A pilot project testing the effectiveness of three weed control methods on the removal of Lantana camara in Forty Mile Scrub National Park,Queensland, Australia

Lantana (Lantana camara) poses a serious threat to the biodiversity of the dry rainforest vegetation at Forty Mile Scrub National Park, Queensland, Australia, by outcompeting native species and increasing vulnerability to fire. This pilot study tests the effectiveness of three weed control methods (hand pulling, a glyphosphate‐based foliar spray herbicide and a picloram‐ and triclopyr‐based basal bark herbicide) in removing lantana and their success in reducing lantana fuel loads. The foliar spray herbicide was the most effective in killing lantana, while manual pulling resulted in the largest decrease in fuel height. We suggest that foliar spraying will be most efficient for combating large infestations of lantana, while hand pulling techniques are recommended for creating firebreaks or when minimizing damage to native species is paramount.     

Spatial pattern of dry rainforest colonizing unburnt Eucalyptus savanna

Abstract The spatial pattern of dry rainforest and savanna tree species was analysed in a 1.56‐ha plot within an unburnt eucalypt savanna woodland in north Queensland, Australia. Rainforest colonization constituted only 1.3% of the basal area and mostly consisted of individuals less than 3 m high. The distribution of rainforest trees was highly clumped around the large savanna eucalypt trees. Ecological mechanisms generating the clumped distribution are discussed in light of evidence from this study and the literature. Herbaceous biomass was not reduced under trees, suggesting that relief from grass competition has not favoured rainforest colonization under tree crowns. Edaphic facilitation through nutrient enrichment under savanna tree crowns appears to be only minor on the moderate fertility soils of the area. The highly clumped pattern of colonizing dry rainforest may be a consequence of seeds dropped from birds roosting in savanna trees.     

Structure and biomass along an Acacia zanzibarica woodland–savanna gradient in a former ranching area in coastal Tanzania

Questions: What factors influence the density, size and growth form of trees in secondary Acacia zanzibarica woodlands on a former humid savanna rangeland? How does tree density relate to variation in tree foliage and spines, and woody and grass biomass? Location: Tropical coastal Tanzania (former Mkwaja Ranch, now in Saadani National Park). Methods: We surveyed 97 circular plots (4‐m radius) representing a gradient from open savanna to dense woodland. Within each plot, we measured all trees and estimated the biomass of spines. Foliage biomass of tree and grass layers was estimated on three occasions, twice during the wet season and once in the dry season. Soil samples were taken from each plot and analysed for texture and nutrient content. Interrelationships among various variables were investigated using linear multiple regression and mixed effects models. Results: Tree densities were highest on more nutrient‐rich, heavy soils. Spinescence was highest on trees in open savanna. Biomass of tree foliage in the wet season was best explained by numbers of ant nests and tree live‐wood ratio. Foliage biomass in the dry season was less than half that in the wet season and best predicted by grass biomass. Variables related to biomass of the grass layer were strongly influenced by fire; living grass biomass also decreased with increasing tree density. Conclusions: A. zanzibarica is a tree with a high water demand, and the association with heavy soils is probably due to greater availability of water on these sites. Establishment of A. zanzibarica woodlands significantly reduced grazing resources at Mkwaja Ranch. Under post‐ranching conditions, however, fires and soil conditions predominate. The woodlands may, therefore, represent a transient state of woody density in a still resilient humid savanna.     

Holocene boundary dynamics of a northern Australian monsoon rainforest patch inferred from isotopic analysis of carbon, (14C and δ13C) and nitrogen (δ15N) in soil organic matter

Abstract Soil organic matter (SOM) was sampled from lateritic soil profiles across an abrupt eucalypt savanna–monsoon rainforest boundary on the north coast of Croker Island, northern Australia. Accelerator mass spectrometry dating revealed that SOM that had accumulated at the base of these 1.5 m profiles had a radiocarbon age of about 5000 years. The mean carbon and nitrogen stable isotope composition of SOM from 10 cm deep layers from the surface, middle and base of three monsoon rainforest soil profiles was significantly different from the means for these layers in three adjacent savanna soil profiles, suggesting the isotopic ‘footprint’ of the vegetation boundary has been stable since the mid Holocene. Although there were no obvious environmental discontinuities associated with the boundary, the monsoon rainforest was found to occur on significantly more clay rich soils than the surrounding savanna. Tiny fragments of monsoon rainforest and abandoned ‘nests’ (large earthen mounds) of the orange‐footed scrubfowl, an obligate monsoon rainforest species, occurred in the savanna, signalling that the rainforest was once more extensive. Despite episodic disturbances, such as tropical storm damage and fires, the stability of the boundary is probably maintained because clay rich soils enable monsoon rainforest tree species to grow rapidly and achieve canopy closure, thereby excluding grass and reducing the risk of fire. Conversely, slower tree growth rates, grass competition and fire on the savanna soils would impede the expansion of the rainforest although high rainfall periods with shorter dry seasons may enable rainforest trees to grow sufficiently quickly to colonize the savanna successfully.     

Mycorrhizas in the Kakadu region of tropical Australia

This research represents the first part of a study which aimed to characterize the role of mycorrhizal associations in undisturbed and disturbed habitats in the Alligator Rivers Region of the Northern Territory of Australia. This is a seasonally dry tropical region with a climate consisting of a long dry season and a monsoonal wet season. Intact soil cores were sampled from 22 sites in this region, representing eucalypt savanna woodland, wetland, rocky hill and rainforest habitats. Clover, sorghum and eucalypt seedlings were grown in these cores in bioassays to measure the inoculum potential of vesicular-arbuscular mycorrhizal (VAM) and ectomycorrhizal (ECM) fungi. Propagules of VAM fungi were concentrated in the surface horizon, and were not adversely affected by 6 months dry storage of soil. Bioassays detected VAM fungus propagules at all sites, but these were less numerous in three sites with sparse herbaceous vegetation (a shrub-dominated woodland site, a sandstone area and a disturbed gravel pit without topsoil), than in other woodland sites. Propagules of VAM fungi were particularly numerous in soil from a rainforest habitat, which had much denser plant cover than any of the savanna sites. Propagules of ECM fungi colonized eucalypt seedling roots in some cores from all sites, except two wetland areas and a disturbed area without eucalypt trees. Physical and chemical properties of soils varied between sites and some properties (texture, organic carbon, etc.) were correlated with the inoculum potential of VAM fungi.     

Vegetation,fire and soil feedbacks of dynamic boundaries between rainforest,savanna and grassland

At fine spatial scales, savanna‐rainforest‐grassland boundary dynamics are thought to be mediated by the interplay between fire, vegetation and soil feedbacks. These processes were investigated by quantifying tree species composition, the light environment, quantities and flammability of fuels, bark thickness, and soil conditions across stable and dynamic rainforest boundaries that adjoin grassland and eucalypt savanna in the highlands of the Bunya Mountains, southeast Queensland, Australia. The size class distribution of savanna and rainforest stems was indicative of the encroachment of rainforest species into savanna and grassland. Increasing dominance of rainforest trees corresponds to an increase in woody canopy cover, the dominance of litter fuels (woody debris and leaf), and decline in grass occurrence. There is marked difference in litter and grass fuel flammability and this result is largely an influence of strongly dissimilar fuel bulk densities. Relative bark thickness, a measure of stem fire resistance, was found to be generally greater in savanna species when compared to that of rainforest species, with notable exceptions being the conifers Araucaria bidwillii and Araucaria cunninghamii. A transect study of soil nutrients across one dynamic rainforest – grassland boundary indicated the mass of carbon and nitrogen, but not phosphorus, increased across the successional gradient. Soil carbon turnover time is shortest in stable rainforest, intermediate in dynamic rainforest and longest in grassland highlighting nutrient cycling differentiation. We conclude that the general absence of fire in the Bunya Mountains, due to a divergence from traditional Aboriginal burning practices, has allowed for the encroachment of fire‐sensitive rainforest species into the flammable biomes of this landscape. Rainforest invasion is likely to have reduced fire risk via changes to fuel composition and microclimatic conditions, and this feedback will be reinforced by altered nutrient cycling. The mechanics of the feedbacks here identified are discussed in terms of landscape change theory.     

Diet selection and density estimates of forest buffalo in Campo‐Ma’an National Park,Cameroon

We studied diet selection and density of forest buffalo in the Campo Ma’an National Park of southern Cameroon. The buffalo’s diet in this rainforest comprised 43% grass, including 15%Leptochloa caerulecens. Other species eaten were non‐graminoid monocots (21.3%), mainly Commelinaceae (18.2%), including Palisota spp. and dicotyledons (32.7%), mainly leaves (26.5%). This diet revealed that buffalo collect their food on road verges, logging tracks and along large rivers. This agrees well with the distribution of buffalo tracks in the study area. Mean buffalo density in the forest was only 0.01 buffalo km^?2, whereas the density on the road together with its direct surroundings was 0.4 buffalo per km of road. We estimated the total number of buffalo in the 650 km² southern part of National Park Campo‐Ma’an of only twenty individuals. We suggest that the buffalo’s ancestral niche was not a primeval type of rainforest, as suggested by Kingdon, but an interface between savanna and rainforest. The Campo‐Ma’an buffalo most likely depend to a very large extent on anthropogenic vegetation types. Maintaining some grass cover by annual clear‐cutting along roads and on logging tracks might be necessary for the survival of the buffalo population within the National Park.     

Feeding ecology of Lilian’s Lovebird Agapornis lilianae in Liwonde National Park,Malawi

Lilian's Lovebird Agapornis lilianae is a small, near-threatened parrot resident in mopane Colophospermum mopane woodlands. We investigated its diet and foraging behaviour in Liwonde National Park, Malawi. We expected that Lilian's Lovebirds would show little specialisation for a particular food source but generally feed on available seeds, fruits, flowers and other items as observed in other lovebirds. Lilian's Lovebirds fed on 30 different plant species. Lilian's Lovebirds were observed feeding in six habitat types in Liwonde National Park and adjacent areas during the wet season, and four in the dry season. In the wet season lovebirds (23% of observations) foraged in grassy wetland (dambo) areas the most, whilst in the dry season they foraged in grasslands with tree cover (18%) the most. In mopane woodland, foraging flock sizes differed significantly between the wet (mean = 19.8 ± 1.0 lovebirds) and dry season (mean = 33.6 ± 2.3 lovebirds). Grass seeds were lovebirds’ main food source from December to June. The nutritional analysis of preferred foods showed that grass seeds have a relatively high protein and energy content. Grass seed availability is reduced with savanna burning and so early season burning (before May–June) in areas in and outside the park is not recommended.     

元江干热河谷植物叶片解剖和养分含量特征

研究了元江干热河谷旱田植物(旱季可浇灌,水分较好)和山坡半萨王纳植被中(自然状况,水分较差)共20种的叶片形态解剖特征,以及7种山坡植物叶片养分含量特征.结果表明,山坡植物叶片比叶重大,气孔密度大,气孔长度小,海绵组织/栅栏组织的值小等.元江干热河谷山坡植物叶片养分含量低,1.3%＞Ca＞N＞K＞1%＞Mg＞P＞S.除氮元素外,其它元素种间差别1～3倍.与热带植物群落叶片养元素含量相比,热带雨林＞元江山坡植物＞东南亚沙地旱生林和巴西矮卡廷加群落,表明元江干热河谷植物叶片具有明显的旱生性形态解剖特征,且叶片养分含量也较低.     

Landscape‐scale effects of herbivores on treefall in African savannas

Herbivores cause treefalls in African savannas, but rates are unknown at large scales required to forecast changes in biodiversity and ecosystem processes. We combined landscape‐scale herbivore exclosures with repeat airborne Light Detection and Ranging of 58 429 trees in Kruger National Park, South Africa, to assess sources of savanna treefall across nested gradients of climate, topography, and soil fertility. Elephants were revealed as the primary agent of treefall across widely varying savanna conditions, and a large‐scale ‘elephant trap’ predominantly removes maturing savanna trees in the 5–9 m height range. Treefall rates averaged 6 times higher in areas accessible to elephants, but proportionally more treefall occurred on high‐nutrient basalts and in lowland catena areas. These patterns were superimposed on a climate‐mediated regime of increasing treefall with precipitation in the absence of herbivores. These landscape‐scale patterns reveal environmental controls underpinning herbivore‐mediated tree turnover, highlighting the need for context‐dependent science and management.     

Fire monitoring in savanna ecosystems using MODIS data: a case study of Kruger National Park, South Africa

The heterogeneity of savanna ecosystems is guaranteed by disturbance events like fires, droughts, floods and browsing and grazing by herbivores. For conservation areas with limited space to preserve biodiversity, fire monitoring is crucial. Long periods of satellite remotely sensed data provide an alternative solution to estimate the distribution of different vegetation types and fire-affected patches over time. This study focusses on the application of MODIS data to detect, identify and delineate fire-affected areas in Kruger National Park (KNP), South Africa, for the period 2001–2003. Fire scars on KNP’s savanna were identified using threshold and supervised classification methods on moderate-resolution imaging spectroradiometer (MODIS) with 500-m resolution and 32-day global composites using a combination of band 1 (red), 2 (NIR, near infrared), 4 (green) and 6 (SWIR, short wave infrared). On identified fire scars, the spectral indexes of albedo, normalised difference infrared index (NDII) and normalised difference vegetation index (NDVI) were extracted. The following four broad habitat types were used for this analysis: riparian woodland, dense woodland, mixed woodland and open-tree savanna. The values of albedo, NDII and NDVI during the dry season (June to October) for different years are lower on fire-affected patches. Mixed woodland is the largest habitat burned with 21%, 43% and 2% of the KNP area affected by fire in 2001, 2002 and 2003, respectively. Riparian woodland is the least affected by fire. The supervised classification method has a greater accuracy for fire scars detection in KNP savannas during the dry season. We conclude that MODIS data can be used successfully for fire monitoring in savanna ecosystems.     

Interactive response of herbivores,soils and vegetation to annual burning in a South African savanna

Abstract Despite decades of research, the primary factors determining savanna structure remain elusive – a conundrum termed ‘the savanna problem’. After 47 years of annual burning in Terminalia woodland and Acacia/Combretum savanna on sandy, granite‐derived soils in the southern Kruger National Park, South Africa, a dense cover of trees and shrubs persists on some burnt plots and is largely absent from others. We postulated that intense browsing pressure by antelope and other herbivores prevents recruitment of trees in burnt plots and that herbivores concentrate on plots that are richest in nutrients. Herbivore abundance did not show a relationship with soil macronutrients and we consequently investigated micronutrient status. The reduction in tree cover as a result of annual burning was positively correlated with mass of herbivores (15–1500 kg) (r ² = 0.61, n = 8). This index of herbivore abundance was in turn positively correlated with total Zn (r ² = 0.64, n = 8). Other indices of herbivore abundance showed significant relationships with total clay content and total Mn. We suggest that herbivores concentrate on sites with greater clay content (possibly due to a greater availability of micronutrients), and that tree cover can remain relatively dense under a regime of annual burning if browsing pressure is not intense. The long‐term burn experiments in the Kruger National Park savanna provide a platform for unravelling the savanna problem. Determining possible interactions between soil properties, herbivory and fire is a step in this direction.     

Seasonal leaf dynamics across a tree density gradient in a Brazilian savanna

Interactions between trees and grasses that influence leaf area index (LAI) have important consequences for savanna ecosystem processes through their controls on water, carbon, and energy fluxes as well as fire regimes. We measured LAI, of the groundlayer (herbaceous and woody plants <1-m tall) and shrub and tree layer (woody plants >1-m tall), in the Brazilian cerrado over a range of tree densities from open shrub savanna to closed woodland through the annual cycle. During the dry season, soil water potential was strongly and positively correlated with grass LAI, and less strongly with tree and shrub LAI. By the end of the dry season, LAI of grasses, groundlayer dicots and trees declined to 28, 60, and 68% of mean wet-season values, respectively. We compared the data to remotely sensed vegetation indices, finding that field measurements were more strongly correlated to the enhanced vegetation index (EVI, r ²=0.71) than to the normalized difference vegetation index (NDVI, r ²=0.49). Although the latter has been more widely used in quantifying leaf dynamics of tropical savannas, EVI appears better suited for this purpose. Our ground-based measurements demonstrate that groundlayer LAI declines with increasing tree density across sites, with savanna grasses being excluded at a tree LAI of approximately 3.3. LAI averaged 4.2 in nearby gallery (riparian) forest, so savanna grasses were absent, thereby greatly reducing fire risk and permitting survival of fire-sensitive forest tree species. Although edaphic conditions may partly explain the larger tree LAI of forests, relative to savanna, biological differences between savanna and forest tree species play an important role. Overall, forest tree species had 48% greater LAI than congeneric savanna trees under similar growing conditions. Savanna and forest species play distinct roles in the structure and dynamics of savanna–forest boundaries, contributing to the differences in fire regimes, microclimate, and nutrient cycling between savanna and forest ecosystems.     

Elephant (Loxodonta africana) Disturbance to Riparian Woodland: Effects on Tree-Species Richness,Diversity and Functional Redundancy

The substantial increase in elephant populations across many areas in southern Africa over past decades is prompting concerns about the effects on biodiversity. We investigated the outcomes of elephant disturbance on tree-species presence, density, and richness, and on alpha and beta diversity within riparian woodland in Chobe National Park, Botswana. We enumerated all tree species occurring in 32 plots (0.06 ha) along the Chobe riverfront. Plots were stratified by soil type (nutrient-rich alluvium vs. nutrient-poor Kalahari sand covering alluvium) and elephant impact (high vs. low impact on both soil types). We tested four predictions: elephants reduce tree density, richness, and alpha diversity; beta diversity is greater in vegetation subjected to high elephant impact; elephant impact on tree-species composition is greater on nutrient-poor than on nutrient-rich soil; and the loss or decline of abundant tree species on heavily disturbed sites is offset by an increase in abundance of functionally similar species, ones that are minor on lightly disturbed sites. Elephant browsing substantially affected tree-species composition, reducing density, species richness, evenness, and alpha diversity but had no effect on beta diversity. The dominant species on relatively undisturbed areas were partly replaced by functionally similar species on heavily disturbed sites. Soil type influenced species composition on lightly disturbed sites but was less important at higher elephant densities. Our findings are important for areas with extreme dry-season densities of elephants but should not be extrapolated to infer purported effects of elephants on tree diversity at lower densities.     

Light environments and floristic composition across an open forest-rainforest boundary in northeastern Queensland

Abstract This paper demonstrates the changes in structure, floristics and forest floor light regimes across the boundary between open forest and rainforest at Kirrama in northeastern Queensland. Hemispherical photographs of the canopy were used to estimate spatial and temporal variations in potential (clear-sky) direct and diffuse photosynthetic photon flux density (PPFD = 400–700 nm) across the boundary. Five vegetation zones were identified across the boundary: (A) open forest, (B) tall open forest, (C) tall open forest with a rainforest under-storey, (D) young rainforest, and (E) mature rainforest. During the summer (September–March) direct PPFD declined semi-exponentially across the boundary, while in the winter (April–August), the decline was more linear. However, the decline in diffuse PPFD across the boundary was linear throughout the year. Mean per cent grass cover was positively correlated and mean per cent shrub cover negatively correlated with annual average total PPFD across the boundary. Mean percentages of diffuse radiation relative to that above the forest (diffuse site factors) ranged from 9.8% in the mature rainforest to 66.4% in the open forest, while mean percentages of direct radiation relative to that above the forest (direct site factors) ranged from 2.9–38.3% at the same sites, respectively. Daily PPFD in the rainforest understorey is probably insufficient for the shade intolerant (pioneer) species which prefer the more open environments on the rainforest margin where light intensities and soil temperatures are higher. Towards the rainforest end of zone C, light conditions appear to be similar to those experienced within small treefall gaps in rainforests. Typically, such light conditions are preferred by shade tolerant (primary) tree species that are unable to grow and reproduce successfully in the adjacent shaded understorey and to compete with the fast growing pioneer tree species in the more open end of zone C.     

Structure, floristic composition, and vegetation forming factors of three vegetation types in Senegal

Six 1 ha plots were established in a coastal savanna, called Fathala Forest, in Delta du Saloum National Park, Senegal. Two plots were placed in woodland, two in wooded grassland, and two in transition woodland in order to describe structure and floristic composition of the vegetation. All trees ≥ 5 cm dbh were sampled. The three selected vegetation types showed distinct differences in structure as well as in species composition. Woodland had high density (440–449 individuals per ha), many small trees, and high basal area (13.4 m²per ha). Transition woodland was characterised by low density (54–118 individuals per ha) but many large trees and a relatively large basal area (8.6–12.8 m² per ha). Wooded grassland was characterised by medium sized trees, it had low density (86–102 individuals per ha) and low basal area (3.8–5.7 m² per ha). Species richness ranged between 17 and 27 species per ha in the six plots. Only two species were found in all plots, Daniellia oliveri (Caesalpiniaceae) and Prosopis africana (Mimosaceae). Legumes dominated all plots. Wooded grassland and transition woodland had many characteristics of fire-affected vegetation in contrast to woodland. Today wooded grassland encroaches on woodland and transition woodland. Management of the latter two vegetation types should be given priority as they maintain structural and floristic characteristics that are essential to conserve biodiversity and original features of the vegetation, and they are also important for local people who are allowed to make sustainable use of the vegetation.     

Soil seed bank dynamics of fire-prone wooded grassland, woodland and dry forest ecosystems in Ethiopia

In order to reveal the role of soil seed banks in vegetation recovery after fire in savanna, the spatial distribution and temporal changes in the soil seed banks of regularly burning savanna in Gambella, western Ethiopia, was studied. The seedling emergence technique was employed to determine the species composition and density of the soil seed bank of six sites ranging in fire severity from wooded grassland with frequent fires over woodland with intermediate fire frequency to forest with absence of fires. Species composition and density of seeds in the soil were compared between seasons, depths and sites with different types of standing vegetation. Fourteen plant species were recorded in the soil seed bank from the grassland and woodland sites and 6 from the dry forests; 60 % of the taxa in the soil seed bank were annuals and 40 % were perennials. The soil seed banks were largely dominated by graminoids and 48–97 % of the soil seed bank in the grasslands and woodlands was of a single grass species, Hyparrhenia confinis , which was absent from the dry forests. The soil seed pools ranged from less than 100 to 4700 seeds per m² depending upon the season. The soil seed bank of graminoids was nearly empty after the onset of the rainy season whereas seeds of broadleaved herbs and woody species able to germinate were still found after this time. Floristic composition, representation of life forms and density of seeds in soil did not correspond closely with that of the standing vegetation, but within graminoids there was a strong similarity between the soil seed bank and the standing vegetation. The current fire regime of Ethiopian savanna woodlands appears to maintain the dominance of graminoids over broadleaved herbs and woody plants both as seeds in the soil and in the standing vegetation.     

Soil chemical factors and grassland species density in Emas National Park (central Brazil)

Studies of grasslands on specific soil types suggest that different nutrients can limit biomass production and, hence, species composition and number. The Brazilian cerrado is the major savanna region in America and once covered about 2 million km(2), mainly in the Brazilian Central Plateau, under seasonal climate, with wet summer and dry winter. In view of the importance of soil chemical factors in the distribution of the vegetation forms within the Cerrado domain and which may influence the number of species, we analyzed some soil characteristics in three herbaceous vegetation forms -- hyperseasonal cerrado, seasonal cerrado, and wet grassland -- in Emas National Park, a core cerrado site, to investigate the relationship between number of species and soil characteristics. We collected vegetation and soil samples in these three vegetation forms and submitted the obtained data to multiple linear regression. We found out that aluminum and pH were the best predictors of species density, the former positively related to species density and the latter negatively related. Since the predictable variation in species density is important in determining areas of conservation, we can postulate that these two soil factors are indicators of high species density areas in tropical grasslands, which could be used in selecting priority sites for conservation.     

A comparison of foliar nutrient concentration in trees from monsoon rainforest and savanna in northern Australia

Abstract Analysis of foliar nutrient concentrations revealed that two mesophyllic monsoon rainforest trees had higher concentrations of most nutrients in leaves than eight savanna species. One of the tested monsoon rainforest species with sclerophyllous leaves had similar nutrient concentrations to the savanna tree species. There were positive or no significant correlations between live foliar nutrient concentration and the percentage of nutrients withdrawn prior to litterfall. The nutrient concentration of litter was similar for most nutrients among tree species of monsoon rainforest or savanna. The results of this study suggest that the relative fertility of surface soils of monsoon rainforest compared with savanna is not determined by contrasting nutrient cycling strategies whereby monsoon rainforest tree species enrich their soils with relatively nutrient rich leaf litter relative to savanna tree species.     

Water,land, fire,and forest: Multi‐scale determinants of rainforests in the Australian monsoon tropics

The small rainforest fragments found in savanna landscapes are powerful, yet often overlooked, model systems to understand the controls of these contrasting ecosystems. We analyzed the relative effect of climatic variables on rainforest density at a subcontinental level, and employed high‐resolution, regional‐level analyses to assess the importance of landscape settings and fire activity in determining rainforest density in a frequently burnt Australian savanna landscape. Estimates of rainforest density (ha/km²) across the Northern Territory and Western Australia, derived from preexisting maps, were used to calculate the correlations between rainforest density and climatic variables. A detailed map of the northern Kimberley (Western Australia) rainforests was generated and analyzed to determine the importance of geology and topography in controlling rainforests, and to contrast rainforest density on frequently burnt mainland and nearby islands. In the northwestern Australian, tropics rainforest density was positively correlated with rainfall and moisture index, and negatively correlated with potential evapotranspiration. At a regional scale, rainforests showed preference for complex topographic positions and more fertile geology. Compared with mainland areas, islands had significantly lower fire activity, with no differences between terrain types. They also displayed substantially higher rainforest density, even on level terrain where geomorphological processes do not concentrate nutrients or water. Our multi‐scale approach corroborates previous studies that suggest moist climate, infrequent fires, and geology are important stabilizing factors that allow rainforest fragments to persist in savanna landscapes. These factors need to be incorporated in models to predict the future extent of savannas and rainforests under climate change.