Context dependency in interference competition among birds in an endangered woodland ecosystem

Much research has quantified species responses to human‐modified ecosystems. However, there is limited work on how human‐modified ecosystems may reshape competitive interactions between species. Using a 19‐year study across 3 million ha, we aimed to answer the question: Are levels of interference competition between bird species context dependent and influenced by habitat structure and productivity? We focussed on the hyper‐aggressive behaviour of the Noisy Miner (Manorina melanocephala), which is recognized as a key threatening process for other woodland bird species in Australia. Whether environmental conditions such as amount of forest cover and net primary productivity (NPP) mediate the Noisy Miners' impact remains untested at large spatiotemporal scales.

In ecosystems heavily modified by human actions, altered environmental conditions, such as native vegetation loss and fragmentation, can directly shape patterns of species occurrence (Betts et al., 2019;Haddad et al., 2015). There are "winners and losers" following ecosystem modification (e.g., Colléony & Shwartz, 2020), with some species responding positively, and others negatively, and responses mediated by species traits and tolerances Keddy & Laughlin, 2021). While much research has quantified diverse species responses to altered environmental conditions in human-modified ecosystems, relatively limited attention has focused on how altered environmental conditions may reshape competitive interspecific interactions (Scheele et al., 2017).
A key form of competitive interaction is interference competition (Grether & Okamoto, 2022), which has been broadly defined as the aggressive exclusion of one or more species by a dominant competitor (e.g., Beggs et al., 2020;Brian, 1956;Case & Gilpin, 1974;Tannerfeldt et al., 2002;Van Lanen et al., 2011). Most studies of interference competition explore interactions between ecologically and functionally similar sympatric species (e.g., Dhondt, 2011;Haynes et al., 2014). Over the past two decades, several studies in Australia's endangered temperate woodland ecosystems have identified the negative impacts of the hyper-aggressive Noisy Miner (Manorina melanocephala), a native honeyeater weighing ~60 g, on an array of other species (Crates et al., 2020;Mac Nally et al., 2012;Maron et al., 2013) (Table 1).
Hyper-aggressive behaviour by the Noisy Miner (see Tables S1,   S3, S4) is considered to be a form of interference competition (e.g., Beggs et al., 2020) and the mechanisms of interference competition by the Noisy Miner have been relatively well studied at fine spatial scales (e.g., Beggs et al., 2020;Dow, 1977). Repeated harassment by the Noisy Miner can curtail foraging by small-bodied woodland birds (Beggs et al., 2020) (see Figures S1-S8, Tables S3, S5)  To address this knowledge gap about how interference competition is mediated by environmental conditions at the landscape level, we asked the overarching question: Are levels of interference competition between the Noisy Miner and other bird species context dependent and influenced by habitat structure and productivity? To answer this question, we used extensive data on bird species gathered over 19 years in patches of temperate woodland in south-eastern Australia spread across an area of about 3 million ha to construct statistical occupancy models of native bird occurrence and related this to Noisy Miner presence, tree cover, NPP, and time. The temperate woodlands of south-eastern Australia are among the most degraded and extensively cleared biomes globally (Fischer et al., 2009) and the Noisy Miner is a disturbance-adapted species and can be abundant in these highly fragmented and degraded environments (Crates et al., 2020;Mac Nally et al., 2012) (see Table S2). More specifically, we sought answers to four more focussed questions related to site occupancy by woodland birds.  (Beggs et al., 2020;Mac Nally et al., 2012;Maron et al., 2013; Table S6). We expected bird species known to be actively harassed by the Noisy Miner (see Table 1) to show negative associations with Noisy Miner occurrence.
2. Does the amount of tree cover influence site occupancy by woodland birds? Positive relationships between the amount of habitat and the occurrence of species are a long-understood general pattern in ecology (e.g., Rosenzweig, 1995;Watling et al., 2020). Indeed, the amount of nearby tree cover as a proxy for habitat availability has been demonstrated to influence site occupancy by birds in temperate woodlands (Cunningham et al., 2014), including species of conservation concern (Montague-Drake et al., 2009). We predicted that positive tree cover effects for most woodland bird

K E Y W O R D S
Box-Gum Grassy Woodlands, eastern Australia, joint species distribution models, long-term datasets, modified agricultural landscapes, net primary productivity, Noisy Miner, temperate woodlands, vegetation cover species would be observed here, although the opposite trend would characterize open-country taxa not dependent on treed environments.

Does productivity influence site occupancy by woodland birds? NPP
can have strong effects on the availability of food resources such as nectar, pollen, and insect prey. These factors may, in turn, affect habitat suitability for many species (Morrison et al., 2006).
We postulated that woodland bird site occupancy would be highest in the most productive sites and where there was greatest variability in NPP over time.
4. Is interference competition mediated by context (habitat availability and/or productivity)? Some authors have speculated that large Australian honeyeaters (including the Noisy Miner) may engage in interference competition to defend resources in typically resource-limited environments (Beggs et al., 2020;Low, 2014). If resource availability is a driver of interference competition and TA B L E 1 Twenty one species of birds recorded as victims of harassment by the Noisy Miner in the study by Beggs et al. (2020). Species are listed in descending order of frequency of harassment by the Noisy Miner. Note that eight species that were analysed in this investigation were not observed in the fine-scale behavioural study by Beggs et al. (2020). An asterisk indicates birds smaller in size than the Noisy Miner. Birds of conservation concern (sensu Montague-Drake et al., 2009) are marked with the symbol #. We acknowledge that the data summarized in Table 1 are biased as species which have vacated an area because of the actions of the Noisy Miner would not be present for harassment events to be recorded.
allows aggressive exclusion of competitors (e.g., Brian, 1956;Case & Gilpin, 1974), then the negative impacts of interference competition may dissipate in higher productivity areas. Evidence for such a pattern would manifest as a two-way interaction between productivity and Noisy Miner presence in statistical models of bird site occupancy (see Figure 1). For example, the strength of negative associations between the presence of the Noisy Miner and site occupancy by other species of birds on low-productivity sites may be reduced on high-productivity sites. A complicating factor here, however, is that Noisy Miner colonization is also associated with productive sites (Maron et al., 2013;Montague-Drake et al., 2011). Past studies have suggested that the Noisy Miner can be edge associated and is less likely to occur in areas with high amounts of tree cover (Barati et al., 2016;Catterall et al., 2002).
On this basis, a two-way interaction between tree cover and Noisy Miner presence would correspond to greater effects of interference competition on those species negatively associated with the Noisy Miner in areas with low tree cover, but reduced effects where tree cover is high (see Figure 1).

| Study area
Our investigation covered sites within an area of approximately 3 million ha of the wheat-sheep belt of south-eastern Australia ( Figure 2). Much of this area was formerly dominated by temperate woodland (Hobbs & Yates, 2000), but the southern half of the region has been cleared of an estimated 85% of its original tree cover (Keith, 2004). The temperate woodlands of the wheat-sheep belt are among the most heavily modified agricultural regions worldwide (Fischer et al., 2009). They have suffered from severe land degradation which includes secondary salinity, soil erosion, weed invasion, and extensive biodiversity loss (Lindenmayer et al., 2022).
We used data gathered at 259 long-term monitoring sites that support old growth or regrowth Box-Gum Grassy Woodlands.
These sites are part of three distinct ongoing long-term studies; (1) the Nanangroe Natural Experiment which commenced in 1998 and encompasses 54 sites , (2)

| Bird counts
We completed spring surveys on our 259 sites between 2002 and 2020 (see Table A1). Our survey protocols entailed repeated fiveminute point interval counts (sensu Pyke & Recher, 1983) at the 0 m, 100 m, and 200 m points along the permanent transect. In each year of our surveys, each site was surveyed at least twice by two different experienced ornithologists on different days to account for observer heterogeneity (see Cunningham et al., 1999). We deemed a species to be present at a site if it was recorded at any one of the points along a transect on a given day. We then used the count by the second observer to facilitate detection/occupancy modelling. We completed counts between 5.30 and 9.30 am and did not undertake surveys on days of poor weather. We detected birds by both sight and call and estimated the distance to each individual bird. We restricted our analyses to birds recorded within 50 m of the observer.
We recorded wind and time of day for each survey for inclusion in subsequent detection/occupancy analyses (see below).
We recorded 185 species of birds from 2271 site × year surveys completed between 2002 and 2020. Of these, we selected 31 individual species with more than 350 detections for statistical analyses (see Table A2). The species modelled ranged from small-(~6 g) to large-bodied taxa (~370 g), and common species to species of conservation concern in temperate woodlands (

| Tree cover and NPP
We obtained data on the amount of native tree cover in the landscape surrounding each of our 259 long-term sites. We used data on the percentage tree cover within 500 m of the centroid of a given  (2002) of our study (termed cross-sectional net primary productivity) and between 2002 and 2020 (termed deviation in NPP). The MOD17A3HGF product derives annual NPP from the sum of all 8-day net photosynthesis products from the given year based on the quality control label for every pixel. We accessed these data by a point sample request using the Application for Extracting and Exploring Analysis Ready Samples (AppEEARS Team, 2020), with each point representing the geographic coordinates at the centre of each of our 259 2 ha sites. In Figure A1, we provide histograms and density plots for average NPP and percentage tree cover, which were our non-time varying covariates, and boxplots by survey year for deviation in NPP.

| Occupancy
We used a Bayesian multi-species occupancy/detection (BMSOD) model to quantify changes in occupancy of the individual bird species as a function of the presence of the Noisy Miner, the amount of tree cover, the average NPP at a site, annual deviation in primary productivity, and time. While we outline that our data were collected in three large-scale studies, they shared broad similarities in bird taxa, environmental conditions, and consistency with which we collected data. For these reasons, we treated the combined data as one larger investigation and elected not to fit a categorical variable for "study" in our analyses, which would have made our already complex models more complicated. structures here and refer the reader to the above-mentioned references for further details.
Let z i,j,t be the true presence (1) or absence of a species (0) i, at site j, and year t. We assumed that: where i,j,t gives the probability of occurrence of species i at site j and year t. 0i is the intercept for species i. 1i to 6i are the species-specific associated regression coefficients representing the linear and quadratic effects for the year the survey was completed (Y jt , Y 2 jt ), and the linear effects of an indicator variable for whether or not Noisy Miners were detected (NM jt ), the NPP annual mean at the site (NPP j ), the difference between the yearly NPP and the annual mean for the site (ΔNPP jt = NPP jt − NPP j ), the tree cover at the site (TC j , note this variable is not time dependent), and a site level random effect which allows for dependence between years (u j ). These regression coefficients are assumed to be drawn from independent normal distributions with a community mean l and variance T l , where l is one of the seven regression coefficients mentioned in the previous formula above.
Let y i,j,t,k be the observed detection (1) or non-detection (0) of species i at site j, year t, and replicate k. The observation process is assumed to be Bernoulli conditional on the true state, z i,j,t as follows: where p i,j,t,k is the probability of detecting species i, at site j, in year t, for replicate k. 0i is the intercept for species i, 1i and bi are the speciesspecific associated regression coefficients representing the linear effects for the time of day the replicate was completed (TD j,t,k ), and an indicator variable for whether it was windy during the replicate. As in the occupancy component of the model, these regression coefficients are assumed to be drawn from independent normal distributions with a community mean l and variance T l , where l is one of the three regression coefficients mentioned in the previous formula above.
We assigned independent normal priors for the occurrence ( 0 to 6 ) and detection ( 0 to 2 ) community-level regression means and independent inverse-Gamma priors for each of diagonal elements of T 0 to T 6 and T 0 to T 6 .
We generated samples from the posterior distribution by running three Markov chains for 33,000 iterations with a warm up/burn in of 3000. We retained every 15th iteration (thinning factor = 15) giving 6000 samples from the posterior. We used the Gelman-Rubin R (Gelman & Rubin, 1992) statistic, examined trace plots and looked at auto-correlations to assess whether or not the chains showed adequate mixing. All chains showed adequate mixing.
We considered seven additional models that allowed for twoway interactions between NM and NPP, NM and ΔNPP, and NM and TC. We considered an additional eight models where ΔNPP was measured on a relative scale rather than an absolute scale, that is, We used k-fold cross-validation information criteria (KFoldIC, k = 10) to choose the most parsimonious model for the occupancy component of the multivariate occupancy detection model (Gelman et al., 2014;Vehtari et al., 2017).
We constructed a single-species occupancy/detection model for the Noisy Miner using a simplified version of the multi-species model described above. We used the same detection variables as previously described, but included a quadratic effect of survey year, tree cover, average annual NPP, and deviation in annual NPP (current year -average annual NPP). We did not perform model selection for the Noisy Miner model. Given the very high detection probability for the Noisy Miner (the species is large, and typically calls loudly and repeatedly, and is very active), we elected to use raw detection values in our BMSOD rather than estimates from a single-species occupancy model.
To assess model fits, we used posterior predictive checks via the function ppcOcc in the spOccupancy package and reported Bayesian p-values assuming that values being more than .05 indicated an adequate fit.

| Species richness
To test the impacts of the Noisy Miner on the whole bird community, we supplemented our occupancy modelling with generalized linear mixed modelling (GLMMs) of species richness of all birds. We also modelled Chao's estimator of species richness (S) which enabled us to account for undetected species (Chao & Chiu, 2016) by calculating estimates for each sample using the 'estimateR' function in the vegan package (Oksanen et al., 2020) in R (R Core Team, 2021). To test these two richness response variables, we constructed Bayesian generalized linear mixed models using the 'brms' package (Bürkner, 2017) in R (R Core Team, 2021), assuming negative binomial error distributions.
We constructed models using the same fixed, interactive, and random effects as those used in the BMSOD described above. We assigned default uninformative priors and generated samples from posterior distributions by running four Markov chains for 2000 iterations with a warm-up/burn in of 1000. We determined the most parsimonious models using KfoldIC scores (k = 10). We used the Gelman-Rubin R (Gelman & Rubin, 1992) statistic, and examined trace plots to assess whether or not the chains showed adequate mixing. We assessed model fit using posterior predictive checks via the pp_check in the bayesplot package (Gabry et al., 2019) (Gabry & Mahr, 2022).

| Probability of site occupancy by the Noisy Miner
After accounting for detection, our Noisy Miner model showed the species was more likely to occur on sites with low levels of tree cover in the surrounding landscape, and on sites with low average NPP.
In addition, there was evidence of a quadratic effect of year, with a decline in site occupancy in the middle years of our study followed by recovery towards the end (Figure 3).

| Main effects
Our BMSOD model revealed a negative association between increasing wind speed and the probability of detection for most bird species. We also found evidence of some associations between the time of day a survey was conducted and the probability of detecting a species ( Figure A2).
Our most parsimonious model according to the k-fold crossvalidation contained evidence of interactions between the amount of tree cover and Noisy Miner detection, and average NPP and Noisy Miner detection (Table A3). Posterior predictive checks indicated that the model was a reasonable fit for all individual species (Table A4). We summarize these interacting effects in more detail below, after first describing main effects in the model.
We found an association between the detection of the Noisy Miner and the probability of occupancy of 27 of the 31 species we modelled. Of these, nine species were positively associated with the Noisy Miner (none of conservation concern), whereas twice as many (18 species) were negatively associated with the Noisy Miner (including six species of conservation concern) (Figure 4).
The vast majority of species negatively associated with the Noisy Miner were small-bodied birds (see Figure S4), whereas some of those positively associated with the Noisy Miner were large-bodied species (Figure 4). However, there were exceptions to this general pattern with large-bodied taxa like the Crimson Rosella (Platycercus
were quadratic effects of year in the temporal patterns in the probability of occupancy for some species (Figure 4).

| Interacting effects of the amount of tree cover and Noisy Miner presence/absence
There was no association between tree cover and overall probability of occupancy, but at the individual species level, associations with tree cover varied from strongly positive to strongly negative ( Figure 4). Six species were characterized by a negative association with the Noisy Miner ( Figure 4). The probability of site occupancy in some species increased with increasing tree cover but the pattern varied between sites where the Noisy Miner was present relative to where it was absent. As an example, the Grey Shrike-thrush (Colluricincla harmonica) was more likely to occur at sites with low levels of tree cover in the absence of the Noisy Miner than when the Noisy Miner was present.
No such differences in site occupancy by the Grey Shrike-thrush were apparent at high levels of tree cover irrespective of whether the Noisy Miner was present or absent ( Figure 4). We documented a similar pattern for the Brown Treecreeper and the Black-faced Cuckoo-Shrike ( Figure 5). In other cases (e.g., the Crested Shrike-tit, Rufous Songlark, and Willie Wagtail Rhipidura leucophrys), the probability of site occupancy declined with increasing tree cover in the absence of the Noisy Miner, but increased in the presence of the Noisy Miner ( Figure 6).
In a further pattern, the Red-rumped Parrot (Psephotus haematonotus) was an example of a species where the probability of site occupancy declined with increasing tree cover but this effect was more marked in the absence of the Noisy Miner than when it was present ( Figure 5).

| Interacting effects of NPP and Noisy Miner presence/absence
We found evidence of a positive association between NPP and the probability of bird site occupancy overall (Figure 4). At the individual species level, seven of our 31 species exhibited a positive association with NPP, whereas six exhibited a negative association (Figure 4).
The interaction between NPP, and the presence/absence of the Noisy Miner was important for seven species (Figures 4 and 7).
Five of these seven species exhibited a negative association with the Noisy Miner (see Figure 4). As an example, the probability of site occupancy in the White-plumed Honeyeater (Ptilotula penicillata) and Black-faced Cuckoo-shrike ranged from close to zero at low levels of productivity in the presence of the Noisy Miner to 25% in the absence of the Noisy Miner. Probability of occupancy for these species increased markedly with increasing site productivity both with and without the Noisy Miner (Figure 7). The opposite pattern was found for the White-winged Triller (Lalage tricolour) for which probability of occupancy was highest on lowproductivity sites in the absence of the Noisy Miner and declined with increasing productivity, approaching similarly low levels of site occupancy to those on high-productivity sites in the presence of the Noisy Miner (Figure 7).

F I G U R E 4
The fixed effects estimates of Noisy Miner detection, Year (and its quadratic), Tree Cover, average annual NPP, and relative NPP on the probability of bird species occupancy. Interactive effects are shown in Figure 5. Estimates are on the logit scale and species have been ordered by body size with Australian Raven being the largest species and Weebill being the smallest. Error bars represent 95% credible intervals. Effects were interpreted as being significant if their credible intervals did not cross the zero-effect line (du Prel et al., 2009). Large points denote significant estimates whereas small points denote non-significant estimates.

| Species richness
Our most parsimonious models for species richness and Chao's S

| DISCUSS ION
Competition between species can be an important driver of community structure (Dhondt, 2011;Lack, 1944;Van Lanen et al., 2011), yet is notoriously difficult to study in the wild (Grether & Okamoto, 2022). In heavily modified agricultural landscapes in eastern Australia, research over the past decade has indicated there is strong interference competition between the native honeyeater, the Noisy Miner, and co-occurring native bird species (Beggs et al., 2020;Mac Nally et al., 2012;Maron et al., 2013;Westgate et al., 2021). We found an association between the detection of the Noisy Miner and the probability of occupancy of 27 of the 31 species we modelled. Most of these associations were negative (18 species) including for a number of species of conservation concern. We found that the potential for interference competition exerted by the Noisy Miner on other bird species was tempered by increasing tree cover and/or increasing NPP. We discuss our key findings below and conclude with commentary on the implications of our results for bird and woodland conservation.

| Does the occurrence of the Noisy Miner influence site occupancy by other species of woodland birds?
We found that overall species richness and more than half (18) of the 31 species modelled exhibited a negative association with the Noisy Miner. An interesting result from our analyses was that negative associations were not limited to small-bodied species, but spanned a range of body sizes, including several taxa larger than the Noisy Miner (see Table A2 for body sizes). In addition, of the suite of small and large-bodied species known to be victims of harassment by the F I G U R E 5 The estimates of the interactive effects of Noisy Miner detection and Tree Cover, and Noisy Miner detection and average annual NPP on the probability of bird species occupancy. Fixed effects are shown in Figure 4. Estimates are on the logit scale and species have been ordered by body size with Australian Raven the largest species and Weebill the smallest. Error bars represent 95% credible intervals. Effects were interpreted as being significant if their credible intervals did not cross the zero-effect line (du Prel et al., 2009). Large points denote significant estimates whereas small points denote non-significant estimates.

F I G U R E 6
Relationships between bird site occupancy, the presence/absence of the Noisy Miner, and the amount of tree cover in the 500 m surrounding a long-term site. Posterior means are given by lines and the shaded areas are 95% credible intervals. Panels with a red star denote that the interactive effects of the species are significant ( Figure 5). Bird images are used with permission from the Canberra Ornithologists Group or the Macaulay Library (Table A7). Images of birds are not to scale.

F I G U R E 7
Relationships between bird site occupancy, the presence/absence of the Noisy Miner, and average annual net primary productivity. Posterior means are given by lines and the shaded areas are 95% credible intervals. Panels with a red star denote that the interactive effects of the species are significant ( Figure 5). Bird images are used with permission from the Canberra Ornithologists Group or the Macaulay Library (Table A7). Images of birds are not to scale.
Noisy Miner (Beggs et al., 2020), we found that some were negatively associated with the Noisy Miner and others positively associated with the species.
An interesting outcome of our analysis was positive cooccurrence between the Noisy Miner and some of the larger-bodied bird species. These larger species are of broad ecological interest, but unlike smaller-bodied taxa, none are of conservation concern.
They are typically granivores or omnivores and are generalists in agricultural landscapes that can use both woodland patches and the surrounding heavily cleared grazing pastures and cropped areas.
While many of these larger species are also targeted by Noisy Miner aggression, they appear able to tolerate such behaviour (and often reciprocate). In other cases, species associated with the Noisy Miner,

| Are interference competition effects on woodland birds mediated by tree cover and NPP?
Our analyses revealed that bird site occupancy was influenced by both the amount of tree cover and NPP with such effects varying between positive and negative among different species. Site occupancy in seven species was influenced by an interaction between the amount of tree cover in the landscape and whether the Noisy Miner was present or absent. The effect of the Noisy Miner on patterns of bird site occupancy was often markedly different at low levels of tree cover relative to high levels of tree cover. These patterns were sometimes negative and sometimes positive, depending on the bird species being examined. This indicates that interference competition by the Noisy Miner may vary in strength both with levels of tree cover and among species. There was a broadly similar outcome for the two-way interaction between the presence/absence of the Noisy Miner and NPP. That is, there was interspecific variation in response to the presence/absence of the Noisy Miner depending on productivity. Notably, we also found an increase in species richness and Chaos' S with increasing tree cover and increasing annual average NPP (ΔNPP).
We propose that two mechanisms might at least partially explain why interference competition is mediated by tree cover or NPP for more than a third of the species examined. For responses to the amount of tree cover, it is possible that more extensive areas of woodland provide more opportunities for some woodland birds to escape harassment by the Noisy Miner (see Table S7). Our anecdotal observations suggest that the Noisy Miner is most likely to attack other birds when they have a direct line of sight to them.
This might be obscured when there is more tree cover in a landscape. This hypothesized explanation may be broadly consistent with previous work at a site-level which has shown that the Noisy Miner often avoids areas with a dense understory in replanted areas (Lindenmayer et al., 2010), as well as in regrowth and old growth woodland with a dense understorey (Westgate et al., 2021). This explanation also appears to be congruent with the findings of other studies which suggest the Noisy Miner is an edge-associated species (Barati et al., 2016;Catterall et al., 2002), with greater areas of interior woodland habitat free of Noisy Miners likely to occur where there is greater overall tree cover. Increasing levels of tree cover also may contribute to greater levels of environmental heterogeneity in the landscape (Lindenmayer & Fischer, 2006), and this in turn may provide more opportunities for more species to coexist (e.g., see Stein et al., 2014). Notably, other studies have revealed that vegetation cover or other environmental conditions can influence species interactions (Miller & Mullette, 1985;Rieman et al., 2006;Tannerfeldt et al., 2002). For instance, fish predation on tadpoles is greater when aquatic vegetation complexity is lower (Babbitt & Tarr, 2002). In another study, shoreline geography influenced site occupancy, co-occurrence, and competition between the Yellowbilled Loon (Gavia adamsii) and Pacific Loon (Gavia pacifica) (Haynes et al., 2014). Tavella and Cagnolo (2019)  interactions. They demonstrated that the more aggressive species became increasingly dominant and competitive on sites that had been burnt relative to unburnt areas (Tavella & Cagnolo, 2019).
In cases where NPP appears to mediate species responses to Noisy Miner occurrence, a second mechanism explaining our findings may be associated with resource availability (see Tables S8-S16).
Aggressive behaviour in many Australian honeyeaters, including the Noisy Miner, has been attributed to a need to protect resource patches in an otherwise resource-limited landscape (Low, 2014). Beggs et al. (2020) found that patch-scale removal of Noisy Miners was associated with a doubling of the rate of foraging by several species of small woodland birds. It is possible that the need to engage in harassment might be most pronounced where productivity is lowest, and resources need to be most aggressively defended against real or perceived competitors. In contrast, if resource availability is a driver of interference competition and allows aggressive exclusion of competitors (e.g., Brian, 1956;Case & Gilpin, 1974), then the negative impacts of interference competition may dissipate in higher productivity areas, as observed for some species in this study.

| Caveats
For the purposes of this study, we considered a negative relationship in co-occurrence between the Noisy Miner and other species of woodland birds to be evidence of interference competition.
However, a limitation of our investigation is that it is based on correlative co-occurrence patterns, similar to several other studies of competition (e.g., Haynes et al., 2014). This means that some of the negative co-occurrence patterns we observed may have occurred for reasons that are not directly related to interference competition such as broad similarities or differences in habitat requirements. However, it is not possible to do a true experiment on Noisy Miner exclusion over large spatial scales and prolonged time periods in fragmented landscapes with connectivity to other sources of populations of the Noisy Miner. This is because of the rapidity with which areas where birds have been culled are recolonized by the Noisy Miner (see Beggs et al., 2019).
Our occupancy analyses focused on the bird species for which there were sufficient data available for analysis (>350 detections). We are aware that many other, less commonly recorded species, including a number of species of conservation concern and vulnerable to Noisy Miner competition, could not be analysed for species-specific responses. Indeed, we note that the data summarized in Table 1 et al., 2012). However, the Noisy Miner is a highly social species and typically exists in large groups (Dow, 1979). Hence, the presence of the Noisy Miner in woodlands will very often correlate strongly with large colonies of the species.

| Management implications
Our findings suggest that the most-effective conservation outcomes for some woodland birds (including a number of conservation con-

| CON CLUS IONS
Species associations and interactions can play important roles in influencing temporal and spatial co-occurrence, and can ultimately shape the assembly of biotic communities (Chesson, 2018;Dhondt, 2011;Grether & Okamoto, 2022;Haynes et al., 2014;Lack, 1944) (Van Lanen et al., 2011. Variations in environmental conditions can lead to shifts in these associations and hence affect patterns of species presence. Based on analyses of extensive data on woodland birds gathered on sites across more than 3 million ha in the past 19 years, we uncovered evidence that interference competition between the Noisy Miner and other woodland bird species is both context dependent and species dependent. This conclusion is supported by the observation that some bird species exhibited a different association with tree cover or productivity depending on whether the Noisy Miner was present or absent. In addition, differences in the probability of occupancy of some species that were apparent at a given level of tree cover or productivity, and which varied depending on Noisy Miner presence or absence, became more muted on, for example, sites with higher tree cover or productivity, irrespective of whether the Noisy Miner did or did not occur. Possible mechanisms for such patterns include: (1) the reduced physical ability of the Noisy Miner to harass other birds because of structural changes associated with increasing amounts of tree cover, and (2) a reduced need to compete with other species where levels of NPP are higher and resource availability is greater.

ACK N O WLE D G E M ENTS
We thank the Canberra Ornithologists Group and the Macaulay Library for granting us permission to use their bird image libraries for figures. Tabitha Boyer and Luke Gordon assisted admirably with many editorial aspects of manuscript preparation. We thank two reviewers for insightful comments which considerably improved an earlier version of the manuscript.

FU N D I N G I N FO R M ATI O N
This project was funded by Sustainable Farms at ANU, which is made possible by funding from The Ian Potter Foundation and Department of Agriculture, Water and the Environment.

CO N FLI C T O F I NTE R E S T
No authors have any conflict of interest to declare in the writing and publication of this research.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13680.

DATA AVA I L A B I L I T Y S TAT E M E N T
All paper data is available in the appendices to this paper, with the remainder deposited in DRYAD at: https://datad ryad.org/stash/ share/ nPlLl 5nuzT 9nHwL kPbhy ZHs7p d9t6H k6AIU TgNjPkUc [https://doi.