispersal by Martu peoples promotes the distribution of native plants in arid Australia

The study investigates whether and how highly mobile hunter-gatherer peoples act as seed dispersal agents for non-domesticated plants, and how such actions shape plant distributions and community structure outside of agricultural contexts. In Australia—where pre-contact subsistence centered on foraging and cultural burning—Aboriginal peoples are thought to have influenced the distribution of many economically and culturally significant plants via fire regimes, bioturbation, and seed scattering. Despite mounting evidence that sedentary agricultural communities can disperse useful wild plants, much less is known about dispersal by mobile foragers. Focusing on Martu people in the Western Desert (Western Australia), the authors test how past and present landscape use (as proxies for anthropogenic seed dispersal) and anthropogenic fire regimes affect the presence and abundance of four edible species: Solanum diversiflorum, Solanum centrale, Eragrostis spp., and Scaevola parvifolia (control). They predict strong anthropogenic dispersal signals for S. diversiflorum and Eragrostis, weaker dispersal for S. centrale (due to edible seeds and clonal growth), and no dispersal effect for Scaevola; all may respond to landscape fire.

Prior work on non-agricultural human-mediated plant dispersal has emphasized modern passive spread of weeds and indirect evidence of legacy effects near settlements and archaeological sites, including hyperdominance of useful taxa. In Australia, studies suggest human dispersal roles for Adansonia gregorii, Araucaria bidwillii, and Castanospermum australe. Ethnographic accounts report Aboriginal handling, caching, and transport of fruits (e.g., Solanum spp.), with anecdotal seed discard around camps and vehicle-enabled translocations. Broader work links Indigenous cultural burning to increased plant diversity and availability of culturally significant species, and archaeological studies show wild plant concentrations and potential human founder effects near sites. Few controlled studies have directly tested impacts of highly mobile hunter-gatherers on wild plant distributions; exceptions include increased densities of wild yams near Baka camps. This study addresses that gap for arid Australia.

Design: Hypothetico-deductive controlled comparison combining ethnographic observation, ecological field surveys, remote sensing, and statistical modeling to test anthropogenic seed dispersal and fire effects on plant distributions in Martu homelands (Great and Little Sandy Deserts; study subset ≈42,000 km²). Study species: Two bush tomatoes (Solanum diversiflorum with inedible seeds, Solanum centrale with edible seeds), seed grasses (Eragrostis setifolia/eriopoda), and control forb (Scaevola parvifolia). All occur in early–mid successional sandplain habitats. Ethnography: 800 residence days (2000–2017) with 385 foraging days recorded; focal observations of foraging, transport, processing, and sharing. Documented harvest and processing behaviors relevant to seed dispersal for each species (e.g., S. diversiflorum seed discard at camps, during vehicle travel, and around hearths; Eragrostis threshing/winnowing losses; limited potential for S. centrale and Scaevola dispersal by people). Dispersal-site survey: In 2018, surveyed 12 dinner camp hearths (used 2002–2017): 6 with known S. diversiflorum processing, 6 controls without. Each included a 50×50 m plot centered on the hearth and a paired comparison plot ≥500 m away. Transect surveys: In June–July 2003, ten 10 km × 10 m belt transects radiating from past/present camping places, stratified by landscape use intensity. Presence/absence and stem counts aggregated to 300 m² plots (n=3130 total; n=2924 early–mid successional used for modeling due to fire-dependence within ≈3.5 years since fire). Anthropogenic dispersal proxies (past/present land use): - Distance to nearest ethnohistoric/archaeological site (compiled from public datasets, informant recall, mapped water sources; n≈192 archaeological + 179 residential/foraging + 21 water sites; sites typed as major vs minor using ethnography and 1953 aerial mosaics). - Site type (major vs minor). - Water permanence (ephemeral to near-permanent; 1–4 scale) at nearest water source. - Contemporary land use intensity via winter fire density (proxy for hunting/foraging intensity): kernel density of winter season fire footprints (1973–2003) from 30 m Landsat time series. Anthropogenic fire (engineering) proxies: - Fire season (winter vs summer). - Fire frequency (number of burns in prior 30 years). - Time-since-fire (TSF) diversity (Shannon diversity of TSF age classes within 3 km). Environmental covariates and controls: - TSF per plot (field-estimated and reconciled with remote fire maps), green/brown/total vegetation fractional cover (TERN/Landsat), elevation (SRTM DEM), NDMI (long-term Landsat average), rainfall (12-month cumulative; 5 km BoM resampled), soil properties (Soil and Landscape Grid of Australia: organic carbon, available water capacity, % sand, clay, silt, P), slope and aspect. Statistical analysis: Information-theoretic model selection in R. Presence modeled with GLMMs (binomial; transect as random effect). Abundance (stems when present) modeled with GLMMs (Poisson; observation-level random effects to handle overdispersion). Three global model sets constructed: dispersal proxies only, engineering proxies only, and combined with environmental covariates. Stepwise AIC-based selection explored interactions (e.g., land use × water permanence, season × fire frequency). Reported standardized odds ratios (2 SD), 95% CIs, p-values, and Nakagawa’s R² for fixed vs fixed+random effects. Model diagnostics included checks for collinearity and spatial autocorrelation. Presence models restricted to plots burned ≤3.5 years prior (n=2429).

Ethnographic and dispersal-site evidence: - S. diversiflorum: Frequent seed discard during harvest, transport, and processing at hearths, along vehicle tracks, and in communities; plants established around recent hearths. In 2018 surveys, S. diversiflorum present at 83% (5/6) of past consumption hearths vs 2/6 control hearths; mean density at consumption sites 1.902 stems/ha [95% CI 0.961, 1.424] vs 0.251 stems/ha [95% CI -0.691, 1.931]. - S. centrale: High dietary value but low dispersal potential by people (seeds edible; minimal loss); likely dispersed by animals and via clonal growth. - Eragrostis: Moderate dispersal potential via seed loss during threshing/winnowing; observed tarpaulin use reduces but does not eliminate losses. - Scaevola: Very low human-dispersal potential; often noted as bustard food and seldom harvested. Transect presence by time since fire (n=3130 plots): - S. diversiflorum present in 7.06% (67) of 0–1.75 yr post-fire plots; 0% beyond 5.75 yr. - S. centrale: 4.11% (39) at 0–1.75 yr; 0% beyond 5.75 yr. - Eragrostis: 31.82% (302) at 0–1.75 yr; declines sharply thereafter. - Scaevola: 4.95% (47) at 0–1.75 yr; persists at low levels across TSF classes. Global models (Table 3): - Engineering model: • S. diversiflorum: Winter fires (OR 3.02 [1.48, 6.15]) and higher TSF diversity (OR 3.08 [1.75, 5.45]) increase presence. • S. centrale: Fire frequency strongly positive (OR 4.06 [1.48, 11.16]). • Eragrostis: Complex interactions; winter alone negative (OR 0.37 [0.28, 0.49]); season × fire frequency positive (OR 2.73 [1.56, 4.77]); TSF diversity negative (OR 0.56 [0.42, 0.74]). • Scaevola: Weak winter positive (OR 1.59 [1.04, 2.44]). - Dispersal model: • S. diversiflorum: Strong proximity to sites (site distance OR 0.27 [0.15, 0.49]); strong effect of land use (OR 17.01 [6.50, 44.50]); positive water permanence (OR 3.84 [1.04, 14.24]); strong interactions land use × water (OR 2836.02 [314.78, 25551.36]) and site distance × land use (OR 0.01 [0.00, 0.09]). Fixed-effects R² ≈ 44.9%. • S. centrale: Dispersal predictors weak; broad CIs. • Eragrostis: Weak negative with distance (OR 0.69 [0.49, 0.98]); some effects near major sites and ephemeral waters; interactions matter. • Scaevola: Presence more likely farther from sites (site distance OR 3.49 [1.94, 6.29]); however, wide CIs and uncertainty noted. Best presence models (Table 5): - S. diversiflorum: Site distance strong negative (OR 0.11 [0.05, 0.25]); land use positive (OR 6.37 [1.95, 20.81]); water permanence positive (OR 13.73 [4.78, 39.42]); major site negative (OR 0.08 [0.01, 0.41]); strong interactions: land use × season (OR 27302.32 [1282.01, 581445.45]); land use × water permanence (OR 23.88 [4.04, 141.09]); site distance × land use (OR 0.00 [0.00, 0.01]). Fixed-effects R² ≈ 0.52. - S. centrale: Fire frequency positive (OR 2.93 [1.19, 7.18]); soil carbon negative; vegetation cover positive; overall weaker dispersal signature. Fixed-effects R² ≈ 0.29. - Eragrostis: Strong effects of fire-season/TSF and interactions; site distance × land use (OR 0.03 [0.01, 0.09]) negative; land use × season (OR 11.60 [3.04, 44.20]) positive; time since fire strongly negative (OR 0.00 [0.00, 0.01]); several environmental covariates significant. Fixed-effects R² ≈ 0.30. - Scaevola: Farther from sites (site distance OR 2.32 [1.25, 4.30]); fire frequency negative (OR 0.16 [0.04, 0.61]); water permanence negative; vegetation cover components positive. Fixed-effects R² ≈ 0.32. Abundance models (Table 6; performance weaker than presence, but significant LR tests): - S. diversiflorum: Higher abundance near sites (site distance OR 0.21 [0.11, 0.41]) and near permanent water (OR 4.59 [2.88, 7.32]); soil carbon and total vegetation cover negative. Fixed-effects R² ≈ 37.8%. - S. centrale: Abundance associated with winter season (OR 0.24 [0.07, 0.79]) and lower with higher fire frequency (OR 0.18 [0.06, 0.52]); interactions with season. Fixed-effects R² ≈ 25.6%. - Eragrostis: Abundance higher in winter (OR 2.09 [1.48, 2.95]); strong interactions of site distance × water permanence (OR 4.54 [1.99, 10.38]) and TSF × site distance; positive with total vegetation cover and NDMI; negative with soil carbon. Fixed-effects R² ≈ 27.5%. - Scaevola: Higher abundance farther from sites (site distance OR 2.05 [1.05, 4.00]); interactions with TSF and fire frequency; some negative effects of soil carbon. Fixed-effects R² ≈ 33.9%. Overall: Strongest evidence for human-mediated dispersal in S. diversiflorum; Eragrostis shows dispersal signatures modulated by seasonal use of ephemeral waters and ritual aggregations; S. centrale patterns align more with fire (engineering) than dispersal; Scaevola shows no positive dispersal or fire effects and may be suppressed near human sites due to effects on animal dispersers.

The findings support the hypothesis that mobile Martu foragers act as significant seed dispersers of culturally important non-domesticated plants, especially Solanum diversiflorum. Ethnographic observations and dispersal-site surveys demonstrate substantial seed deposition linked to foraging and processing practices, resulting in successful plant establishment. Statistical models show S. diversiflorum presence and abundance concentrated near minor occupation sites, high contemporary land use, and permanent water—consistent with epizoochorous seed transport to camps and along travel corridors. Eragrostis distributions are consistent with human-mediated dispersal combined with fire-mediated habitat dynamics and seasonal social use patterns: higher presence/abundance near sites with ephemeral waters aligns with wet-season harvesting and ritual aggregations where heavy seed processing and loss would occur. In contrast, S. centrale—with edible seeds and clonal growth—shows weaker dispersal signals but strong positive associations with fire frequency, suggesting anthropogenic landscape engineering via cultural burning promotes its presence through vegetative reproduction. The control, Scaevola, exhibits reduced presence near sites, potentially due to reduced activity of nonhuman dispersers (e.g., bustards, bettongs) around human-visited areas. These results demonstrate that even unintentional behaviors during foraging and transport can produce lasting legacy effects on plant community structure in arid landscapes historically shaped by people. Recognizing human roles beyond intentional cultivation refines interpretations of cultural landscapes and informs conservation baselines that incorporate millennia of human–plant mutualisms.

This study provides controlled, multi-evidence support that Martu peoples’ foraging, transport, and processing practices disperse seeds and structure the distributions of key wild food plants in arid Australia. Strong anthropogenic dispersal effects were detected for Solanum diversiflorum and, in interaction with fire and seasonal land use, for Eragrostis. Solanum centrale distributions appear more influenced by anthropogenic fire regimes than by human seed dispersal. The control species, Scaevola parvifolia, shows no positive dispersal or engineering response near human sites. By integrating ethnography, ecological surveys, remote sensing, and model-based inference, the work highlights unintentional yet consequential human–plant mutualisms in non-agricultural contexts and underscores the need to include people and cultural burning in conservation and restoration strategies for rangelands and native Solanum. Future research should: (1) test genetic signatures of anthropogenic founder effects; (2) experimentally quantify seed survival and recruitment from specific dispersal pathways (hearths, tracks, communities); (3) refine multi-scale fire–vegetation models to predict plant responses under varying cultural burning regimes; and (4) assess how restoring animal dispersers and cultural practices together can optimize plant conservation.

- Dispersal-site survey sample size was small (n=12 hearths; 6 with S. diversiflorum processing), limiting precision. - Some model coefficients, especially for S. centrale and Scaevola dispersal proxies, had wide confidence intervals and high uncertainty. - Abundance models generally performed worse than presence models (wider CIs; more variance explained by random effects), limiting inference on density drivers. - Potential biases in the site database: remote, poorly sampled landscape; possible omission of older or less visible water-associated camps; informant recall biases. - Remote sensing constraints: 30 m Landsat resolution; TSF mapping accuracy ≈90%; discrepancies at fire boundaries required reconciliation with field estimates. - Collinearity and complex interactions among covariates (e.g., land use, water permanence, fire regime) complicate causal attribution. - Presence models restricted to plots burned ≤3.5 years prior to avoid confounding with successional declines; results may not generalize beyond early–mid post-fire stages.