Multiple Long‐Term, Landscape‐Scale Data Sets Reveal Intraspecific Spatial Variation in Temporal Trends for Bird Species

ABSTRACT

Quantifying temporal changes in species occurrence has been a key part of ecology since its inception. We quantified multidecadal site occupancy trajectories for 18 bird species in four independent long‐term, large‐scale studies (571 sites, ~1000 km latitude) in Australia. We found evidence of a year × long‐term study interaction in the best‐fitting models for 14 of the 18 species analysed, with differences in the temporal trajectories of the same species in multiple studies consistent with non‐stationarity. Non‐stationarity patterns in occupancy were not related to the distance from a species niche centroid; species in locations further from their niche centroid did not demonstrate differing temporal trajectories to those closer to their niche centroid. Furthermore, temporal trajectories of species were not associated with climatic values for each study relative to their niche. Our findings demonstrate the need for multiple long‐term studies across a species range, especially when tailoring conservation decisions for populations.

Quantifying temporal changes in species occurrence has been a key part of ecology since its inception. We quantified multidecadal site occupancy trajectories for 18 bird species in four independent long‐term, large‐scale studies (571 sites, ~ 1000 km latitude) in Australia. We provide evidence of non‐stationarity with differences in the temporal trajectories of the same species in multiple studies and a year × long‐term study interaction in the best‐fitting models for most taxa.

1. Introduction

A key part of ecology is quantifying temporal patterns of change in populations of species and the factors influencing them (Elton 1927; Krebs 2009; Krebs et al. 2024). Long‐term studies can be important for doing this (Likens 1989; Lindenmayer et al. 2012; Hughes et al. 2017). However, given the logistical and financial constraints of long‐term studies (Lindenmayer et al. 2012), they are often restricted in terms of their geographic scope. This can be problematic as long‐term study sites are often selected for particular characteristics, such as supporting high numbers of a given target species (Fournier et al. 2019), or to test specific scientific hypothesis. Either of these factors may mean that key trends in measures like the abundance of species and/or species‐level site occupancy from a given individual study may not be representative of patterns across the full range of that taxon (Symstad et al. 2003; Soranno et al. 2014; Fournier et al. 2019).

Given potential limitations of inferences from a single investigation, studies of temporal change in different populations conducted across the distribution of that species can be particularly informative (Krebs 2009; Compagnoni, Evers, and Knight 2023). Indeed, two broad patterns of temporal change have been identified from analyses of long‐term data of populations of the same species. One of these is synchrony—where the same species in different populations exhibit the same demographic parameters and/or the same temporal trajectory in measures such as abundance or site occupancy (Bjørnstad, Ims, and Lambin 1999; Bellamy, Rothery, and Hinsley 2003; Liebhold, Koenig, and Bjornstad 2004). Synchrony in population dynamics can occur due to spatial proximity, similarities in climatic conditions at different sites within the distribution of a given species (e.g., Bellamy, Rothery, and Hinsley 2003), or because of long‐distance dispersal between populations (Ranta et al. 1997; Peltonen et al. 2002; Mortelliti et al. 2015). Synchrony in population dynamics may occur across very large spatial scales (e.g., hundreds to thousands of kilometres) (Koenig 2002; Liebhold, Koenig, and Bjornstad 2004; Vallés‐Medialdea et al. 2022). There are many examples of synchrony including those (among others) of mammals in north‐western North America (Moran 1953; Krebs 2009), birds in North America (Koenig 2001) and the United Kingdom (Bellamy, Rothery, and Hinsley 2003), and waterfowl in Spain (Vallés‐Medialdea et al. 2022).

An alternative spatio‐temporal pattern to synchrony is non‐stationarity—where there is spatial variation in the demographic parameters or temporal trajectories of different populations of the same species (Rollinson et al. 2021; Pease, Pacifici, and Kays 2022; Murphy and Jarznya 2023). Non‐stationarity can be broadly defined as spatial or temporal variation in the direction and/or the magnitude of a relationship between a response variable (such as animal occurrence or site occupancy) and explanatory variables (e.g., climate attributes, landscape cover or other factors) (Schmidt et al. 2014; Rollinson et al. 2021; Pease, Pacifici, and Kays 2022). Non‐stationarity may be common (Bini et al. 2009) due to spatial variation in climate, environment and plant and animal responses to these factors (Rollinson et al. 2021) such as local adaptation. For example, Pease, Pacifici, and Kays (2022) found non‐stationarity in populations of White‐tailed Deer (Odocoileus virginianus) in North Carolina (USA) that was associated with an interactive effect of predation pressure and extent of forest cover. Ranius et al. (2024) described marked differences in dispersal rates and ranges between northern and southern populations of the Hermit Beetle (Osmoderma eremita) and attributed these differences to temperature regimes. Murphy and Jarznya (2023) documented evidence of non‐stationarity for the majority (90%) of overwintering North American bird species they modelled and identified the spatial extent of suitable habitat as a key driver of demographic change.

Despite the potential prevalence of intraspecific spatial non‐stationarity (Bini et al. 2009), empirical studies of it appear to be comparatively uncommon (but see Schmidt et al. 2014; Pease, Pacifici, and Kays 2022; Murphy and Jarznya 2023). This is likely due to a lack of multiple long‐term data sets to quantify such patterns for a given species across its range (Pease, Pacifici, and Kays 2022), although census data collected over extensive geographic regions such as those associated with initiatives like the Audubon Christmas Bird Count in North America make more studies of non‐stationarity possible (Murphy and Jarznya 2023). The current paucity of studies on non‐stationarity is a major knowledge gap because efforts to stem species losses are dependent on a strong understanding of which populations of a given species are declining in what places. This can, in turn, help tailor conservation decisions for particular populations of a given species in space and over time (Rollinson et al. 2021; Murphy and Jarznya 2023).

In the empirical study reported here, we explored evidence for synchrony and non‐stationarity in bird species site‐occupancy trajectories across 571 permanent field sites within four independent long‐term (20–25 years), large‐scale studies located in four major ecosystem types spanning ~1000 km of latitude in eastern Australia. We established these long‐term studies to quantify individual bird species responses to key drivers of change in each long‐term study such as natural disturbances (e.g., wildfire), and management treatments or interventions (e.g., logging, vegetation restoration). When combined, these studies offer an important opportunity to test for intraspecific synchrony and non‐stationarity, as well as to examine some of the potential drivers of such patterns. A novel aspect of our study was the exploration of patterns of synchrony versus non‐stationarity using occupancy modelling, accounting for imperfect detection. We examined whether the spatio‐temporal patterns in bird site occupancy were associated with the location of long‐term study sites within the bioclimatic niche space of each bird species. Drawing broadly on the abundant‐centre hypothesis (sensu Martínez‐Meyer et al. 2013), locations close to the centre of a species' niche might be expected to support optimal conditions for that taxon. Therefore, that species would be more abundant towards its niche centroid (see Osorio‐Olvera et al. 2020), although this is generally a weakly supported pattern observed in some studies (e.g., Dallas, Decker, and Hastings 2017; but see the re‐analysis by Soberón et al. 2018). Hence, populations of species at sites close to the niche centroid would be less likely to exhibit a declining trajectory in site occupancy than locations distant to the niche centroid.

The locations of the different studies in our investigation (and their niche position relative to the niche centroid for a given species) may or may not support optimal temperature and/or rainfall conditions for that species. That is, a species niche centroid may not occur at the centre of its geographic range (Pironon et al. 2017). Given this, together with the fact that climate attributes have been used to explain both synchrony (Bellamy, Rothery, and Hinsley 2003; Liebhold, Koenig, and Bjornstad 2004) and also non‐stationarity (see Murphy and Jarznya 2023), we sought to determine if the spatio‐temporal patterns in bird site occupancy that we observed were associated with the equivalent measures of climate at species' niche centroids in a third part of our analysis.

The three key, inter‐related questions which motivated our study are outlined below.

• Q1. Is there evidence of synchrony or non‐stationarity in site occupancy trajectories for a suite of bird species common to our long‐term studies?

Due to the large spatial coverage of our long‐term studies (and hence between‐study heterogeneity in climatic and environmental conditions), we predicted that the temporal trajectories in site occupancy for the majority of individual bird species would be characterised by non‐stationarity. Conversely, synchrony might be apparent in wide‐ranging generalist species with similar life‐history attributes (such as seasonal migrants with a broad diet encompassing invertebrates, nectar and pollen).

• Q2. Is intraspecific spatial variation in long‐term occupancy patterns (viz non‐stationarity) related to the distance of study sites from each species' niche centroid?

At the outset of this investigation, we hypothesised that, in niche space, a population of a species on our sites far from its niche centroid would be more likely to exhibit a declining trajectory in occupancy. That is, there would be a negative relationship between a species' occupancy trajectory and the distance of a given long‐term study to that species niche centroid.

• Q3. Are long‐term occupancy patterns driven by species tracking their climate niches?

If species are tracking their climate niches as documented in other studies (e.g., see Antão et al. 2022; Lawlor et al. 2024), then it is expected there would be differences in temporal trajectories of occupancy between hotter versus cooler edges, and the drier versus wetter edges of that particular species' distribution. On this basis, and under a warming climate (Canadell et al. 2021), site occupancy in a given species might be, for example, less likely to be declining in locations that are cooler or wetter relative to mean climatic conditions characteristic of their niche centroid. This could, in turn, drive spatio‐temporal patterns in bird site occupancy consistent with non‐stationarity.

2. Methods

We explored evidence for non‐stationarity in trajectories of species site occupancy across different ecosystem types by analysing four large‐scale, long‐term time‐series data sets from eastern Australia: (1) the Nanangroe study, (2) the South West Slopes restoration study, (3) the Booderee National Park study and (4) the Victorian Central Highlands study (Table 1; Figure 1).

TABLE 1.

Summary of data from four long‐term studies in south‐eastern Australia (see Figure 1). Total bird counts are for the 18 species with sufficient data across the long‐term studies (and occurred in at least 5% of surveys) to enable statistical analyses (see text and Table S1).

Long‐term study	Years	Sites	Key experimental contrasts used in study	Site × year surveys	Total surveys	Total birds observed within 50 m of an observer
Booderee National Park	2003–2007, 2009–2022	129	Broad vegetation type: Forest, Heathland, Rainforest, Sedgeland, Shrubland, Woodland	2169	4338	18,927
Nanangroe	1998, 2000, 2005, 2007, 2009, 2011–2017, 2020, 2022	111	Temperate woodland Location in landscape: Farm	1397	2794	13,395
			Inner plantation
			Outer plantation
South‐west slopes restoration	2002, 2004, 2006, 2008, 2009, 2011, 2013, 2015, 2017, 2019, 2021	218	Temperate woodland	1990	3980	5967
			Patch type:
			Coppiced regrowth
			Natural regrowth
			Planting
			Old growth
Victorian Central Highlands	2004, 2005, 2007, 2009–2014, 2016, 2017, 2019, 2022	112	Tall, wet eucalypt forest	1011	2022	11,708
			Stand age:
			1926/1939
			1960/1990 2009
			Mixed ages
			Old growth
			Old growth/mixed combination
	Total	571		6567	13,134	49,997

FIGURE 1.

Location of the four major long‐term studies of birds in south‐eastern Australia and example images of some of their respective treatments.

The Nanangroe study has been ongoing since 1998 on 111 sites. It entails bird counts in temperate woodland patches before and after the surrounding landscape (previously dominated by cleared grazing paddocks) was subject to exotic Radiata Pine (Pinus radiata) plantation establishment. This study also includes a set of matched woodland ‘control’ patches where the surrounding areas have not been subject to plantation conversion (Lindenmayer et al. 2019b). The South West Slopes restoration study started in 2002 and involves quantifying bird responses on 218 sites in four broad kinds of temperate woodland on farmland: replantings, regrowth woodland (coppice and natural) and old growth woodland dominated by box‐gum eucalypt trees exceeding 120 years old (Lindenmayer et al. 2018). The Booderee National Park study started in 2003 and has focussed on bird counts on 129 sites dominated by different kinds of broad vegetation structure: heathland, woodland, rainforest, sedgeland, shrubland, rainforest and dry‐sclerophyll forests. These sites include areas long unburnt by wildfire (> 50 years) (Lindenmayer et al. 2016). The Victorian Central Highlands study started in 2004 and encompasses bird counts on 112 replicated sites in tall, wet ash‐type forests of different stand age (time since last stand‐replacing disturbance). Old growth ‘control’ sites in this study are where there has been no fire for > 120 years (Lindenmayer et al. 2019a).

2.1. Field Surveys of Birds

Between 1998 and 2022, we undertook annual bird surveys following broadly the same protocol in each of our four long‐term studies in the Austral spring during the main breeding season and when migratory bird species have arrived. The field protocols involved repeated 5‐min point interval counts (sensu Pyke and Recher 1983) along a permanent 100–200 m transect at each site, with three survey points on a transect in each study except Booderee National Park where there were two per transect.

In each year, we surveyed sites with two different experienced ornithologists on different days to enable analyses to account for imperfect detection. We conducted surveys between 5:30 AM and 10:30 AM, but not on days of poor weather. We estimated the distance from the observer for each bird count; only those birds within 50 m of an observer were recorded. We documented wind, temperature and cloud cover during each survey, allowing us to account for detectability in our statistical analysis (see below).

3. Statistical Analysis

Of the 223 species of birds recorded in our four long‐term studies, we detected 18 species with sufficient frequency (at least 5% of all detections, for sites and years) and across at least three of the four studies, to build statistical models to quantify their trajectories. These species were a mixture of residents and migrants, large and small birds and insectivores, granivores and nectarivores (life history and other data such as whether species were of conservation concern are summarised in Table S1).

• Q1. Is there evidence of synchrony or non‐stationarity in site occupancy trajectories for a suite of bird species common to our long‐term studies in south‐eastern Australia?

To quantify trajectories in site occupancy over time for each species, we constructed a series of Bayesian multi‐season single‐species occupancy/detection models implemented in the ‘spOccupancy’ package (Doser et al. 2022) in R (R Core Team 2023). The spOccupancy package contains functions to fit occupancy/detection models using a Pólya‐Gamma data augmentation (Polson, Scott, and Windle 2013) and Markov Chain Monte Carlo algorithm for computational efficiency. The models we constructed quantified the probability of occupancy for the species in question while also taking into account factors that influence detection (MacKenzie et al. 2018). The multiseason nature of these models assumes that the true system state, represented by occupancy, changes over time and can therefore be used to analyse trends over time (Doser and Kéry 2024). In all our models, we included wind, cloud cover, temperature and the individual observer ID as detection covariates. The models also enabled us to use a first‐order autoregressive temporal covariance structure (AR1) to allow dependence in the temporal random effects, instead of assuming that each time period was independent. To allow for dependence at each site, over multiple years, we also included a site‐level random effect in all models.

We completed our analyses of temporal data in a sequential way and at different spatial scales to enable examination of site occupancy trajectories for each species: (1) when combining all data for an overall trend over time; (2) for trends over time in each study; and (3) for trends over time for each of the most important experimental contrasts within a given study (see Table 1). When combining all data, we used an offset term for the number of survey points per sample (two in Booderee National Park, three in the other studies). For each stage of our analysis, and for each species, we compared the null model that contained no predictors and a model with time as linear and quadratic predictors. We used k‐fold cross‐validation, interpreting the most parsimonious model within two k‐fold information criterion scores (KFoldIC) of the lowest score as the best‐fit candidate model (Gelman, Hwang, and Vehtari 2014; Vehtari, Gelman, and Gabry 2017).

For the data combined between studies, we also tested for the effects of study as a predictor and the interactions of study and the linear and quadratic effects of time. For the within‐study treatments, we fitted additional models testing the effects of the treatments (experimental contrasts) as a predictor, and the interactions of treatments and the linear and quadratic effects of time. For each model, we ran three Markov chains, each of 200 batches of 50 in length, resulting in 10,000 samples and a warm‐up/burn‐in of 3000. We retained every fourth iteration (thinning factor = four). This resulted in 5250 posterior samples. We examined trace plots and used the Gelman‐Rubin $\widehat{R}$ (Gelman and Rubin 1992) statistic to confirm whether or not the chains showed adequate mixing.

We interpreted an interactive effect of study and time as an indication of non‐stationarity because this would indicate that there was strong evidence for differing species trajectories between studies. Conversely, we interpreted the absence of this interaction to indicate evidence of synchrony in species trajectories.

• Q2. Is intraspecific spatial variation in long‐term occupancy patterns related to the distance of study sites from each species' niche centroid?

We sought to determine if there was a relationship between the species occupancy trajectory in each study location and that location's distance in niche space from the niche centroid of that species. We extracted trajectory effect sizes (i.e., the posterior estimates and standard deviations for the change in occupancy over time) of models where we fitted the linear effect of year as a predictor without its quadratic effect. These effect sizes, we assumed, represented whether each species had increased, decreased or remained stable over time in each of the studies.

Prior to constructing species niche models, we completed the following steps. We downloaded point location occurrence data from the Atlas of Living Australia (ALA) for the 18 bird species analysed (Table S1). We spatially thinned the data to ensure only one record per pixel (pixel size = 30 arcseconds or approx. 1 km²), thereby substantially reducing sampling bias while retaining the greatest amount of useful occurrence information (Aiello‐Lammens et al. 2015). We then cropped the data to polygons of known species' area of extent using species range maps provided by BirdLife International and Handbook of the Birds of the World (DEECA 2023). We extracted 18 bioclimatic variable raster layers for Australia from the WorldClim data set (Fick and Hijmans 2017) using the ‘geodata’ package (Hijmans 2023) in R. We then conducted Pearson correlation analysis on these raster layers using the layerStats function in the raster package (Hijmans 2023) in R and removed layers with a correlation coefficient of 0.5 or above. This left three bioclimatic variables to use in our niche hypervolume analysis (annual mean temperature, mean temperature of driest quarter and annual precipitation).

We constructed niche hypervolumes using the ‘hypervolume’ package (Blonder et al. 2023) specifying the support vector machine method (SVM) for each species using the location‐specific values for the three bioclimatic variables. To account for the error associated with the stochastic nature of hypervolume construction, we generated 500 hypervolumes for each species in each study, extracting the centroid values for each iteration. This enabled us to calculate the mean centroid values and their associated error. We then extracted bioclimatic values for all sites in all four studies and calculated the mean values and standard deviations of the bioclimatic variables for each study. We then calculated the mean distances between the niche centroid bioclimatic values for each species and the site bioclimatic values. To determine the combined uncertainty of these distances, we used the law of propagation of uncertainty. This involved calculating the partial derivatives of the distance function with respect to each bioclimatic value and then applying the individual standard deviations of the bioclimatic values to quantify how these uncertainties contribute to the overall uncertainty in the mean distance estimates. Finally, we constructed a Bayesian linear mixed model using the ‘brms’ package (Bürkner 2021) in R to quantify whether distance from species' niche centroids was associated with species temporal trajectories. We specified measurement errors in both the response and the predictor, using the ‘mi()’ and ‘me()’ functions in the ‘brms’ package. The simple relationship from this approach was:

$$y_{\mathit{se}}~\mathrm{{\rm N}}\left({\tau }_{\mathit{se}},{\sigma }_{{\tau }_{\mathit{se}}}^2\right)$$

$${\tau }_{\mathit{se}}={\beta }_0+{\beta }_1{D}_{\mathit{se}}+u_s+v_e$$

where $y_{\mathit{se}}$ is the observed trajectory over time, ${\tau }_{\mathit{se}}$ is the unobserved mean trajectory and ${\sigma }_{{\tau }_{\mathit{se}}}$ is the standard error of the occupancy trajectory of species s in study e, ${\beta }_0$ is the intercept, ${\beta }_1$ is the regression coefficient representing the linear effect of distance from centroid (D _se ) and $u_s$ and $v_e$ are the species‐ and study‐level intercept random effects which allow dependence of repeated measures for each species s and long‐term study e. We scaled the distance from centroid variable D _s and its associated measurement error to have a mean of zero and standard deviation of one.

• Q3. Are long‐term occupancy patterns driven by species tracking their climate niches?

We sought to determine if species' trajectories were influenced by whether the three bioclimatic variables at each of the four study sites were greater or lower than the centroid value for that variable calculated across each species' entire niche. For example, if the annual precipitation value for the Crimson Rosella at its niche centre (centroid) was 1000 mm, while the mean annual precipitation at Nanangroe was 800 mm, the departure from this species' niche centre at Nanangroe was therefore 200 mm. We quantified the effects of the differences between species' centroid values and study values for the bioclimatic variables by fitting Bayesian linear mixed models testing these as predictors of species trajectories. This enabled us to determine whether trajectories in site occupancy for species across the studies were related to a site's climate conditions, relative to each species' average values for each variable. Using annual precipitation as an example, the model relationship was as follows:

$$y_{\mathit{se}}~\mathrm{{\rm N}}\left({\tau }_{\mathit{se}},{\sigma }_{{\tau }_{\mathit{se}}}^2\right)$$

$${\tau }_{\mathit{se}}={\beta }_0+{\beta }_1{P}_{\mathit{se}}+u_s+v_e$$

where $y_{\mathit{se}}$ is the observed trajectory over time, ${\tau }_{\mathit{se}}$ is the unobserved mean trajectory and ${\sigma }_{{\tau }_s}$ is the standard error of the occupancy trajectory of species s in study e, ${\beta }_0$ is the intercept, ${\beta }_1$ is the regression coefficient representing the linear effect of departure of annual precipitation at a given study e for a species s (P _se ) and $u_s$ and $v_e$ are the species‐ and study‐level intercept random effects which allows dependence of repeated measures for each species s and long‐term study e. We scaled the departure of annual precipitation variable P _se and its associated measurement error to have a mean of zero and standard deviation of one.

We compared combinations of models that included the three bioclimatic variables, including the null model, using Leave‐one‐out cross validation information criterion scores (LOOIC). We considered all models within two scores of the lowest scored model as the best fits (Gelman, Hwang, and Vehtari 2014; Vehtari, Gelman, and Gabry 2017). Again, we included species‐level and study‐level random effects. For all models in Q2, we ran four Markov chains for 10,000 iterations, including 5000 warm‐up/burn‐in iterations. We used the Gelman‐Rubin $\widehat{R}$ (Gelman and Rubin 1992) statistic and examined trace plots to assess whether the chains showed adequate mixing.

4. Results

4.1. Evidence for Synchrony or Non‐stationarity

We uncovered strong evidence of non‐stationarity as reflected by marked differences in the temporal trajectories of the same species in two or more of our long‐term studies (Figure 2); a year × long‐term study interaction characterised the best fitting models for 14 of the 18 species we analysed (Table 2). The four species which exhibited synchrony were the Red Wattlebird (Anthochaera carunculata), Noisy Friarbird (Philemon corniculatus), White‐throated Treecreeper (Cormobates leucophaea) and Superb Fairy‐wren (Malurus cyaneus).

FIGURE 2.

(A) Trajectories for bird site occupancy from 1998 to 2022 using data from all four studies combined. Posterior means are given by lines and the shaded areas are 95% credible intervals. Bird images are used with permission from the Canberra Ornithologists Group or the Macauley Library (see Table S2). Images are not to scale (for model details see Table S3). (B) Trajectories for bird site occupancy from 1998 to 2022 for each of the four studies individually. Trajectories are predicted from the interaction model for each bird species for each study (for model tables see Table S4).

TABLE 2.

Model selection results for Q1. Models where the study × time interaction was the most parsimonious model are bolded. Formulae are abbreviated where T is time and C is study.

Species	Model formula	KFoldIC score	Delta KFoldIC
Brown Thornbill	T × C	11,339.90	0.00
	C	11,351.49	11.59
	T + C	11,351.81	11.91
	Null	11,924.50	584.60
	T	11,931.79	591.89
Crimson Rosella	T × C	12,072.13	0.00
	T + C	12,082.38	10.25
	C	12,083.17	11.03
	T	13,375.28	1303.14
	Null	13,381.89	1309.76
Eastern Spinebill	T × C	9296.87	0.00
	C	9361.80	64.93
	T + C	9371.96	75.09
	T	10,415.58	1118.70
	Null	10,440.28	1143.41
Eastern Yellow Robin	T × C	8010.85	0.00
	C	8028.86	18.01
	T + C	8029.91	19.06
	T	8434.83	423.98
	Null	8439.98	429.13
Golden Whistler	T × C	8026.34	0.00
	T + C	8028.80	2.45
	C	8040.92	14.58
	T	9327.90	1301.56
	Null	9364.72	1338.38
Grey Fantail	T × C	14,052.74	0.00
	C	14,065.57	12.84
	T + C	14,070.77	18.04
	Null	17,091.41	3038.68
	T	17,095.39	3042.65
Grey Shrike thrush	T × C	13,494.80	0.00
	T + C	13,545.02	50.22
	C	13,550.45	55.65
	Null	13,761.73	266.92
	T	13,765.12	270.31
Noisy Friarbird	C	7607.69	0.00
	T × C	7611.24	3.56
	T + C	7616.16	8.47
	T	7959.99	352.30
	Null	7963.48	355.80
Red Wattlebird	T × C	11,355.32	0.00
	T + C	11,356.74	1.42
	C	11,368.42	13.10
	T	11,560.85	205.54
	Null	11,564.65	209.33
Rufous Whistler	T × C	10,740.73	0.00
	C	10,770.27	29.54
	T + C	10,771.19	30.46
	T	10,960.67	219.94
	Null	10,961.17	220.44
Silvereye	T × C	10,390.75	0.00
	T + C	10,417.46	26.71
	C	10,418.10	27.35
	T	10,615.72	224.97
	Null	10,626.37	235.63
Spotted Pardalote	T × C	8698.24	0.00
	T + C	8705.69	7.45
	C	8710.79	12.55
	T	8908.74	210.50
	Null	8920.64	222.40
Striated Pardalote	T × C	10,780.03	0.00
	T + C	10,791.48	11.45
	C	10,794.16	14.12
	Null	10,905.23	125.20
	T	10,907.57	127.54
Striated Thornbill	T × C	8498.96	0.00
	C	8514.20	15.24
	T + C	8516.92	17.96
	T	9009.86	510.90
	Null	9011.53	512.57
Superb Fairy wren	T + C	11,005.61	0.00
	C	11,009.06	3.45
	T × C	11,028.48	22.86
	T	11,532.09	526.47
	Null	11,535.94	530.32
White browed Scrubwren	T × C	10,814.52	0.00
	T + C	10,818.37	3.85
	C	10,821.45	6.93
	T	11,706.09	891.57
	Null	11,731.84	917.32
White throated Treecreeper	T	8990.25	0.00
	T × C	8991.27	1.02
	Null	8995.18	4.93
	T + C	9023.86	33.61
	C	9025.94	35.69
Yellow faced Honeyeater	T × C	11,332.69	0.00
	T + C	11,421.55	88.86
	C	11,430.34	97.65
	T	11,798.28	465.59
	Null	11,806.07	473.38

Beyond differences between studies, we also found evidence for between‐treatment differences in temporal patterns of occurrence for some species within a given study. The proportion of species characterised by a within‐study treatment × year interaction varied from 5% (South West Slopes study) to 50% (the Nanangroe study) (Table S5; Figures S1 and S2). However, within‐study experimental contrast effects were less common relative to between‐study differences in temporal trajectories for all species (i.e., 14 of 18 species or ~78%) (see Tables S4 and S5).

4.2. Relationships Between Temporal Trajectories in Site Occupancy and the Location of Study Sites Relative to a Species' Niche Centroid

We found no evidence that the distance of sites in each long‐term study from the niche centroid of a given species was associated with the temporal trajectory in site occupancy of that species (Figure 3). That is, taxa were not more likely to exhibit an increasing occupancy trajectory in studies located closer to that species' niche centroid and, vice versa, taxa did not show a negative trend in the probability of site occupancy at locations were further from their niche centroid (Figure 3).

FIGURE 3.

Bayesian linear mixed model testing the effect of distance from hypervolume centroid on the long‐term trajectories of bird species in each of the studies (slope = 0.11, standard error = 0.09). Error band represents standard errors. The legend shows the different species for each of the four major studies.

4.3. Relationships Between Temporal Trajectories in Site Occupancy and Climate Variables

Our analyses revealed that the temporal trajectories in site occupancy of species across each study were not associated with the bioclimatic values for each study relative to each species' niche centroid values. That is, none of the models tested in Q3 were better fits, as determined using LOOIC scores, than the null model, which excluded the departure from bioclimatic variables. For example, species site occupancy trajectories did not increase in studies that were located in a relatively cool or wetter portion of a species' niche space as may be expected as the climate warms (Tables S6).

5. Discussion

The distribution of many species encompasses a number of different populations. These different populations may exhibit similar temporal trajectories; that is, synchrony (Bjørnstad, Ims, and Lambin 1999; Liebhold, Koenig, and Bjornstad 2004). In other cases, different populations may exhibit marked differences in temporal trajectories; a form of non‐stationarity (Schmidt et al. 2014; Rollinson et al. 2021; Pease, Pacifici, and Kays 2022). Determining where populations are exhibiting synchrony versus non‐stationarity matters. For example, it can be important for identifying when it is appropriate to scale‐up management strategies from an individual site to larger spatial scales (Rollinson et al. 2021) and/or when to apply particular conservation actions for particular populations (such as strict protection vs. restoration) but not others.

Here, we explored evidence for synchrony versus non‐stationarity for a series of long‐term bird data sets by quantifying site occupancy of the same bird species in different, independent, long‐term and large‐scale studies, spanning ~1000 km of latitude within four major ecosystem types in eastern Australia. We did this for 18 species that varied markedly in life history and other attributes (see Table S1). We also explored the factors potentially influencing non‐stationarity such as the location of study sites in niche space and climate. Our detailed empirical analyses identified 14 bird species that exhibited evidence of non‐stationarity, with four species showing spatio‐temporal patterns in bird site occupancy broadly consistent with synchrony. We found that neither patterns of non‐stationarity nor synchrony were related to location in niche space or climate variables. In the remainder of this paper, we discuss our findings further, including the importance of spatially distributed long‐term monitoring efforts to document temporal changes in species site occupancy over time.

• Q1. Is there evidence of synchrony versus non‐stationarity in site occupancy trajectories for a suite of bird species common to all four long‐term studies in south‐eastern Australia?

We found evidence of a year × long‐term study interaction in the best‐fitting models for 14 of the 18 species we analysed and hence there were marked differences in the temporal trajectory of the same species across different studies—consistent with the concept of non‐stationarity. Only four species exhibited spatio‐temporal patterns in bird site occupancy broadly consistent with synchrony. Thus, the prevalence of non‐stationarity was far greater than synchrony, consistent with suggestion by Bini et al. (2009) that non‐stationarity is common, although other authors have reported that synchrony is described more often than non‐stationarity (Liebhold, Koenig, and Bjornstad 2004). The reasons for the patterns of non‐stationarity identified in this study remain unclear. Inter‐site distance might be one factor affecting the prevalence of non‐stationarity. Distances between sites of less than 100 km are typically those over which synchrony is reported (Koenig 2001), but our investigation encompassed studies that spanned more than 1000 km (Figure 1) which may, in part, explain the prevalence of non‐stationarity in our study. Climatic factors can have important effects on both synchrony (Liebhold, Koenig, and Bjornstad 2004) and non‐stationarity (Vallés‐Medialdea et al. 2022). For example, Compagnoni, Evers, and Knight (2023) described how locations for studies in the conterminous USA needed to be at least 316 km apart to limit correlations in precipitation. However, we recognise that patterns of correlation in climate attributes are likely to vary substantially between regions. Interactions between key drivers of species occurrence are another reason sometimes proposed for non‐stationarity (e.g., Pease, Pacifici, and Kays 2022). We note that interactions between climate and vegetation cover type have previously been quantified for birds in the South West Slopes study. For example, replanted areas in this study were found to be more important for small‐bodied woodland birds than old growth woodland, particularly during drought periods and in areas with long‐term climatic conditions characterised by comparatively high rainfall (Lindenmayer et al. 2018).

We note that the 14 species which exhibited non‐stationarity were a mixture of residents and migrants, large and small birds, insectivores, granivores and nectarivores (Table S1). Similarly, the four species for which there was apparent synchrony were large‐bodied species (Red Wattlebird, Noisy Friarbird), an intermediate‐sized species (White‐throated Treecreeper), and a small species (Superb Fairy‐wren). In addition, these four species were either residents or migrants/blossom nomads. They also had different diets, either nectar or pollen feeders to insectivores (see Table S1). Thus, there appeared to be no readily identifiable life‐history relationships underpinning the patterns of non‐stationarity that we observed. Recent studies have discussed how life‐history strategies may not always be a good surrogate for sensitivity to environmental change (Rademaker et al. 2024). It would be interesting to explore whether rare species, including range‐limited taxa, also exhibit synchrony or non‐stationarity, but we were unable to conduct such analyses because of limited data for these taxa.

The effects we uncovered for spatial variation in species' site occupancy trajectories between long‐term studies were more widespread than those between the experimental contrasts within studies (see Tables S4 and S5). The reasons for such findings remain unclear explicitly, but are likely associated with between‐ecosystem differences in climate, vegetation structure and plant species composition that may influence the temporal availability of resources, including habitat and food availability (e.g., patterns of flowering, fruiting, populations of insect prey). This suggests that patterns of resource availability may be more similar between experimental treatments within a given ecosystem than between the different ecosystems encompassed by our four long‐term studies.

• Q2. Is intraspecific spatial variation in long‐term occupancy patterns (viz non‐stationarity) related to the distance of study sites from each species' niche centroid?

At the outset of this investigation, we predicted that sites closer to species' niche centroids (which are assumed to represent optimal conditions under the abundant‐centre hypothesis (Brown 1984; but see Dallas, Decker, and Hastings 2017)) would be less likely to exhibit negative (declining) trajectories in site occupancy. We found no evidence that position within a species' niche was associated with its temporal trajectory in site occupancy. Notably, our hypothesis about declining trajectory in site occupancy in study areas far from a species' niche centroid was broadly based on the abundant‐centre hypothesis. However, a recent review highlighted the mixed empirical evidence in support of the abundant‐centre hypothesis (Dallas et al. 2020) and attributed this to several factors such as non‐equilibrial population dynamics, species interactions, landscape structure, poor quality data, and inconsistent methodology. We suggest that the first three factors may be relevant to the results reported here given: (1) temporal variation in patterns of site occupancy associated with temporal variation in weather (e.g., see Lindenmayer et al. 2018) potentially contributing to non‐equilibrial population dynamics, (2) marked differences in landscape structure between the different studies (see Table 1), and (3) well‐quantified patterns of species interactions associated with interference competition such as via the actions of a hyper‐aggressive native honeyeater (Lindenmayer et al. 2023). The latter two factors affecting support for the abundant‐centre hypothesis (poor quality data and inconsistent methodology) are unlikely to be important. This was because of intensive and extensive field data collection efforts, coupled with the careful application of similar field protocols employed by the same field staff over the duration of our four long‐term studies. More broadly, a lack of an association between trajectories in species site occupancy and distance from their niche centroids highlights the inherent complexities in understanding the underlying reasons for non‐stationarity.

• Q3. Are long‐term occupancy patterns driven by species tracking their climate niches?

We predicted that if species are tracking their climate niches as documented in other studies (e.g., see Antão et al. 2022; Lawlor et al. 2024), then there would be differences in temporal trajectories in site occupancy between sites at the hotter versus cooler edges and the drier versus wetter edges of their distribution (see also Dallas, Decker, and Hastings 2017; Santini et al. 2019). However, our analyses revealed no associations between climate drivers and non‐stationarity in site occupancy trajectories. As with our analysis of the distance from centroid, the reasons for the absence of these relationships remains unclear. Notably, a recent review indicated that many species have not shifted their distributions in response to climate change or shifted in ways opposite to that expected under changes in temperature (Lawlor et al. 2024). It is possible that other landscape‐scale drivers, such as changes in the amount of habitat or between‐species interactions, and that are not typically studied in range shift analyses, might outweigh the potentially negative effects of climate change‐based modifications in average temperature and rainfall.

5.1. Caveats

We acknowledge that our focus in this investigation was on the most common species for which we had sufficient data in at least three of four long‐term studies. Different patterns might have manifested if we had sufficient data to analyse the trajectories of rare species. Notably, in the case of at least one study—the temperate woodlands on the South West Slopes (see Table 1), rarer bird species have been increasing over time whereas more common species have declined (Lindenmayer et al. 2018). Other studies have shown that common species are often declining at regional and national levels (Inger et al. 2014; Jansen et al. 2020; van Klink et al. 2024). We also acknowledge that our focus here was on birds, which are a relatively mobile taxonomic group, although many species we analysed are sedentary or residents and do not move over long distances once they have established a territory. Nevertheless, the results for less mobile groups such as reptiles might have been different from those we have reported here for birds.

5.2. Ecological Implications

Our analyses contained strong evidence of non‐stationarity with markedly different trajectories for a given species depending on the scale of analysis. We also found evidence of different species temporal trajectories between treatments within studies, between studies and overall (using data combined from all studies) (Figure 2; Figures S1 and S2). Together, these results highlight that a broad generalisation of species trajectories across their entire range risks masking informative trends at smaller scales, such as at the regional, landscape or site level. Failure to determine the influence of non‐stationarity and hence quantify intraspecific differences in population trajectories between different locations could lead to Simpson's Paradox (sensu Simpson 1951), where the averaged pattern for a particular species across a number of sites masks differences in population changes between sites (e.g., Pease, Pacifici, and Kays 2022). Therefore, coarse‐scale analyses based on highly diverse (spatially and ecologically) composite data sets may fail to detect key threatening processes in some places, or conversely, fail to reflect species management successes through targeted interventions in others (e.g., replanting treatment effects found in temperate woodlands in the South West Slopes study; Lindenmayer et al. 2018). In other cases, a need to account for the spatial scale of data aggregation can be important when reporting key trends in populations for a given species, including as part of listings of conservation status. Scale effects and the subsequent risks of misinterpretation of trends are an issue recognised in many studies of ecological phenomena (Samuel et al. 2000; Maas‐Hebner et al. 2015; Crawford et al. 2021). In the case of some of the studies in the investigation reported here, we employed a nested hierarchical study design to guide sampling at site, farm and landscape scales (e.g., see Cunningham et al. 2007) and ensure robust spatial extrapolation from small to larger scales. We suggest that such kinds of hierarchical design criteria might be useful in other studies to facilitate bridging from local‐scale ecology to macro‐ecology.

Finally, our results have implications for the design of monitoring programmes (and other studies of non‐stationarity; e.g., Rollinson et al. 2021) as they underscore the value of documenting temporal patterns of site occupancy for a species across different, spatially separated environments. However, the importance of gathering ecologically, spatially and temporally diverse data sets that can be utilised independently to reveal finer scale differences in species trajectories has been largely overlooked in many articles on the design of monitoring programmes (e.g., Lindenmayer et al. 2012). Distinguishing local fluctuations from consistent patterns of decline across a species' range is critical to inform management decisions (Lindenmayer et al. 2012), and appropriately target resources for ecological interventions and conservation efforts in an era of global change in which many species are thought to be experiencing major declines. Other authors have suggested that spatial replication in long‐term monitoring of plant populations is needed to document their relative responses to climatic variability (Compagnoni, Evers, and Knight 2023). We argue that where analyses of spatio‐temporal data indicate consistent declines across the full range of a species, then concerted management actions will likely be required, including to support threatened species listing (Mace et al. 2008).

Author Contributions

D.L. led all four long‐term research programs and conceived the study. All co‐authors assisted with further conceptualization of the work. M.J.E. led the statistical analysis. D.L. led the initial writing of the manuscript with all co‐authors contributing to edits in manuscript preparation and revision.

Conflicts of Interest

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/ele.14531.

Supporting information

Data S1.

Data Availability Statement

We have now made the data publicly available for download, as requested, via this link: https://doi.org/10.5061/dryad.34tmpg4s1.