Inducing whole-assemblage change by experimental manipulation of habitat structure
R. Mac Nally, Australian Centre for Biodiversity: Analysis, Policy and Management, School of Biological Sciences, Monash University, Melbourne, Victoria 3800, Australia. Tel.: +61 3 9905 5642. Fax: +61 3 9905 5613. E-mail: Ralph.MacNally@sci.monash.edu.au
For many years, ecologists have developed models linking occurrence of individual taxa to specific habitat characteristics (Dueser & Shugart 1979; Rotenberry & Wiens 1980; Holt, Warren & Gaston 2004). Models often are logistic representations of the probability of occurrence of a taxon as a function of habitat elements, such as availability of large trees, tree hollows, refugia, physical or chemical gradients, specific foods, etc. (Fleishman et al. 2003; Gillies et al. 2006). A common approach is to find locations occupied by the taxon and to choose randomly a selection of unoccupied places to develop a ‘contrast’ between the ‘present’ and ‘absent’ locations (Heithaus et al. 2006).
Other workers are interested in more general biotic responses to habitat features, although usually still concentrating on a limited taxonomic range, such as butterflies (e.g. Fleishman et al. 2001), birds (e.g. Fauth, Gustafson & Rabenold 2000), fish (e.g. Crook & Robertson 1999) or invertebrate assemblages (e.g. Jenkins 2005). Relatively complex multivariate models generally are needed to encapsulate the variety of habitat features that appear to define the habitat preferences of individual species or that controls assemblage richness or composition (Mac Nally 2000; Barry & Elith 2006).
The multivariate modelling approach makes it difficult to assess the relative impact or importance of individual habitat elements on assemblages. Ranking habitat elements vis-à-vis their potential for influencing the composition or richness of assemblages is an important step towards understanding how various structural or floristic elements affect assemblages. While the coefficients (with respect to their standard errors) in multivariate models often are thought to provide relative ‘explanatory power’, this is disputable without assessment using validation tests with newly collected data (Mac Nally & Fleishman 2004). Experimental validation often is intractable because models, especially for terrestrial vertebrates, are derived for relatively large spatial scales (> 10s km2) (Loo, Mac Nally & Lake 2007).
a precisely manipulable structural element: fallen timber
Manipulation of many habitat-structural elements in forests and woodlands is difficult to do in a quantitative fashion. The addition of ‘aliquots’ of trees (including important elements such as hollows), shrubs and understorey, generally cannot be done precisely because these components need to regrow or to develop. Differences in site characteristics (e.g. soil fertility or moisture) and often very long-term horizons for regrowth (Vesk & Mac Nally 2006) make such experiments effectively intractable. Therefore, habitat-manipulation experiments generally are done by ‘deletion’ rather than by ‘addition’ of amounts of habitat elements (e.g. Lohr, Gauthreaux & Kilgo 2002). For example, comparisons may be made between plots that have 50% or 25% of the shrubs or trees remaining with ‘control’ plots to assess the effect of that habitat component on biodiversity.
Stripping fallen timber from forests, woodlands and rivers has been a major human activity and cause of ecological change in much of the world (Maser & Sedell 1994). The ecological role of fallen timber in rivers, for example, has been an increasingly important issue in many countries (Crook & Robertson 1999; Gerhard & Reich 2000). The ecological impact of fallen timber loss on forest floors also has an increasing profile (Harmon et al. 1986; Hagan & Grove 1999). A torrent of recent papers has illustrated the severity of the problem of tree loss and associated diminution of fallen timber in many parts of the world (see Bragg et al. 2005). There now are many studies of vertebrates that attribute diminished local diversity to reduced availability of fallen timber, but most studies provide only correlative evidence (e.g. small mammals, Thompson, Baker & Ter-Mikaelian 2003; birds, Lohr et al. 2002; amphibians, Moseley, Castleberry & Ford 2004; reptiles, Bury 2004).
The relatively discrete nature of fallen timber, which has led to its extensive harvesting and widespread depletion, also means that it is one of the few habitat structural elements that is very precisely manipulable (Moseley et al. 2005; Heinemann & Kitzberger 2006). Here, we describe results of a ‘mesoscale’ manipulation of fallen timber in river red gum Eucalyptus camaldulensis floodplain forests in northern Victoria, Australia (Mac Nally 2001). There was 1 year of pre-manipulation bird surveys, followed by experimental distribution of c. 1000 Mg of fallen timber to construct treatments of differing fallen timber loads. Monitoring of effects on the avifauna was conducted for three subsequent years.
We previously have reported on responses of a vulnerable individual species of bird (brown treecreeper Climacteris picumnus Mathews 1912; Mac Nally, Horrocks & Pettifer 2002a; Mac Nally 2006) and also of the only ground-dwelling marsupial found on the floodplains [Antechinus flavipes (Waterhouse) 1838; Mac Nally & Horrocks 2002, Lada et al. 2007]. Here, we generalize those studies by considering whether whole-assemblage structure is affected by wood-load changes. We ask the question: Does variation in one easily identified and precisely manipulable habitat-structural element produce discernible assemblage-wide effects? and if there are identifiable effects, which wood loads produce the most substantial changes and which species of birds respond most to the alterations? We consider the outcomes of this experiment in terms of the implications for optimum management of the floodplain forest for avian biodiversity as a complex spatial and temporal problem.
The experiment was conducted on Gunbower Island (35°42′23″S 144°12′13″E), which lies in an anabranch between the Murray River and Gunbower Creek near Cohuna, in north-central Victoria, Australia. The island formerly flooded almost every year, but with more extreme water extractions and flow regulation, flooding now is much rarer (Crabb 1997). Gunbower Island is intensively exploited for firewood and for post and railway-sleeper (stays) production, so silviculture and wood management are critical issues vis-à-vis biodiversity.
Thirty-four 1-ha plots were marked out (a plan of distribution of plots and treatments is available from the authors). Vegetation characteristics were measured prior to wood-load manipulation. The following variables were measured for each 1-ha plot: river red gums (Eucalyptus camaldulensis Dehnhardt) of these size classes: (1) saplings [> 2 m tall and < 10 cm diameter at breast height (d.b.h.)]; (2) small (10–39 cm d.b.h.); (3) medium (40–59 cm d.b.h.); (4) large (60–79 cm d.b.h.); and (5) very large (> 80 cm d.b.h.). We also measured: (6) hollows (i.e. number of trees with hollows, including coppice-hollows); (7) coppices (number of coppiced trees); (8) stags (number of standing dead trees); and (9) stumps (number). There are very few shrubs and grasses, and forbs change rapidly due to inundation (Ballinger, Mac Nally & Lake 2005). Therefore, these characteristics were not measured.
Wood-load measurements (average 27 Mg ha−1) were done prior to manipulation. We measured all pieces of fallen timber within the 1-ha plot, taking end diameters and lengths of pieces and computing volumes. These were translated into mass by using the mean density of 0·6 Mg m−3 (Robinson 1997). The 34 plots were randomly allocated to eight treatments prior to moving the wood. Sites allocated to different treatments were interspersed to reduce possible spatial effects. Five treatments corresponded to loads of 0 Mg ha−1, 20 Mg ha−1, 40 Mg ha−1, 60 Mg ha−1 and 80 Mg ha−1 (designated 0L, 20L, 40L, 60L, 80L) of aged fallen wood (≥ 10 cm diameter). River red gum is a highly resilient timber that decays slowly (Robinson 1997), so that most of the fallen timber was of Decay Class I or II (i.e. bark largely intact and structural integrity mostly solid; Harmon & Sexton 1996).
Following four rounds of pre-manipulation bird surveys beginning in May 1999, timber was moved in late March 2000. On all of these plots, fallen timber already on the plots was disturbed so that all timber was dislodged from previous footings. There was an undisturbed control treatment (designated UC), where no equipment or persons traversed plots during wood moving. There also was a disturbed control (DC), where no timber was added or removed but there was dislodgement of existing wood and activity on the plots commensurate with the load manipulation plots. Surveys of plots of this treatment were discontinued after six post-manipulation surveys owing to logistic constraints and little evident differences between the UC and DC avifaunas. The other treatment was the imposition of 40 Mg ha−1 of tree crowns on to plots (40H). The 40H treatment emulated current silvicultural practices, where the bole is removed for timber and the crown is deposited for up to 3 year before harvesting the main branches for firewood. Thus, new fallen timber in these production forests often is in the form of crowns. Existing timber was removed from these plots and fresh crowns deposited. There were four replicate plots for each treatment, apart from 0L, for which there were six. In all manipulated plots, timber was evenly distributed over the whole hectare (100 m × 100 m). The middle 50 m × 50 m part of each plot was marked with metal stakes and formed the focus for bird surveys. One thousand Mg of timber was repositioned during this operation, requiring six people for 8 days, hydraulic tandem trailers, a bulldozer and a log-harvesting machine (Mac Nally 2001).
bird survey schedule and method
There were 18 rounds of surveys, each over 7 days, in the following months: May, July and October 1999, January, May, July and September 2000, January, March, April, July, September and November 2001, January 2002, February, April, May and July 2003. The first four rounds were prior to wood-load manipulations, while the latter 14 followed the experimental changes.
Surveys involved the observer (the second author) sitting at one corner of the interior square for 45 min and recording all birds active within the half of the square closest to the observer (a right-triangular area of 0·125 ha). Given that the average number of individual birds observed was 4·61 (see below) and special effort was made to keep track of individual birds, we are confident that there was little double-counting of individuals. While activities of birds outside this small central part of plots were recorded, these data are not considered here. The habitats are uniformly open with little shrub or grass cover so that there were few impediments to definitive sightings of birds using the interior squares. Surveys were conducted either between dawn and 11.00 h Australian Eastern Time or from 14.00 h to 1 h before sunset.
We analysed four sets of data, two of which were species richness and summed densities [= total number of individual birds observed on the plot, i.e. total individuals (0·125 ha)−1] of all birds. The other two sets were richness and summed densities of birds known to use fallen timber and the ground. We used independent foraging data (Mac Nally 1994) to characterize species that might be expected to respond most strongly to fallen timber loads. Twenty-five species had ≥ 30% of foraging actions on the ground, on fallen timber or on low shrubs or perches (Mac Nally 1994); the list of species is available upon request.
The data consisted of an array of 30 sites by 18 survey periods. We used a Poisson model to analyse these data (Gelman et al. 1995). Most of the data were small, non-negative values (< 5), so the use of a counts distribution like the Poisson seemed reasonable. The model is (Mac Nally 2006):
Yj(m)k(n) ~ Poisson(µj(m)k(n)); log(µj(m)k(n)) = αj(m)k(n=1) + αj(m)k(n≠1) (eqn 1)
The Ys are assumed to be Poisson-distributed with means µ and are the observed numbers in plot j in survey k, with the j(m) indicating that site j belongs to treatment m (namely, 0L, ... , UC). The k(n) subscript assigns survey k to period n (n = 1 = pre-manipulation, n = 2 = post-manipulation). The αs for n = 2 are multiplicative constants and, when exponentiated, indicate the degree to which numbers increase or decrease relative to the values for the given treatment before manipulation (i.e. n = 1). From these coefficients, one can estimate the average change in numbers in period 2 relative to period 1 and relate these to changes in the UC treatment sites over the same comparisons:
ΔYj(m)k(n≠1) = (exp(αj(m)k(n≠1)) – 1) × exp(αj(m)k(n=1)), ∂Yj(m)k(n≠1) = ΔYj(m)k(n≠1) – ΔYj(m=UC)k(n≠1) (eqn 2)
The ΔYs are the estimated net increases in numbers in period 2 relative to period 1 for each treatment, while the ∂Ys are these differences relative to the changes that occurred in the UC treatment sites over the same periods. Therefore, the ∂Ys provide changes in numbers corrected for changes in numbers recorded in the control sites.
We used the WinBUGS Bayesian analysis software (public domain, version 1·4, Spiegelhalter, Thomas & Best 2003). Noninformative Normal priors were used for the α parameters (0 means, 10−6 precisions). In all analyses, means and medians of posterior distributions of parameters were similar, indicating symmetric probability distributions. Burns-in of 10 000 and posterior samples of 20 000 were used. We computed Bayes’ factors using the posterior probability distributions (PPD) of the model parameters and comparisons. For a noninformative prior, where the ratio of posterior probabilities for the parameter is unity, the Bayes’ factor is the ratio of PPD/(1 – PPD). Kass & Raftery (1995) argued that Bayes’ factors > 10 (or Bayes’ factors < 1/10 for negative parameters) are strong evidence supporting one hypothesis over another (here that the parameter differs from 0), and Bayes’ factors between 3 and 10 signal positive evidence. We use this inferential framework here.
The flexibility of the WinBUGS programming environment allowed us to compute posterior probability distributions of differences (i.e. post hoc comparisons) in the ∂Y values for all pairs of treatments. For example, during modelling, one can compute the distribution for ∂Y80L,period 2 – ∂Y60L,period 2) to assess whether increases in the 80-L treatment were greater than in the 60-L treatment (both relative to the UC treatment change). These posterior distributions were used to assess whether before–after changes (mainly increases) were greater in some treatments than in others using the Bayes’ factors approach.
Habitat structure consisted of multivariate data. Systematic differences in habitat structure (other than fallen timber loads) were examined by using the analysis of similarity (ANOSIM) routine of PRIMER 5 (Clarke & Gorley 2001), which is designed to compare treatments consisting of multivariate data. Variables first were standardized (subtract means and scale by standard deviations within variables) and then Euclidean distances were computed. This distance matrix was used in the ANOSIM permutations.
similarity-percentage (simper) analyses
We identified those species contributing most to differences between pre- and post-manipulation summed densities by using the SIMPER routine of PRIMER 5. This analysis is appropriate because it partitions Bray–Curtis dissimilarities among pre-defined groups (pre- and post-manipulation surveys in this case) among variables (i.e. bird species). The partition involves the calculation of the fraction of the total dissimilarity among groups attributable to each variable (species). Here, we restrict our attention to species contributing ≥ 5% to the between-group dissimilarities.
Only two pairs of treatments may have differed in habitat structure: (1) 0L vs. 80L, and (2) 0L vs. 20L (both 0·05 < P < 0·10 under permutation). More importantly, there was little evidence of differences in structure between any of the treatments with the control (UC) (all P > 0·43 under permutation). These results indicate that the ∂Y values comparing changes before and after manipulation with the UC changes are unlikely to be related to initial habitat differences per se, and so, are attributable to the manipulation of fallen timber loads.
general characteristics of the avifauna
Fifty-two species of birds were recorded during the 18 survey rounds. There were 2489 records, with the average records per individual survey = 4·61 (n = 30 × 18 = 540 surveys). The average species richness per individual survey = 2·49. There were < 10 records for 24 species.
whole-assemblage responses to manipulation
In four of the treatments, there were substantially greater increases in records of all bird species following manipulation than in the control plots (UC) (Table 1a). This amounted to as much as an extra 1·52 records per survey per replicate plot for the 80L treatment. For comparison, the average number of records per survey was 4·61, so a relative increase of 1·52 (above the change in UC) can be regarded as a 33% relative increase. Using the Bayes’ factors criteria, the treatments could be ranked from biggest increase through to smallest, and this ranking is shown in Table 2(a). The biggest increases were in 80L and 40H, which were similar to each other. Similar inferences were drawn for records for ground-using species (Tables 1c and 2c). Changes in mean species richness were largely similar to the total records, although the magnitude of the statistical evidence of change for the 80L and 40H was much more marked (Tables 1b,d and 2b,d). In these treatments, there were on average > 0·36 more species recorded per survey per replicate plot than in the control. There was some evidence for relative declines in species richness in 0L and 20L treatments of at least 0·18 species per survey per replicate plot (Tables 1b,d and 2b,d).
Table 1. Values of ∂Y (post-manipulation change relative to change in the control treatment [UC]) for Bayesian analyses of (a) all records, (b) species richness of all species, (c) records of species known to forage on ground or in low strata, and (d) species richness of the latter. Data are mean ± SD of model calculations. Records refer to numbers (0·125 ha survey unit)−1 (see text)
|0L||0·29 ± 0·33||–0·18 ± 0·27||0·34 ± 0·29||–0·19 ± 0·25|
|20L||0·12 ± 0·35||–0·29 ± 0·30||0·06 ± 0·32||–0·28 ± 0·27|
|40H||1·46* ± 0·38||0·36* ± 0·31||1·37* ± 0·35||0·43* ± 0·28|
|40L||1·07* ± 0·37||0·08 ± 0·31||0·87* ± 0·34||0·10 ± 0·28|
|60L||0·49* ± 0·36||–0·01 ± 0·30||0·55* ± 0·32||0·07 ± 0·27|
|80L||1·52* ± 0·37||0·45* ± 0·31||1·13* ± 0·33||0·40* ± 0·28|
Table 2. Pairwise comparisons of ∂Y [post-manipulation change relative to change in the control treatment (UC)] for Bayesian analyses of (a) all records (b) species richness of all species (c) records of species known to forage on ground or in low strata and (d) species richness of the latter
|(a) All records||(80L ≈ 40H) > (40L ≈ 60L) ≥ UC ≥ (20L ≈ 0L)|
|(b) All species richness||(80L ≈ 40H) > (40L ≈ 60L) ≥ (UC ≈ 0L) ≥ 20L|
|(c) Ground records||(80L ≈ 40H) > 40L ≥ 60L ≥ 0L ≥ (20L ≈ UC)|
|(d) Ground species richness||(80L ≈ 40H) > (40L ≈ 60L ≈ UC) ≥ (20L ≈ 0L)|
most responsive species
Species that contributed most to differences among the avifaunas of the 40H and 80L plots before and after wood-load manipulation are shown in Table 3. Four species seem especially important (Latin names and authorities listed in Table 3): white-plumed honeyeater, brown treecreeper, yellow rosella and white-winged chough. We present comparable data on numerical changes for these species on the UC plots for reference (Table 3). While the white-plumed honeyeater also increased in the UC plots, densities increased more than twice as much in the 80L treatment plots as in the UC plots and by almost 50% more in the 40H plots (Table 3). The brown treecreeper changed little in the UC plots (0·64 birds ha−1), but increased by > 3·9 birds ha−1 in the 80L and 40H plots. The yellow rosella changed by virtually nothing in UC plots, but increased by > 2·5 birds ha−1 in the 80L and 40H plots. The white-winged chough occurred in much greater densities (5·76 birds ha−1) in the 40H treatment, far in excess of the effectively status quo in the UC plots.
Table 3. Bird species contributing most to differences in avifaunal densities in pre- vs. post-manipulation surveys in the two treatments with substantial changes in abundance (80L and 40H) based on SIMPER analyses, with changes for UC provided for comparison. SIMPER based on summed occurrences for all species
|40H||White-plumed honeyeater Lichenostomus penicillatus (Gould) 1837||21||0·71 ± 0·04||5·68 ± 0·32|
|White-winged chough Corcorax melanorhamphos (Viellot) 1817||18||0·72 ± 0·06||5·76 ± 0·48|
|Brown treecreeper Climacteris picumnus Mathews 1912||13||0·64 ± 0·04||5·12 ± 0·32|
|Yellow rosella Platycercus elegans flaveolus (Gould) 1837|| 8||0·31 ± 0·02||2·48 ± 0·15|
|Willie wagtail Rhipidura leucophrys (Viellot) 1817|| 6||0·27 ± 0·01||2·16 ± 0·09|
|Little friarbird Philemon citreogularis (Gould) 1837|| 5||–0·22 ± 0·01||–1·76 ± 0·08|
|80L||White-plumed honeyeater||28||1·06 ± 0·05||8·48 ± 0·39|
|Brown treecreeper||11||0·49 ± 0·02||3·92 ± 0·15|
|Yellow rosella|| 9||0·44 ± 0·02||3·52 ± 0·15|
|Sacred kingfisher Todirhamphus sanctus Vig. & Horsf. 1827|| 5||–0·16 ± 0·01||–1·28 ± 0·08|
|UC||White-plumed honeyeater||–||0·48 ± 0·03||3·84 ± 0·40|
|White-winged chough||–||0·01 ± 0·02||0·08 ± 0·16|
|Brown treecreeper||–||0·08 ± 0·02||0·64 ± 0·16|
|Yellow rosella||–||–0·01 ± 0·01||–0·08 ± 0·08|
|Willie wagtail||–||0·13 ± 0·04||1·04 ± 0·32|
|Little friarbird||–||–0·02 ± 0·04||–0·16 ± 0·32|
|Sacred kingfisher||–||–0·01 ± 0·02||–0·08 ± 0·14|
The four analyses (Tables 1 and 2) produced a broad agreement on the rank-ordering of the relative effects of the treatments on richness and summed densities of birds in plots following the manipulation. The order was: (80L ≈ 40H) > (40L ≈ 60L) ≥ UC ≥ (20L ≈ 0L). While these results reflect overall assemblage-level changes, three species appeared to be especially important in generating the changes (at least in densities), namely the white-plumed honeyeater, brown treecreeper and yellow rosella (Table 3). Each bird is ‘characteristic’ of river red gum forests throughout much of the Australian inland (honeyeater), the floodplains of the Murray River (rosella) or eastern Australia (treecreeper). The increases in density of the latter two species were not unexpected given their dependence on fallen timber as a foraging substrate (treecreeper) and use of grass seeds and fallen seeds (rosella), where greater amounts of fallen timber may afford more cover.
The strong and consistent response of the white-plumed honeyeater was unexpected given that this species typically gleans invertebrates from arboreal leaves and twigs. Independent data showed that the white-plumed honeyeater foraged only 0·4% on the ground or on logs (Mac Nally 1994). A closely related congener of the white-plumed honeyeater, the fuscous honeyeater Lichenostomus fuscus (Gould 1837), also was thought to be primarily a canopy foliage-gleaning nectarivore (Mac Nally 1994). However, more detailed analyses of diurnal patterns of foraging showed that individuals of this species switch mainly to insectivory, and much of this at lower strata, in afternoons (Timewell & Mac Nally 2004). Therefore, it is possible that the white-plumed honeyeater also has such behavioural and ecological plasticity, which may become expressed when the nature of habitat structure changes.
Given the rapidity of the responses of the avifauna to the experiment (≤ 3 years), these effects are most likely to be mainly due to a spatial concentration of birds from surrounding, lower wood-load locations. We do not have information on whether differential habitat use relates to differential reproductive success (see Walters, Ford & Cooper 1999), but it seems possible that the scales of manipulation (1 ha) may be too small for these apparent habitat preferences to be expressed as differential reproductive outputs. Nevertheless, the current results provide a basis for conducting further experiments with fewer, selected treatments (e.g. 0L, 80L, UC, 40H) pitched at larger spatial scales than those considered here. It seems feasible to increase the size of plots to perhaps 5 ha each, which may be sufficient for studying reproductive responses of at least some of the species (see Cooper, Walters & Ford 2002).
All aged-wood treatments with loads ≤ 20 Mg ha−1 were associated with virtually no important changes in bird richness and densities (possibly even declines), with the changes in the 40L and 60L treatments falling between those for the lower loads and those for the 80L treatment (Tables 1 and 2). The average pre-manipulation wood-load was 27 Mg ha−1, while the average regional loads for much of the southern Murray–Darling basin, within which Gunbower Island is located, are c. 19 Mg ha−1 (Mac Nally et al. 2002b). These results suggest that the current average wood-loads across vast areas of southern Australian floodplain forests are much lower (by 50–60 Mg ha−1 compared with 80 Mg ha−1 loads) than those that would provide conditions for richer, more abundant avifaunas. Further experiments with loads greater than 80 Mg ha−1 would help to establish whether that level is effectively perceived by birds as being similar to pre-European-settlement levels, or whether greater amounts would continue to elevate avian richness and abundances.
It is possible that the strong effects reported here relate to the relatively simple habitat structure of river red gum floodplain forests, which effectively are monocultures with little vegetation layering and open subcanopies and lower strata. Whether similarly clear patterns occur for more structurally complex woodland or forest types would contribute to our knowledge of the relative significance of fallen timber as a controller of assemblages, and also, whether components of habitat structure (e.g. multilayered vegetation, strongly developed shrub layer) mask or suppress effects of other components.
heterogeneity and complexity in habitat structure
Two diverse treatments consistently produced the greatest increases in richness and densities, namely, the 80L and 40H treatments. The 80L treatment effect is interpretable as the influence of a greater availability of a key habitat-structural element, presumably leading to greater habitat complexity (sensu McCoy & Bell 1991). The 40H treatment engendered changes consistently greater than those in the 40L treatment. These differences may be due to the twigs and leaves attached to the boughs. However, the effects continued for 3 years, even though the foliage dried and detached over that time.
These outcomes perhaps should not be unexpected given that the ‘natural’ way in which fallen timber would be delivered is through death and collapse of whole trees. Trees also formerly would have been undercut by flood-waters producing whole-tree falls (Mac Nally & Parkinson 2005). Current management often results in plots with either aged fallen logs and large limbs (such as in the 80L treatment) or, through bole-harvesting, tree-crowns (40H treatment). It is tempting to speculate that an 80 Mg ha−1 treatment of 40 Mg ha−1 aged timber and 40 Mg ha−1 of crowns may create a local hotspot of avian richness and activity exceeding even the effects seen in the 80L and 40H treatments.
landscape design and temporal management
The outcomes of the current study are only the first step in effective management of habitat structure for bird diversity in these forests. Different wood-loads now can be viewed as having demonstrably different degrees of effective habitat quality from the perspective of the focal organisms based on species richness and utilization (Church et al. 2000). This characterization would be even more compelling if one could gather information on reproductive outcomes in different treatments (Mac Nally 2006). What would be an optimal configuration of fallen timber loads over the 20 000 ha of Gunbower Island? It seems possible that tools developed for reserve design (e.g. McCarthy, Thompson & Possingham 2005; Moilanen 2005) might be used. The problem is not just a spatial one, but also a temporal one. Optimal decisions must be made in the context of a temporal sequence of management actions (Meir, Andelman & Possingham 2004), in this case, provision of relatively large amounts of a mercantile commodity – timber. There has been little consideration in much of the reserve design literature about the decline in quality of habitats through time. This is especially important in the current context because, while durable, river red gum timber decays eventually and this will lessen the effective quality of a site with a given load. Moreover, much of the island has been held in an arrested state of maturity to service production requirements, so there are few large, old trees available to naturally provide fallen timber as existing stocks decay. Therefore, planning must have a temporal component involving a strategy by which replacement wood ‘comes on line’ at appropriate times to sustain the optimum design and placement.
We gratefully acknowledge the support of the Australian Research Council through Grant Nos F19804210 and A19927168 and the generous support of the Hermon Slade Foundation (http://www.hermonslade.org.au/background.html, Grant HSF 02–3). We thank Erica Fleishman, Andrew Bennett and Sam Lake for either reading the manuscript or contributing to parts of the work. Work was conducted under the specifications of the Monash University Animal Ethics Committee application BSCI/2002/11. The North-western Region of the Victorian Department of Natural Resources and Environment (Robert Price) was especially co-operative in performing the manipulation. This is contribution 82 from the Australian Centre for Biodiversity: Analysis, Policy and Management.