Integrating models across temporal and spatial scales to simulate landscape patterns and dynamics in mountain pasture-woodlands

Context

Pasture-woodlands are semi-natural landscapes that result from the combined influences of climate, management, and intrinsic vegetation dynamics. These landscapes are sensitive to future changes in land use and climate, but our ability to predict the impact on ecosystem service provisioning is limited due to the disparate scales in time and space that govern their dynamics.

Objectives

To develop a process-based model to simulate pasture-woodland landscapes and the provisioning of ecosystem services (i.e., livestock forage, woody biomass and landscape heterogeneity).

Methods

We modified a dynamic forest landscape model to simulate pasture-woodland landscapes in Switzerland. This involved including an annual herbaceous layer, selective grazing from cattle, and interactions between grazing and tree recruitment. Results were evaluated within a particular pasture, and then the model was used to simulate regional vegetation patterns and livestock suitability for a ~198,000 ha landscape in the Jura Vaudois region.

Results

The proportion of vegetation cover types at the pasture level (i.e., open, semi-open and closed forests) was well represented, but the spatial distribution of trees was only broadly similar. The entire Jura Vaudois region was simulated to be highly suitable for livestock, with only a small proportion being unsuitable due to steep slopes and high tree cover. High and low elevation pastures were equally suitable for livestock, as lower forage production at higher elevations was compensated by reduced tree cover.

Conclusions

The modified model is valuable for assessing landscape to regional patterns in vegetation and livestock, and offers a platform to evaluate how climate and management impact ecosystem services.

     

Predicting carnivore occurrence with noninvasive surveys and occupancy modeling

Terrestrial carnivores typically have large home ranges and exist at low population densities, thus presenting challenges to wildlife researchers. We employed multiple, noninvasive survey methods—scat detection dogs, remote cameras, and hair snares—to collect detection–nondetection data for elusive American black bears (Ursus americanus), fishers (Martes pennanti), and bobcats (Lynx rufus) throughout the rugged Vermont landscape. We analyzed these data using occupancy modeling that explicitly incorporated detectability as well as habitat and landscape variables. For black bears, percentage of forested land within 5 km of survey sites was an important positive predictor of occupancy, and percentage of human developed land within 5 km was a negative predictor. Although the relationship was less clear for bobcats, occupancy appeared positively related to the percentage of both mixed forest and forested wetland habitat within 1 km of survey sites. The relationship between specific covariates and fisher occupancy was unclear, with no specific habitat or landscape variables directly related to occupancy. For all species, we used model averaging to predict occurrence across the study area. Receiver operating characteristic (ROC) analyses of our black bear and fisher models suggested that occupancy modeling efforts with data from noninvasive surveys could be useful for carnivore conservation and management, as they provide insights into habitat use at the regional and landscape scale without requiring capture or direct observation of study species.     

Connecting mountains and desert valleys for black bears in northern Mexico

Context

Black bear connectivity studies are scarce in the southern distribution where the species is endangered. The identification of corridors is a strategy to promote conservation in human-modified landscapes.

Objectives

Assess and validate long-distance corridors in the southern black bear distribution using resistance models, occurrence records, and radio-telemetry of an individual that dispersed between the Sierras Madres of Mexico.

Methods

We acquired black bear occurrence records from several sources and telemetry records from one dispersal individual in northern Mexico. We generated ensemble habitat suitability models and resistance landscape surfaces to generate cumulative resistant kernel and least-cost paths to identify connectivity core areas and corridors of importance through Natural Protected Areas. Finally, we assessed long-distance corridors.

Results

We developed three habitat suitability models for black bears southern range; one matches the current distribution of the species. When including radio-tracking records, the landscape resistance is reduced to arid sites with low habitat suitability. We used least resistance connectivity surfaces to merge subpopulations within each Sierra Madre. The long-distance corridor models indicate narrow routes that require individuals with plastic behavioral dispersal capacity. Almost 20% of the connectivity core areas are within Natural Protected Areas. These are the first large-scale corridors using resistance layers in the southern black bear distribution.

Conclusions

Corridors can be functional for a range of temperate and dry habitat species. Landscape connectivity models should include the monitoring of dispersal individuals to identify the plasticity of organisms and the tangible barriers for them.

     

Seismic Cutlines,Changing Landscape Metrics and Grizzly Bear Landscape use in Alberta

Besides providing habitat to the grizzly bear (Ursus arctos) and other wildlife, the Rocky Mountain foothills of Alberta, Canada hosts considerable mining, seismic oil and gas exploration and production, and forest harvesting activities. Worldwide, such human activities influence the configuration and composition of the landscape. We assessed seismic cutline effects on landscape structure and grizzly bear use during early summer of 1999 and 2000. We studied five female and two male bears, which were GPS-collared in the spring following den emergence. The area available to this population was stratified into 49 km² hexagon-shaped sub-landscapes. The scale of this stratification was determined by patterns of bear movement. Fourteen compositional and configurational landscape metrics were calculated within each landscape unit, and bear use points were pooled or ‘binned’ within each unit. Landscape use was related to landscape metrics using a Generalized Linear Model (GLM). We found that seismic cutline proportion did not explain landscape use by grizzly bears; however secondary effects of cutlines on landscape structure did. Declining use was mainly associated with increasing proportions of closed forest, and increasing variation of inter-patch distances, while use was mainly increasing with increasing mean patch size. An earlier investigation had demonstrated that adding seismic cutlines to grizzly bear habitat caused increases in the variation of inter-patch distances. Since the landscape structure of this grizzly bear population will continue to change as a function of increased levels of resource extraction activities in the near future, it is crucial to further study the detailed meaning of landscape structure at the large and small scale for effective conservation efforts.     

The use of herders’ accounts to map livestock activities across agropastoral landscapes in Semi-Arid Africa

Improved understandings of the agricultural and range ecologies ofsemi-arid Africa require better information on the spatiotemporal distributionof domestic livestock across agropastoral landscapes. An empirical GIS-basedapproach was developed for estimating distributions of herded livestock acrossthree agropastoral territories (around 100 km² each)over a two-year period. Algorithms developed from regression analyses of herdtracking data (with R²s 0.67) are used to transform a morecomprehensive but incomplete set of data generated from herders accounts oftheir herds grazing itineraries (400 herds following 6500 itineraries). Theresulting characterization registers 40 000 days of livestock activitiesacross694 land units (averaging 70 ha) over the study period. This studydemonstrates that rural producers knowledge of their daily extractionpracticescan be translated to fine-grained characterizations of extraction densitiesacross mixed landscapes. The spatiotemporal distribution of livestock that isrevealed by this approach diverges strongly from that predicted bycommonly-usedpoint-diffusion estimation procedures. Instead, the distribution reflects localpatterns of land use, topography, vegetation, settlements, and water points.Grazing and nongrazing times spent in land units are not spatially correlatedand the seasonality of grazing pressure is spatially variable. Therefore, theecological impacts of livestock grazing are spatially variable at fine scalesand there is a significant potential for livestock-mediated nutrient transfersacross agropastoral landscapes. The georeferenced data produced by thisapproachnot only will help evaluate the impact and sustainability of differentmanagement practices but also provides a strong empirical base for improvedspatial modeling of herded livestock.This revised version was published online in May 2005 with corrections to the Cover Date.     

Interaction of food resources and landscape structure in determining the probability of patch use by carnivores in fragmented landscapes

Studies on the distribution of mammalian carnivores in fragmented landscapes have focused mainly on structural aspects such as patch and landscape features; similarly, habitat connectivity is usually associated with landscape structure. The influence of food resources on carnivore patch use and the important effect on habitat connectivity have been overlooked. The aim of this study is to evaluate the relative importance of food resources on patch use patterns and to test if food availability can overcome structural constraints on patch use. We carried out a patch-use survey of two carnivores: the beech marten (Martes foina) and the badger (Meles meles) in a sample of 39 woodland patches in a fragmented landscape in central Italy. We used the logistic model to investigate the relative effects on carnivore distribution of patch, patch neighbourhood and landscape scale variables as well as the relative abundance of food resources. Our results show how carnivore movements in fragmented landscapes are determined not only by patch/landscape structure but also by the relative abundance of food resources. The important take-home message of our research is that, within certain structural limits (e.g. within certain limits of patch isolation), by modifying the relative amount of resources and their distribution, it is possible to increase suitability in smaller/relatively isolated patches. Conversely, however, there are certain thresholds above which an increase in resources will not achieve high probability of presence. Our findings have important and generalizable consequences for highly fragmented landscapes in areas where it may not be possible to increase patch sizes and/or reduce isolation so, for instance, forest regimes that will increase resource availability could be implemented. Electronic supplementary material The online version of this article (doi:) contains supplementary material, which is available to authorized users.     

An integration of habitat evaluation,individual based modeling,and graph theory for a potential black bear population recovery in southeastern Texas,USA

Population recovery is difficult for species that require large contiguous areas of habitat, particularly within areas of heterogeneous land ownerships. Ecologically, potential for recovery success requires assessment of quantity, quality, and distribution of available habitat. Our objective was to evaluate habitat for a possible Louisiana black bear recovery in southeastern Texas. First, we categorized land cover and identified remote areas of highly suitable habitat. Next, we used the individual based simulation model J-walk to estimate ability of female black bears to move among remote habitat patches. Then, we applied graph theory to J-walk output to evaluate overall connectivity of remote habitat. An estimated 225,626 ha of remote habitat were identified in 901 patches, most of which was located within the eastern half of the study area. Network analysis showed specific areas where targeted conservation efforts may help black bear population expansion throughout the study region. Ultimately, enough habitat area exists to sustain a black bear population and it is best connected among public and private lands largely within the eastern half of the study area. Habitat evaluation will need to be revisited if black bears establish themselves locally and actual habitat use data become available. Regardless, our analysis demonstrates an important first step that may be incorporated into a larger adaptive management framework, updated, and replicated as more-detailed habitat suitability and land use data are available.     

Scale dependence of felid predation risk: identifying predictors of livestock kills by tiger and leopard in Bhutan

Context

Livestock predation by tiger and leopard in Bhutan is a major threat to the conservation of these felids. Conflict mitigation planning would benefit from an improved understanding of the spatial pattern of livestock kills by the two predators.

Objectives

We aimed to identify the landscape features that predict livestock kills by tiger and leopard throughout Bhutan. Our goals were to: (1) identify the predictors that have the largest influence in determining livestock kills, (2) assess the influence of scale across the different predictors evaluated and identify the scale at which each was most important.

Methods

We used livestock kills obtained from compensation records of tiger (n = 326) and leopard (n = 377) across Bhutan between 2003 and 2012 to run predation risk models with MaxEnt algorithm, using a multi-scale modeling approach (1, 2, 4, 8 and 16 km).

Results

Human-presence (density of settlements and roads) and land-cover (percentage of tree cover and meadow patches) were the main variables contributing to livestock kills by both species. Livestock kills were likely driven by a trade-off between livestock density and predator ecology, and the balance of this trade-off varied with scale. Risk maps revealed different hotspots for tiger and leopard kills, and analysis showed both species preferentially killed equids over other livestock types.

Conclusions

Our results highlight the importance of evaluating scale when investigating the spatial attributes of livestock kills by tiger and leopard. Our findings provide guidance for reducing conflict between humans and large felids throughout the country.

     

The effects of the number,size and isolation of patches along a gradient of native vegetation cover: how can we increment habitat availability?

Habitat availability—or how much habitat species can reach at the landscape scale—depends primarily on the percentage of native cover. However, attributes of landscape configuration such as the number, size and isolation of habitat patches may have complementary effects on habitat availability, with implications for the management of landscapes. Here, we determined whether, and at which percentages of native cover, the number, size and isolation of patches contribute for habitat availability. We quantified habitat availability in 325 landscapes spread across the state of Rio de Janeiro, in the Atlantic Forest hotspot, with either high (>50 %), intermediate (50–30 %), low (30–10 %) or very low (<10 %) percentage of native cover, and for six hypothetical species differing in inter-patch dispersal ability. Above 50 % of native cover, the percentage of cover per se was the only determinant of habitat availability, but below 50 % the attributes of landscape configuration also contributed for habitat availability. The number of patches had a negative effect on habitat availability in landscapes with 50–10 % of native cover, whereas patch size had a positive effect in landscapes with <10 % of native cover. The different species generally responded to the same set of landscape attributes, although to different extents, potentially facilitating decision making for conservation. In landscapes with >50 % of native cover, conservation actions are probably sufficient to guarantee habitat availability, whereas in the remaining landscapes additional restoration efforts are needed, especially to reconnect and/or enlarge remaining habitat patches.     

Using diffusion models to simulate the effects of land use on grizzly bear dispersal in the Rocky Mountains

Timber harvests proposed for Trail Creek Watershed, in southwestern Montana, U.S.A., have been opposed because grizzly bear (Ursus arctos horribilis) dispersal from northern Montana wildernesses into the Greater Yellowstone Ecosystem may be less likely. We used an individual-based model to simulate grizzly bear responses to: 1) region-level management practices represented by ownership patterns, and 2) watershed-level changes in habitat availability due to proposed harvests and road building. We assigned permeabilities (i.e., values that represent how easily a bear can move through a patch) to ownership blocks (region-level) and habitat patches (watershed-level) based upon a literature review, and used a correlated random-walk diffusion model to simulate movements. Simulated bears were placed into rasterized landscapes in a stratified random manner. At the regional level, bears moved 1500 times (i.e., 1530 km), and their destinations were tallied. At the watershed level, the number of moves required for bears to leave the watershed were tallied. Sensitivity analyses were used to determine the variability of the results with respect to changes in some parameters of interest (i.e., permeabilities of private lands, harvest permeabilities, and disturbance indices).With the permeability of private land set at 50 (range: 0 to 99), simulated grizzlies did not disperse from the Scapegoat and Bob Marshall Wildernesses into Yellowstone National Park (0 of 10000 simulated individuals). Under the assumptions of this model, a linkage between the wildernesses in northern Montana and Yellowstone does not appear to exist. However, a significant number of simulated grizzlies (41%) dispersed from Anaconda Pintler Wilderness, which is near Trail Creek Watershed, into the wilderness ES in eastern Idaho. A linkage may exist between these sites.At the watershed-level, removal of forest habitat under proposed Harvest I (1.77% of the watershed cut) led to simulated grizzlies using slightly more moves (i.e., 5.6%, P=0.042) to exit the watershed than under existing conditions. Harvests of 3.5% of the watershed (planned Harvest II) did not alter the number of moves required to exit the watershed (P=0.068). When disturbances associated with roads and harvests were also examined, large increases in number of movements required to exit the watershed occurred ( 151%, P=0.002). These analyses suggest that grizzly bears would be disturbed while timber harvests were ongoing, but that long-term changes in movement would not occur if roads were closed following harvests. The analyses demonstrate the utility of applying individual-based diffusion models to landscape-level movements of animals, and identifies the need for telemetry studies to determine movement rates through specific habitats.     

Habitat structure mediates spatial segregation and therefore coexistence

Understanding the mechanisms driving diversity in nature is an important and ongoing challenge in our changing world. To efficiently protect ecosystem diversity it is crucial to explain why and how species coexist. Over the last decades models explaining species coexistence have increased in complexity but usually don’t incorporate a detailed spatial context. However, spatial structure has been shown to affect species coexistence and habitat deterioration is one of the biggest threats to biodiversity. We therefore explore a spatially explicit two-species model and assess the effects of habitat structure on species coexistence using a wide diversity of fractal landscapes. Each species is specialized in a particular habitat type. We find that landscape structure has a major influence on the stability and constitution of a two species system and may be sufficient to explain the coexistence of two species. Well connected and highly structured habitat configurations allow spatial segregation of both species and this decreases local interspecific competition; in our model this is the most important process stabilizing coexistence.     

Sex-specific connectivity modelling for brown bear conservation in the Carpathian Mountains

Landscape Ecology - Romania is currently one of the best-connected landscapes, with high-quality habitats and a high density of brown bears. However, regional development measures are needed for...     

Distinguishing between human-induced and climate-driven vegetation changes: a critical application of RESTREND in inner Mongolia

Changes in the spatiotemporal pattern of vegetation alter the structure and function of landscapes, consequently affecting biodiversity and ecological processes. Distinguishing human-induced vegetation changes from those driven by environmental variations is critically important for ecological understanding and management of landscapes. The main objectives of this study were to detect human-induced vegetation changes and evaluate the impacts of land use policies in the Xilingol grassland region of Inner Mongolia, using the NDVI-based residual trend (RESTREND) method. Our results show that human activity (livestock grazing) was the primary driver for the observed vegetation changes during the period of 1981?C2006. Specifically, vegetation became increasingly degraded from the early 1980s when the land use policy??the Household Production Responsibility System??led to soaring stocking rates for about two decades. Since 2000, new institutional arrangements for grassland restoration and conservation helped curb and even reverse the increasing trend in stocking rates, resulting in large-scale vegetation improvements in the region. These results suggest that most of the degraded grasslands in the Xilingol region can recover through ecologically sound land use policies or institutional arrangements that keep stocking rates under control. Our study has also demonstrated that the RESTREND method is a useful tool to help identify human-induced vegetation changes in arid and semiarid landscapes where plant cover and production are highly coupled with precipitation. To effectively use the method, however, one needs to carefully deal with the problems of heterogeneity and scale in space and time, both of which may lead to erroneous results and misleading interpretations.     

The Forgotten Rural Landscapes of Central and Eastern Europe

Interactions between nature and man – the underlying forces in landscape – have over time caused diversity. Usually, geographers and landscape ecologists deal with spatial diversity; in this paper, we would like to also consider temporal diversity. We argue that Central and Eastern European landscapes (using the examples of Estonia, Hungary, Poland and Slovenia) are much more diverse in time (layers) than Western European ones. This difference requires the use of different indicators in order to measure and study landscapes and special problems, threats, and possibilities of management and future development – but most important is the consideration of different perceptions. We also show that this diversity reduces the readability of landscapes, creating miscommunication and a transformation of meanings. We further argue that the link between humans and landscape is lost in Central and Eastern European countries due to temporal diversity, and that this link will be created anew in a globalizing world. To overcome alienation, we need slightly different classifications/typologies for each country in this region, with the aim of a sound future management of cultural landscapes.     

Species Dynamics in Disturbed Landscapes: When does a Shifting Habitat Mosaic Enhance Connectivity?

Although landscape ecology emphasizes the effects of spatial pattern on ecological processes, most neutral models of species–habitat relationships have treated habitat as a static constraint. Do the working hypotheses derived from these models extend to real landscapes where disturbances create a shifting mosaic? A spatial landscape simulator incorporating vegetation dynamics and a metapopulation model was used to compare species in static and dynamic landscapes with identical habitat amounts and spatial patterns. The main drivers of vegetation dynamics were stand-replacing disturbances, followed by gradual change from early-successional to old-growth habitats. Species dynamics were based on a simple occupancy model, with dispersal simulated as a random walk. As the proportion of available habitat (p) decreased from 1.0, species occupancy generally declined more rapidly and reached extinction at higher habitat levels in dynamic than in static landscapes. However, habitat occupancy was sometimes actually higher in dynamic landscapes than in static landscapes with similar habitat amounts and patterns. This effect was most pronounced at intermediate amounts of habitat (p = 0.3?0.6) for mobile species that had high colonization rates, but were unable to cross non-habitat patches. Differences between static and dynamic landscapes were contingent upon the initial metapopulation size and the shapes of disturbances and the resulting habitat patterns. Overall, the results demonstrate that dispersal-limited species exhibit more pronounced critical behavior in dynamic landscapes than is predicted by simple neutral models based on static landscapes. Thus, caution should be exercised in extending generalizations derived from static landscape models to disturbance-driven landscape mosaics.     

Land-use history and topographic gradients as driving factors of subalpine Larix decidua forests

European larch (Larix decidua Mill.) forests in the Alps are cultural landscapes that have been shaped by humans for centuries through traditional management. Biological and historical data sources were employed, and a multi-scale approach was adopted to capture the influence of factors affecting the structure of these forests. Landscape and stand scale dynamics were analyzed in four watersheds (c. 13,000 ha) of the western and central Italian Alps that have experienced different land-use intensities. Observed landscape changes were generalized using path analyses developed from a common conceptual model. Stand structure and a range of environmental variables were sampled in 203 circular plots, and land use and anthropogenic variables were derived from thematic maps and aerial photographs. We used multivariate statistical analyses (ordination and SEM models) to relate forest structure, anthropogenic influences, land uses, and topography. The most commonly observed land cover transition was an expansion of forests at the expense of open areas. All studied watersheds were dominated by larch forests, but their structure and spatial pattern differed greatly. Anthropogenic variables were less important at Ventina, the least accessible site, but emerged as fundamental to explain stand structure in the other study sites. Complexity of topography and proximity to roads had influenced past human activities mainly in the most accessible sites. Regeneration density was greatest at lower elevations and closer to human settlements. Quantification of the role played by forest harvesting and cattle grazing in past centuries is critical for understanding how global change factors may influence future dynamics of mountain forests in the European Alps and similar cultural landscapes worldwide.     

Identification of functional corridors with movement characteristics of brown bears on the Kenai Peninsula,Alaska

We identified primary habitat and functional corridors across a landscape using Global Positioning System (GPS) collar locations of brown bears (Ursus arctos). After deriving density, speed, and angular deviation of movement, we classified landscape function for a group of animals with a cluster analysis. We described areas with high amounts of sinuous movement as primary habitat patches and areas with high amounts of very directional, fast movement as highly functional bear corridors. The time between bear locations and scale of analysis influenced the number and size of corridors identified. Bear locations should be collected at intervals ≤6 h to correctly identify travel corridors. Our corridor identification technique will help managers move beyond the theoretical discussion of corridors and linkage zones to active management of landscape features that will preserve connectivity.     

Differentiating anthropogenic modification and precipitation-driven change on vegetation productivity on the Mongolian Plateau

Context

The Mongolian Plateau, comprising Inner Mongolia, China (IM) and Mongolia (MG) is undergoing consistent warming and accelerated land cover/land use change. Extensive modifications of water-limited regions can alter ecosystem function and processes; hence, it is important to differentiate the impacts of human activities and precipitation dynamics on vegetation productivity.

Objectives

This study distinguished between human-induced and precipitation-driven changes in vegetation cover on the plateau across biome, vegetation type and administrative divisions.

Methods

Non-parametric trend tests were applied to the time series of vegetation indices (VI) derived from MODIS and AVHRR and precipitation from TRMM and MERRA reanalysis data. VI residuals adjusted for rainfall were obtained from the regression between growing season maximum VI and monthly accumulated rainfall (June–August) and were used to detect human-induced trends in vegetation productivity during 1981–2010. The total livestock and population density trends were identified and then used to explain the VI residual trends.

Results

The slope of precipitation-adjusted EVI and EVI2 residuals were negatively correlated to total livestock density (R² = 0.59 and 0.16, p < 0.05) in MG and positively correlated with total population density (R² = 0.31, p < 0.05) in IM. The slope of precipitation-adjusted EVI and EVI2 residuals were also negatively correlated with goat density (R² = 0.59 and 0.19, p < 0.05) and sheep density in MG (R² = 0.59 and 0.13, p < 0.05) but not in IM.

Conclusions

Some administrative subdivisions in IM and MG showed decreasing trends in VI residuals. These trends could be attributed to increasing livestock or population density and changes in livestock herd composition. Other subdivisions showed increasing trends residuals, suggesting that the vegetation cover increase could be attributed to conservation efforts.

     

Blowing in the wind? Nutrient enrichment of remnant woodlands in an agricultural landscape

Increasing fertiliser use in agricultural landscapes is likely to threaten the viability of remnant native vegetation in many parts of the world. Australia’s prime grain production landscapes have nutrient poor soils, which formerly supported semi-arid woodland. The ecological function and capacity for regeneration of these remnants may be particularly susceptible to nutrient enrichment. The key sources of nutrients are wind and water deposition from crop fertilisation, and manure and feed from sheep. We hypothesised that these sources would result in unequal deposition of nutrients within and among remnant vegetation patches. We surveyed soil nutrients (Total N, Available P and K, C:N ratio, and soil pH) in the edges and interiors of 60 remnant woodland patches of various sizes, and in adjacent cultivated paddocks. Nutrient load was negatively correlated with remnant size and patterns were particularly strong for available P. Small remnant patches (<3 ha) were accumulation zones for nutrients, with levels comparable or higher than within crop lands. The patterns are consistent with the hypothesis that small remnants are strongly enriched as a result of being used for livestock shelter. In larger remnants, the primary cause of enrichment is consistent with edge accumulation of nutrients due to wind and water movement. In large patches, remnant edges, particularly the windward edge, were elevated compared to interiors of large patches. In these semi-arid crop lands, current trends in intensification of cropping and a shift away from livestock may reduce the input of nutrients to small patches but increase the nutrient threat to larger remnants.     

Tree-grass coexistence in a flood-disturbed, semi-arid savanna system

The coexistence of trees and grasses in savanna ecosystems is a contentious phenomenon. Fire and herbivory disturbances are often cited as major structuring forces that create a sustainable tree–grass relationship. However, periodic flooding of savanna patches may also enable coexistence. The aim of this study was to investigate the effects of flood-disturbance on the recruitment patterns of Acacia karroo trees in a semi-arid savanna system in South Africa. We analysed the spatial coincidence of A. karroo seedlings with tussocks of the tall spiny grass Stipagrostis namaquensis in the riverbed and related herbivory intensity to spatial position. The data showed that A. karroo seedlings were significantly positively associated with S. namaquensis (Chi-square test, $ \chi_{1}^{2} = 4 5. 20 $ , n = 118, P < 0.001); A. karroo seedlings growing inside of tussocks experienced less browsing pressure than those growing in the floodplain (Kruskal–Wallis test, H = 11.90, n = 118, P < 0.01); and recruitment success of A. karroo trees was spatially discrete (K–S test, D = 0.78, n = 196, P < 0.01). We suggest that floods create an enemy-free zone, which S. namaquensis colonises and then facilitates successful A. karroo establishment. High levels of A. karroo recruitment in the riverbed may replenish the woodlands fringing the river, which appear to be sink areas for A. karroo seedlings. Thus, the interaction between disturbances at different spatial and temporal scales (flooding versus herbivory) seems to maintain the inherently unstable coexistence of tree and grass species in this ecosystem. These findings also suggested that flood disturbances alter the tree–grass relationship.