Aim

Herbivore exclusion is implemented globally to recover ecosystems from grazing by introduced and native herbivores, but evidence for large-scale biodiversity benefits is inconsistent in arid ecosystems. We examined the effects of livestock exclusion on dryland plant richness and reproductive capacity.

Location

Central Australia.

Methods

We collected data on plant species richness and seeding (reproductive capacity), rainfall, vegetation productivity and cover, soil health, and herbivore grazing intensity from 68 sites across 6500 km² of arid Georgina gidgee (Acacia georginae) woodlands between 2017 and 2020. Sites were on an actively grazed cattle station and two destocked conservation reserves. We used structural equation modelling to examine indirect (via soil or vegetation modification) versus direct (herbivory) effects of grazing intensity by two introduced herbivores (cattle, camels) and a native herbivore (red kangaroo), on seasonal plant species richness and seeding.

Results

Soil health and rainfall were the strongest drivers of variation in richness and seeding. Cattle and camel grazing indirectly led to lower seasonal richness and seeding by reducing soil health. Kangaroos had a small but negative direct impact on richness, but no impact on soil health. Both introduced and native herbivores reduced annual chenopod shrub richness and seeding, whereas only cattle directly reduced perennial shrub richness and seeding. Camels indirectly reduced perennial shrub richness by impacting shrub abundance. Introduced herbivores reduced native grass richness and seeding indirectly via impacts on soil health, whereas forbs responded positively to cattle and camel activity.

Main conclusion

Considering indirect impacts improves evaluations of the effects of disturbances on biodiversity, as focusing only on direct effects can mask critical mechanisms of change. Our results indicate substantial biodiversity benefits from excluding livestock and controlling camels in drylands. Reducing introduced herbivore impacts will improve soil and vegetation condition, ensure reproduction and seasonal persistence of species, and protect native plant diversity.

Site background and selection

Our study occurred across 6600 km² of the north-eastern Simpson Desert, central Australia, on three properties, Ethabuka Reserve (23°51’S, 138°28’E, 2330 km², cattle removed 2004), Pilungah Reserve (23°8’S, 138°26’E, 2130 km², cattle removed 2006, previously known as Cravens Peak) and Carlo Station (23°29’S, 138°32’E, 2110 km², still grazed by cattle). The average annual precipitation of nearby long-term weather stations is 198 mm (Bedourie; 1933–2020) to 211 mm (Boulia; 1887-2020) (Bureau of Meteorology, 2021), but rainfall is highly variable and years with below-average rainfall are more frequent than years with above average rainfall (Dickman, Mahon, Masters, & Gibson, 1999). The area is on the eastern edge of the Simpson Desert, and is dominated by linear sand dunes 5–20 m high vegetated with spinifex (Triodia basedowii E.Pritz.) grassland and scattered shrubs. Georgina gidgee (Acacia georginae F.M.Bailey) woodlands occur patchily on clay swales between dunes as well as on calcareous and alluvial soils fringing the dunefields, with typical patch sizes ranging from 0.5 ha to 10 ha. We selected 17 gidgee woodland patches across the three properties. Patches were selected to represent a range of grazing intensities but located specifically to avoid areas immediately adjacent to watering points, which are often extremely heavily grazed. Each patch had four sites set up at 300-m intervals along a 1 km line, resulting in 16 sites on Carlo Station (in 4 patches; site selection was limited due to the density of artificial watering points in relation to accessible gidgee patches), 24 sites on Pilungah (in 6 patches) and 28 sites on Ethabuka (in 7 patches). Each site comprised a 100-m-long transect. Sites were monitored during 4 trips: September 2018, May and September 2019, and September 2020. Appropriate permissions and licences to conduct the fieldwork were obtained from the Queensland Government (Permits WITK15192514 and WISP15192514).

Grazing intensity

To assess the recent relative abundance of large herbivorous mammals we conducted dung surveys every monitoring session. All dung was removed from transects after counting, ensuring that transects would be clear for the next visit. 

Soil health

At each site, we measured soil surface attributes and ground cover characteristics along a 100-m line transect, with the direction of each transect determined randomly. We took a measurement every 5 m along the transect, resulting in 20 measurements per transect. Plant foliage projected ground cover (vegetation <0.5 m tall) and litter cover (including dropped seeds and dead vegetation) were estimated within 0.5 m x 0.5 m quadrats and recorded as a percentage of the quadrat area. At each quadrat, a pocket penetrometer (Handheld Penetrometers, QA Supplies, Norfolk, VA) was used to measure the strength of the soil surface. For each variable, we averaged all 20 measurements to derive an overall value per 100-m transect per trip per variable. An index of soil health was calculated as the arithmetic mean of the standardized (z-transformed) values of the three surface attributes (soil strength, plant foliage projected ground cover and litter cover) per site per trip.

Woody cover

At each site, we measured gidgee stand structure along a 100-m belt transect. For 68 of the transects, the belt width was 10 m (1000 m²). For the remaining 16 transects the belt was increased to 20 m (2000 m2) to enable at least ten trees to be measured (these transects were more sparsely vegetated and distributed across all three properties). For every tree on the transect, we measured tree canopy diameter at 1.5 m from the ground and canopy density (% cover, visual estimation). We converted canopy diameter to area, representing total canopy cover per tree, and multiplied this by its canopy density to create a derived relative cover value. Relative cover was summed for all trees in the transect to derive a final cumulative canopy cover variable per site. We also quantified the relative abundance of perennial shrubs per 1000 m² at each site, by counting all perennial shrubs along the same 10 m by 100 m belt as the tree structure monitoring.

Plant richness and reproductive capacity

To quantify plant richness, we identified every plant species along the 100-m x 10-m belt transect each trip. To quantify changes in plant condition over time, we estimated fruiting activity each trip for all plants along the 100-m belt transect using a scoring system. An index of reproductive output for each plant species per plot was recorded at each transect. All plant species that were fruiting or seeding were scored on a rank scale for the presence of fruits/seeds, from 0 (fruit/seed absent) to 5 (fruit/seed profuse, all individuals seeding). Each plant species received a relative abundance score (1–5, where 1 is very low abundance/a few individuals present and 5 is relatively high abundance). Diaspores that would be dispersed were recorded as ‘seeds’ and included fruits (e.g. pods on peas rather than the seeds within each fruit) and single seeds. We calculated a reproductive capacity score for each species by multiplying its fruiting or seeding score by its relative abundance score and summed individual species scores to derive a single fruiting score per transect. For example, if on Transect A, species X was given a seeding index of 4 with a presence of 3, then its relative seeding activity would be 12. If on the same transect, two other species were recorded, Y and Z, with relative abundance scores of 10 and 8 respectively, then the seeding activity for transect A would be 30 (12+10+8). The cumulative summed seeding activity scores for each site per trip were then used in statistical analyses.

Rainfall and Productivity

Rainfall was the cumulative rainfall for the previous 12 months before the survey at the closest weather station (there are 12 weather stations across the three properties). Productivity was estimated using the mean greenness for the site over the previous 180 days, derived from the green fraction of a timeseries of single date fractional cover data (Joint Remote Sensing Research Program, 2021). We also calculated a long-term rainfall variable (average of the annual cumulative rainfall over the past 20 years).