SEARCH

SEARCH BY CITATION

Keywords:

  • Acacia harpophylla;
  • Brigalow;
  • hierarchical Bayesian models;
  • individual-based models;
  • multi-level models;
  • Queensland

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

1. Restoration thinning involves the selective removal of stems in woody ecosystems to restore historical or ecologically desirable ecosystem structure and processes. Thinning may also accelerate carbon sequestration in dense regenerating forests. This study considers restoration thinning effects on both structural development and carbon sequestration in a regenerating forest ecosystem.

2. An experimental thinning trial was established in dense Acacia harpophylla regrowth in southern Queensland, Australia. The mean stem density prior to thinning was 17 000 stems ha−1. Four treatments (no thinning and thinning down to 1000, 2000 and 4000 stems ha−1) were applied in a randomized block design. Growth and mortality of a subset of stems was monitored for 2 years. Mixed-effects models and hierarchical Bayesian models (HBMs) were used to test for treatment effects and to explore relationships between neighbourhood density variables and the growth and mortality of stems. The HBMs were subsequently used to parameterise an individual-based simulation model of stand structural development and biomass accumulation over 50 years.

3. The circumference growth rates of stems in thinning treatments were significantly higher than in the control. Woody species diversity and grass cover were also significantly higher in thinning treatments and were strongly negatively correlated with canopy cover. The HBMs confirmed that both growth and mortality were density dependent to some extent.

4. The simulation model predicted a net gain in living above-ground biomass in some thinning treatments (compared with the control treatment) within 20 years after thinning. The 6000 stems ha−1 treatment was predicted to be the optimal thinning density for structural development towards the structure of a nearby mature reference forest.

5.Synthesis and applications. Naturally regenerating woody vegetation provides important habitat for native fauna in fragmented landscapes and represents an efficient means to reinstate habitat connectivity and increase forest area. Many regrowth ecosystems also have considerable potential as land-based carbon sinks. This study demonstrates that restoration thinning can be applied to accelerate stem growth and woody species recruitment and may also accelerate structural development and carbon sequestration in this extensive regrowth ecosystem. The application of restoration thinning to provide dual restoration and carbon benefits should be explored for a wider range of naturally regenerating woody ecosystems.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Thinning of stands has been used in forestry to increase the yield of desirable timber for centuries (Brandl 1992). More recently, thinning has been applied to forest ecosystems in an attempt to restore historical structure and to reinstate ecological processes that have been disrupted by human-induced land use changes (e.g. Allen et al. 2002). This type of thinning is commonly referred to as restoration thinning. An additional, but largely unexplored benefit of restoration thinning in regenerating forests can be the acceleration of carbon sequestration (Vargas, Allen & Allen 2009), suggesting that restoration and carbon sequestration goals may be congruent, although this is more likely over long-time frames (Dwyer et al. 2009). Longer term goals are difficult to assess in recovering forest ecosystems because of the time periods involved, highlighting the need for predictive models to assess goal compatibility and to explore long-term thinning responses at the individual stem and forest stand scales. Although thinning studies are common in the forestry literature, few ecological restoration studies have explored thinning responses at the individual stem scale (but see Stone, Kolb & Covington 1999; Vargas et al. 2009) or predicted long-term structural development and carbon sequestration under different thinning scenarios.

Restoration thinning has been most widely applied and documented in Pinus ponderosa Douglas ex C. Lawson (ponderosa pine) ecosystems in the United States (Allen et al. 2002), but it has also been applied to secondary tropical dry forest in Mexico (Allen et al. 2003; Vargas et al. 2009), shrub-steppe and Picea rubens Sarg. forest in the United States (Olson & Whitson 2002; Rentch et al. 2007) and secondary Acacia koa A. Gray forest in Hawaii (Pearson & Vitousek 2001). In the case of ponderosa pine ecosystems, livestock grazing, fire suppression and logging have promoted the emergence of dense cohorts of saplings. These dense cohorts alter and slow structural development, and alter fire frequency and intensity (Covington & Moore 1994) and nutrient cycling (Kaye & Hart 1998). The implied stalling mechanism in these ecosystems is intense intraspecific competition combined with low levels of density dependent mortality (Allen et al. 2002). Restoration thinning operates at the individual stem scale by reducing local competition and at the stand scale by facilitating structural development. It is therefore important to accommodate multiple spatial scales when investigating thinning responses (Kariuki 2008).

Few studies have explored the effects of restoration thinning on carbon sequestration in naturally regenerating forest ecosystems or jointly considered thinning effects on both carbon sequestration and restoration. In 11-year-old tropical dry forest regrowth, restoration thinning removed almost 5 t C ha−1 in above-ground biomass, but carbon recovery in both above and below-ground pools was rapid (equivalent to unthinned stands within 5 years; Vargas et al. 2009). In the absence of natural disturbances, the thinned tropical dry forest is expected to continue rapid sequestration into the future (Vargas et al. 2009).

This study explores responses to restoration thinning in a dense 29-year-old stand of Acacia harpophylla F. Muell. ex Benth. dominated regrowth in Australia. Our restoration goal was to develop structure and biomass comparable with a minimum acceptable reference ecosystem in the fastest possible time. We selected a reference ecosystem (Fig. 1) with structure and living above-ground biomass equivalent to an adjacent mature forest (never cleared), but chose not to commit to the same level of floristic diversity found in the mature forest given the limited available data on plant species recruitment in this ecosystem.

image

Figure 1.  Possible restoration scenarios: (1) represents the current state of the regrowth ecosystem (moderate biomass, low complexity), (2) represents the minimum acceptable reference ecosystem and (3) represents an adjacent mature forest i.e. the most desirable state. The dark grey line indicates the minimum acceptable trajectory following restoration thinning and the black line indicates the ideal restoration trajectory following clearing. The space in between is the acceptable restoration space. Given the high density of Acacia harpophylla stems, it is very unlikely that floristic complexity would develop without structural change (i.e. the system is unlikely to enter the space to the right of the black line).

Download figure to PowerPoint

Our carbon goal was to maximize the amount of above-ground biomass accumulated over the next 50 years. We focus on above-ground carbon here, but refer to Dwyer et al. (2009) for a brief summary of available information relating to below-ground stocks in Brigalow forests. We hypothesized that: (H1) thinning accelerates the growth rate of remaining stems, (H2) recruitment of woody species and grasses is greater in thinned stands than unthinned stands, (H3) thinning accelerates restoration and carbon sequestration at the stand scale and (H4) thinning reduces the amount of above-ground carbon lost through decay over 50 years. The fourth hypothesis explores the trade-off between carbon lost initially through restoration thinning and carbon lost over time through self thinning. Multi-level models were used to test treatment effects, explore neighbourhood effects on the growth and death of individual stems and to parameterise an individual-based model (IBM) of growth and mortality. We conclude by providing management recommendations to maximize dual carbon and restoration benefits.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Study ecosystem

Brigalow is the common name applied to A. harpophylla and to the ecosystems in which it occurs. Broad-scale pastoral development in northeastern Australia over the last 60 years has caused a dramatic reduction in the extent of Brigalow ecosystems. In mature states, A. harpophylla forms high biomass forests and woodlands up to 20 m tall. Sexual reproduction in A. harpophylla is sporadic; however, stems are capable of prolific resprouting following clearing and dense clonal stands (referred to here as Brigalow regrowth) have emerged extensively throughout working and abandoned pastures (Butler 2009). Brigalow regrowth also has considerable potential for carbon sequestration (Dwyer et al. 2009); however, the characteristic high stem densities can slow biomass accumulation and structural development in the longer term (Dwyer, Fensham & Buckley 2010).

Study area

The thinning trial was established at Bulli State Forest (28°1′S, 150°55′E) in southeast Queensland (Fig. 2). Mean summer and winter maximum temperatures are 32·8 and 17·7 °C respectively (Bureau of Meteorology 2009a). Rainfall is summer dominant but variable, with a mean annual value of 600 mm (Houlder et al. 2000). Annual rainfall (April–March) for 2006/2007, 2007/2008 and 2008/2009 was 267, 715 and 400 mm respectively (Bureau of Meteorology 2009b).

image

Figure 2.  Location and local context of the study site.

Download figure to PowerPoint

The State Forest includes areas of deeper clay soils that support open forest of A. harpophylla and Casuarina cristata Miq., and Brigalow regrowth in disturbed areas. Five hectares of Brigalow regrowth along the western boundary (Fig. 2) was initially cleared by bulldozer in 1967 and again in the early 1980s making the current regrowth c. 29 years old. This dense regrowth is dominated by A. harpophylla (mean density of 17 000 stems ha−1) with stems ranging in height up to 7 m. Scattered individuals of Eucalyptus pilligaensis Maiden and C. cristata are present. Soils are cracking clays (vertosols) with limited development of gilgais (natural depressions). The mature reference stand (Fig. 2) corresponds to the ‘remnant’ site described in Chandler, Buckley & Dwyer (2007) and ecosystem 3 in Fig. 1. Casuarina cristata is dominant in this reference site, comprising ∼75% of total living above-ground biomass. The experimental site is fenced and has not been grazed by livestock since the emergence of dense suckers following the last clearing attempt. The mature reference site is not fenced, but is buffered from adjacent pasture and cropping by a 200 m wide strip of 40-year-old Brigalow regrowth (advanced regrowth; Fig. 2). Refer to Table S1 in Supporting Information for a summary of mature reference forest data.

Experimental design

Sixteen 25 m × 25 m (625 m2) quadrats were established between October 2006 and March 2007 with a minimum buffer distance of 5 m between quadrats and 20 m from clearings. All stems ≥1 cm diameter at 30 cm above-ground level (AGL) were mapped and their diameter measured to the nearest centimetre. Stems were recorded as separate individuals even if they were joined just AGL. This was performed because it was not possible to determine connectivity between all neighbouring stems without excavation.

Treatments were assigned to quadrats using a randomized block design, with four blocks chosen based on pre-thinning stem densities. Four treatments were applied: control (no thinning), thinned to 4000 stems ha−1 (250 stems quadrat−1), thinned to 2000 stems ha−1 (125 stems quadrat−1) and thinned to 1000 stems ha−1 (63 stems quadrat−1). Once mapped, stems were randomly selected for retention. Thirty stems were monitored from an internal 18 m × 18 m quadrat (324 m2), providing a 3·5 m buffer to reduce edge effects. While a wider buffer would have been ideal, this was precluded by the area of regrowth available. The circumference of monitored stems was measured to the nearest millimetre and a band was painted around each stem at 30 cm AGL for all future measurements. Circumference measurements at 30 cm AGL enabled use of a published allometric equation for A. harpophylla (Scanlan 1991). Only stems >7 cm circumference were monitored, smaller stems commonly branched below 30 cm AGL. Larger stems selected for thinning were ringbarked using an angle grinder with wood cutting blade or a manual chain girdler. A pocket knife was used for smaller stems. Ringbarking was undertaken in February and March 2007 when soil moisture was relatively high to minimize the risk of secondary suckering (Johnson 1964).

Circumferences of the 480 monitored stems were measured in early April 2007 and 2009 and mortality was also recorded in 2009. A random subset of 14 stems was measured 10 times to obtain an estimate of measurement error. In 2009 canopy cover, ground cover and woody shrub density were measured in each quadrat. Cover was recorded as present (1) or absent (0) at 10 random points along two 18-m transects and converted to a proportion. The number and species of woody stems (excluding A. harpophylla, herein referred to as ‘non-Brigalow woody stems’) were recorded 1 m either side of the cover transects to provide estimates of density and diversity per quadrat.

Statistical analysis of treatment effects

Mixed effects anova were used to test for treatment differences in circumference growth rates (CGR) and quadrat-scale variables (canopy cover, grass cover and non-Brigalow stem diversity and density). A mixed logistic model with logit link and binomial error distribution was used to test for treatment effects on stem survival (binary, alive or dead in 2009). We calculated the mean annual CGR of each stem as follows:

  • image

Where circ 09, and circ 07, are the circumference measurements of stem i in 2009 and 2007 respectively and the division by two provides the CGR per year. For quadrat-scale variables, block was included as a random effect and for stem-scale variables, both block and quadrat (nested within block) were included as random effects. The following orthogonal a priori contrasts were applied in all models of treatment effects – control vs. others, 1000 vs. 2000 and 4000 and 2000 vs. 4000. CGR was square root transformed and non-Brigalow stem density was log transformed to meet the assumptions of anova. Additional linear mixed effects models were used to explore relationships between canopy cover and other quadrat-scale continuous variables. To make inferences about parameter values in anova and linear models, 10 000 Markov Chain Monte Carlo samples were taken from parameter posterior distributions and 95% highest posterior density (HPD) intervals were calculated. Wald z statistics were consulted for the abovementioned mixed logistic model (Bolker et al. 2009). All analyses of treatment effects were undertaken using the lme4 package (Bates & Maechler 2009) in r 2.2.9 (The R Foundation for Statistical Computing 2009).

Hierarchical Bayesian models of growth and mortality

The experimental design included three spatially nested scales (stem, quadrat and block). Therefore, a multi-level modelling approach was adopted where intercepts were permitted to vary by quadrat and in the growth model, slopes [with respect to ln(circ 07)] were also permitted to vary by quadrat. We herein refer to these parameters as ‘varying quadrat intercepts’ and ‘varying quadrat slopes’. Residual (stem-level) variance was also allowed to vary with treatment in the growth model. Block was not included in either model because variance between blocks was effectively zero for both responses.

The growth model response variable was ln(circ 09) which was modelled as a function of ln(circ 07) and other stem- and quadrat-level explanatory variables. The two circumference variables were log transformed to meet assumptions of normality and constant variance. The mortality model response variable was binary and so a logistic regression with logit link was adopted. Mortality was modelled as a function of stem size (circ 07) and other stem- and quadrat-level explanatory variables. We chose three stem-scale neighbourhood variables to explore the influence of neighbourhood space or density on growth and survival using A. harpophylla stems only. The neighbourhood variables were (1) available space (Theissen polygon area, Mead 1966), (2) neighbour BA (basal area of stems within 3·5 m of target stems) and (3) neighbour density (number of stems within 3·5 m of target stems). A 3·5 m radius was chosen to correspond with the minimum buffer distance applied between quadrat edges and monitored stems. Square root transformations of the three neighbourhood variables were also assessed. At the quadrat scale, we tested mean available space (mean Theissen polygon area in each quadrat), quadrat BA (basal area within each quadrat) and quadrat density (density of stems in each quadrat).

Because the neighbourhood variables were correlated, each combination of one stem-level variable and one quadrat-level variable was assessed separately. We used the Deviance Information Criterion (DIC, Spiegelhalter et al. 2002) to guide model selection. Each combination was fitted, checked for adequate convergence (graphically and using inline image values) and parameter posterior distributions were examined. The combinations that resulted in the lowest DIC values were selected. Interactions between the chosen explanatory variables and stem size were then assessed one at a time. In both the growth and mortality models all interaction terms were estimated near zero with high uncertainty and were not retained. The specification of varying quadrat coefficients was not altered during the model selection process.

Individual-based model

An IBM of growth and mortality was parameterised using the regression coefficients and variance estimates from the growth and mortality hierarchical Bayesian models (HBMs) to simulate stand development and carbon sequestration over a 50-year period following thinning (Fig. 3). The IBM was written specifically for this study and executed in r (The R Foundation for Statistical Computing 2009). We used 400 simulations of seven treatments in each of the 16 quadrats through 25 time steps (50 years). Simulated treatments were an unthinned control and 1000, 2000, 4000, 6000, 8000 and 10 000 stems ha−1. Each stem was randomly assigned a maximum circumference from a distribution of maximum circumferences derived from the mature reference forest quadrat data. A stem asymptote function slowed growth near the maximum stem size and a biomass function (Scanlan 1991) converted circumference to biomass. Decay of dead biomass was also modelled using available estimates of decay constants (Hart 1995; Mackensen, Bauhus & Webber 2003). Refer to Methods S1 and Figs S1–S4 in Supporting Information for a full description of the IBM.

image

Figure 3.  Flow diagram of the individual-based model showing only components relating to living above-ground biomass (t ha−1). For the growth model components, circijt is the predicted circumference in cm, αj is the intercept for the jth quadrat, λ is the population mean growth intercept, ujt − 1 is the quadrat-level predictor at the previous time point, γ is the corresponding regression coefficient, σ2 is the quadrat intercept variance, bj is the slope of the relationship between ln(circ 07) and ln(circ 09) for the jth quadrat, xijt − 1 is the stem-level neighbourhood predictor calculated for stem i in quadrat j at the previous time point, βg is the corresponding regression coefficient, εijk is the random stem effect for treatment k which does not change with time. For the mortality model component, Pr(survivalijt) is the predicted probability of survival, πj is the intercept for quadrat j, zijt − 1 is the stem-level neighbourhood predictor for stem i in quadrat j at the previous time point and βm is the corresponding regression coefficient.

Download figure to PowerPoint

Living, dead and total above-ground biomass accumulation curves were generated for each treatment. Dead above-ground biomass was assumed to be zero at the start of the simulation period. The median predicted size-class distribution of circumferences (5 cm classes) for each treatment at each time step was compared with the continuous distribution of circumferences from the reference forest (pooled across all tree species and all quadrats) using two-sample Kolmogorov–Smirnov (KS) tests. KS tests require that both samples being compared are continuous. We therefore converted the predicted discrete median distributions to continuous samples by randomly assigning each stem a value from a uniform distribution bounded by the stem’s size class (e.g. a stem in the 1–5 cm class could take any value between 1 and 5). Repeated testing indicated that the differences generated by these random draws did not change test results. To test this further, we ran KS tests on the discrete predicted median distributions and a discrete version of the mature forest circumference sample (same 5 cm classes applied). This approach yielded almost identical results to the analysis of continuous samples.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Treatment effects

The CGR of stems in thinned quadrats was significantly higher than in control quadrats, and was significantly higher in 1000 and 2000 stems treatments than in the 4000 stems treatment (Fig. 4a). Mortality was not significantly different in control quadrats compared with thinned quadrats (Wald z = 1·03, P = 0·30), or among thinning treatments (1000 vs. 2000 and 4000 stems: Wald z = −1·03, P = 0·30, 2000 vs. 4000 stems: Wald z = 0·07, P = 0·94). All grasses and non-Brigalow woody species recorded during the final survey were native. The Shannon index of diversity and density of non-Brigalow woody stems was greater in thinned quadrats, but no differences were detected among thinning treatments (Fig. 4b,c). The diversity of non-Brigalow stems in the thinning treatments was not statistically different from the mature reference ecosystem, but was significantly lower in the control treatments (two-sided Mann–Whitney U-tests, not shown). Grass cover followed the same hierarchy as CGR (Fig. 4d). The lower grass cover in the 4000 stems treatment can be explained by the significant negative effect of canopy cover on grass cover (Table 1). The average canopy cover in the 4000 stems treatment was 50% which appears high enough to prevent the proliferation of grass. Canopy cover was also significantly negatively related to grass cover, non-Brigalow stem density and non-Brigalow stem diversity (Table 1).

image

Figure 4.  Plots of treatment effects: (a) box and whisker plot of pooled circumference growth rates (CGR) from the four quadrats for each treatment, (b) bar plot of mean Shannon Index of diversity of non-Brigalow woody stems for each treatment, (c) bar plot of mean non-Brigalow woody stem density for each treatment and (d) bar plot of mean grass cover for each treatment. Letters denote statistically significant groups as determined from mixed effects anova. In (a) thick lines are medians, boxes are interquartile ranges and whiskers indicate total ranges after accounting for outliers. Bars in (b), (c) and (d) are standard errors of treatment means (n = 4).

Download figure to PowerPoint

Table 1.   Summary of linear mixed effects models of various response variables as functions of canopy cover (proportion)
Response (units)Estimate95% HPD intervalsBetween quadrat varianceBetween-block variance
  1. HPD, highest posterior density intervals generated from 10 000 Markov Chain Monte Carlo samples.

Non-Brigalow Shannon Index of diversity−1·075−1·645, −0·5270·0590·014
Ln(non-Brigalow stem density) (stems ha−1)−2·229−3·243, −1·2300·2070·000
Grass cover (proportion)−0·428−0·605, −0·2510·0060·002

Hierarchical Bayesian models of growth and mortality

Model comparisons using DIC values indicated that sqrt(available space) was the best stem-level predictor of both growth and mortality. Quadrat BA was retained in the final growth model at the quadrat-level; however, no quadrat-level predictors proved informative for mortality. Stem size (circ 07) was also not retained in the final mortality model (refer to Supporting Information for DIC values of all models considered (Table S2) and detailed descriptions of the final growth and mortality models (Methods S1)).

In the final growth model (Table 2), measurement error accounted for 1·6–9·6% of the total residual (stem-level) variance depending on the treatment. Sqrt(available space) explained 27–46% of stem-level variance after accounting for measurement error. Quadrat BA explained additional variation in growth (Fig. 5a) and combined with sqrt(available space), explained 78% of variation in size-dependent growth between quadrats. Unexplained variance between stems increased with thinning intensity. In the mortality model sqrt(available space) had a positive effect on survival (Fig. 5b). The average survival rate over the 2-year period across all quadrats was 91·2%, however quadrat survival rates ranged between 77% and 95%.

Table 2.   The final growth model, response variable ln(circ 09). Symbols correspond to those used in Fig. 3. Refer to table footnote for units of variables
Explanatory variable/variance componentSymbolEstimate95% Credible intervals
  1. *Variables mean centred (but not standardized) using the following mean values: ln(circ 07) = ln(12·94 cm) = 2·56, sqrt(available space) = sqrt (3·5 m2) = 1·87, quadrat BA = 8·87 m2 ha−1.

Stem level
 Ln (circ 07)* (overall slope)Not shown0·9820·96, 1
 sqrt(available space)*βg0·01450·0089, 0·0205
 Stem variance (1000 stems ha−1)σ2yk0·00240·0019, 0·0034
 Stem variance (2000 stems ha−1)σ2yk0·00210·0016, 0·0029
 Stem variance (4000 stems ha−1)σ2yk0·00130·0009, 0·0018
 Stem variance (control)σ2yk0·00040·0002, 0·0005
 Measurement varianceNot shown0·000040·00003, 0·00005
Quadrat level
 Overall interceptλα2·682·66, 2·69
 Quadrat BA#γ−0·0025−0·0039, −0·0012
 Variance of quadrat interceptsσ2α0·00040·0002, 0·0011
 Variance of quadrat slopes [slopes for ln(circ 07)]Not shown0·00090·0001, 0·0033
image

Figure 5.  Selected plots from the growth and mortality models: (a) Quadrat BA vs. the varying quadrat intercepts from the growth model (the αj’s in Fig. 3) and (b) available space vs. the probability of survival. The bars in (a) are 95% credible intervals. In (b) the grey lines are the predicted curves for each quadrat (resulting from varying quadrat intercepts – the πj’s in Fig. 3) and the black line is the overall mean predicted relationship. To generate the points in (b), binary survival values were ordered according to ranked available space, binned into groups of 16 and means were calculated for each of the bins.

Download figure to PowerPoint

Individual-based model results

Asymptotic living above-ground biomass values predicted by the simulation model were realistic and were driven largely by the maximum stem circumferences assigned to each stem from the empirically derived distribution of maximum circumferences. The mean living above-ground biomass of the adjacent mature reference forest is 110·3 t ha−1 (SD = 26·7 t ha−1). The median asymptotic values predicted for the 6000–10000 stems treatments were all close to 120 t ha−1 (Fig. 6a, pooled 95% predictive intervals = 114–125·1 t ha−1). The asymptotic value for the 4000 stems treatment was slightly lower, though not significantly so. The 1000 and 2000 stems treatments had significantly lower values because of lower numbers of stems contributing biomass (Fig. 6a,b). All thinning treatments reached asymptotic living above-ground biomass values within the 50 year simulation period, first the more severe treatments (40–44 years after thinning) and then the least severe treatments (48–50 years after thinning). The control treatment did not asymptote within the 50 year period (Fig. 6a), but did attain a maximum living above-ground biomass value of 115·1 t ha−1 (110·5–119·4 t ha−1). Predicted median accumulation curves for the 4000–10000 stems treatments all surpassed the median trajectory of the control treatment within 22 years following thinning.

image

Figure 6.  Median predicted trajectories from 0 to 50 years after thinning: (a) living above-ground biomass accumulation, (b) stem density decline, (c) dead above-ground biomass accumulation and (d) total above-ground biomass accumulation. All line types follow the legend in (a). The solid thick grey horizontal lines in (a) and (b) respectively correspond to the mean living above-ground biomass and mean tree species stem density recorded in the mature reference forest. 95% predictive intervals were omitted to enhance clarity, refer to Figs S2–S4 in Supplementary Information.

Download figure to PowerPoint

Accumulation of naturally dying biomass commenced at the second time step in the model, so dead above-ground biomass declined in all thinning treatments initially because of the decay of thinned stems (Fig. 6c). Dead above-ground biomass declined for the entire simulation period in the 1000 stems treatment because of the decay of large amounts of thinned biomass and lower rates of density dependent mortality. In the other thinning treatments dead above-ground biomass began to increase at varying rates, depending on the severity of thinning. Dead above-ground biomass in all treatments reached asymptotic values within the simulation period and actually began to decline slowly in the final few time steps. Final dead above-ground biomass values were greatest in the control treatment (41·1–46·2 t ha−1), but not significantly greater than the 8000 and 10 000 stems treatments. The predicted total above-ground biomass accumulation curves (Fig. 6d) indicate that the highest values after 50 years would occur in the 8000 and 10000 stems treatments (∼161 t ha−1), but these were not significantly greater than predictions for the control and 6000 stems treatments.

The greatest amount of above-ground biomass lost to decay of dead stems (including thinned biomass) over the simulation period occurred in the control, 8000 and 10 000 stems treatments (Table 3), due mainly to higher rates of self thinning. The most severe treatments lost the least above-ground biomass to decay, but they also accumulated less because of the low number of surviving stems.

Table 3.   Predicted median (95% predictive intervals) values for total above-ground biomass at the end of the simulation period, predicted above-ground biomass lost through decay during the simulation period (including thinned biomass) and predicted net gain in above-ground biomass at the end of the simulation period
Treatment (stems ha−1)Total above-ground biomass after 50 years (t ha−1)Biomass lost over 50 years (t ha−1)Net gain after 50 years (t ha−1)
100095·6 (90·9, 100·3)76·4 (73·4, 79·4)19·2 (17·5, 20·9)
2000121·5 (116·8, 126·7)92·2 (88·4, 96·0)29·3 (28·4, 30·6)
40001145·5 (141·0, 150·1)110·2 (106·5, 114·1)35·3 (34·5, 36·0)
60001154·8 (150·5, 159·8)120·8 (117·1, 124·2)34·0 (33·4, 35·6)
8000159·2 (154·7, 163·2)127·5 (124·1, 131·4)31·7 (30·6, 31·8)
100001161·2 (157·2, 165·3)131·6 (127·9, 135·1)29·6 (29·3, 30·2)
Control159·0 (155·2, 162·8)133·3 (130·5, 136·6)25·7 (24·7, 26·2)

After 50 years, predicted stem densities in the 1000–4000 stems treatments were all below 1000 stems ha−1 (Fig 6a), compared with mean tree density of 1630 ha−1 in the mature reference forest. Predicted final densities in the 6000–10 000 stems treatments were similar to the mature reference site. Although tree densities declined considerably in the control treatment, the median density after 50 years was still 2310 stems ha−1.

Predicted circumference distributions for all treatments were significantly different to that of the mature reference forest over the entire simulation period (not shown). This was because of the absence of larger stems at earlier time points and the absence of very small stems at later time points (recruitment was not modelled). Given the low resolution provided by these significance tests we consulted the KS test statistic (D) for each comparison. Low D values indicate a higher probability that the two circumference samples come from the same distribution. We recorded the years that the value was lower the 0·3 (a low but arbitrary value) to provide a relative indication of when and for how long the predicted distributions achieved a moderate level of similarity to the mature reference forests (Table 4). In general, the duration of moderate similarity increased as thinning intensity decreased. Distributions in the more severe thinning treatments became similar sooner because of more rapid growth of some stems to larger sizes. However, these treatments were unable to maintain similarity for more than 20 years and eventually developed negatively skewed or bimodal distributions (refer Fig S4 in to Supporting Information).

Table 4.   The periods when the Kolmogorov–Smirnov (KS) test statistic (D) was below 0·3 for tests between predicted median circumference distributions and the pooled distribution of mature forest circumferences
Treatment (stems ha−1)Period (years) when D statistic < 0·3Year of highest similarity (D)
  1. Also shown is the year when the D statistic was lowest and hence the distributions were most similar. Values in brackets are corresponding minimum D values.

10006–1610 (0·092)
20006–2012 (0·094)
40008–2614 (0·095)
60008–3216 (0·092)
80008–3618 (0·105)
1000010–4020 (0·108)
Control14–5028 (0·10)

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

There was strong support for all four hypotheses. CGR of stems were higher in the thinned quadrats (H1) and were strongly positively related to the space available to each stem and negatively related to the basal area in each quadrat. Support for hypothesis 2 was unequivocal; 2 years after thinning, all thinned quadrats had significantly greater non-Brigalow woody species diversity, non-Brigalow woody stem density and grass cover than control quadrats. The IBM predicted that thinning to densities between 6000 to 10 000 stems ha−1 will accelerate both structural restoration and carbon sequestration at the stand scale compared with unthinned stands (H3). The IBM also predicted that thinning increases the net gain of carbon over 50 years (H4).

Thinning for restoration

The most suitable reference ecosystem to assess the IBM predictions is the ‘minimum acceptable reference ecosystem’ (ecosystem 2 in Fig. 1) because the predicted stands were A. harpophylla monocultures. We used pooled tree species data from the mature reference forest to represent this minimal acceptable ecosystem. The treatments that consistently remained in the acceptable restoration space with regard to living above-ground biomass and stem densities were the 6000–10 000 stems treatments. These treatments also maintained size-class distributions similar to the mature forest for more than 20 years, providing a longer window of opportunity for recruitment to occur. Minor secondary suckering in A. harpophylla was observed following thinning but was regularly controlled during the monitoring period. Suckering occurs in most Brigalow ecosystems in response to stochastic disturbance events and can be manually stimulated by abrasions to shallow lateral roots (Johnson 1964). Widespread flowering occurred in the mature reference forest in late 2007 during the first major A. harpophylla sexual reproduction event since the 1970s. No flowers, seeds or seedlings were observed in the dense 29 years old regrowth despite the presence of some larger stems that were of equivalent diameter and height to reproductive stems in the reference forest, suggesting that intense intraspecific competition suppresses sexual reproduction. The mean density of seedlings in the mature reference forest immediately following germination was 18 seedlings m−2, but only 14% survived the first year (John M. Dwyer, unpublished). No data are available on the longer term survival of A. harpophylla seedlings.

Good recruitment of non-Brigalow woody species in thinned quadrats contrasts with limited responses recorded in restoration thinning studies in ponderosa pine forests (Laughlin et al. 2008; Nelson, Halpern & Agee 2008). Non-Brigalow woody species diversity and grass cover were strongly negatively related to canopy cover suggesting that light is a limiting factor for woody and grass species recruitment and initial growth, but competition for belowground resources may also be important. Recruitment observed in this study probably represents a best case scenario in terms of connectivity to mature forests and size of the regional species pool. Many regrowth patches occur in heavily cleared parts of the landscape where there are very few patches of mature forest to function as seed sources (Butler 2009).

Although all recorded grass species were native, higher grass cover in thinned quadrats is alarming. The exotic grass Pennisetum ciliare (L.) Link, commonly known as buffel grass, is the dominant pasture species throughout the central and northern Brigalow Belt. It can invade open Brigalow forests and establish a grass-fire feedback that kills A. harpophylla and co-occurring species and promotes further grass invasion (Butler & Fairfax 2003). Severe thinning is therefore likely to facilitate invasion of exotic grasses in regrowth and increase fire risk. Our results indicate that maintaining canopy cover of at least 50% should minimize grass establishment, but higher cover may be required to manage buffel grass invasion.

The microhabitat provided by course woody debris in Brigalow forests is important for the persistence of many threatened reptile species including regional endemics (Richardson 2006). Similarly, the endangered bridled nailtail wallaby Onychogalea fraenata Gould preferentially utilizes hollow logs for shelter in Brigalow and associated ecosystems when available (Fisher 1999). The accumulation of coarse woody debris is therefore as important as forest architecture for supporting the greatest possible range of fauna species. Moore, Russell & Coaldrake (1967) estimated the dead above-ground biomass of a mature Brigalow ecosystem in the same region at ∼130 t ha−1, but this is probably atypically high. The IBM suggested maximum dead above-ground biomass values of 36–44 t ha−1 in the less severe thinning treatments, but these values ignore the dead above-ground biomass present in the system before thinning (probably around 5 t ha−1). The stronger density effects on growth in the control treatment reduced the mean size of dying stems at each time point compared with thinning treatments (not shown). Thinning is therefore likely to generate better quality habitat in the form of large stags and logs sooner than in unthinned stands.

Thinning for carbon sequestration

According to the IBM 6000 stems ha−1 represents the lowest treatment stem density that will develop the maximum asymptotic living above-ground biomass value. This treatment also yielded impressive total above-ground biomass values. The net gain in above-ground biomass over the simulation period was highest in the 4000 and 6000 stems treatments, suggesting that restoration thinning in the early stages of stand development circumvents self thinning of larger stems later in development. Biomass accumulation in unthinned stands is also predicted to be substantial, reaching >90% of the above-ground living biomass capacity within 79 years of the last clearing event.

Inferring process from observed neighbourhood effects

The strong neighbourhood effect on growth is evidence that intraspecific competition in regrowth Brigalow is mainly size symmetric. Size symmetric competition occurs when stems consume resources in proportion to their size or less than in proportion to their size (Schwinning & Weiner 1998) and has mainly been demonstrated in competition for belowground resources (e.g. Cahill & Casper 2000). Physiological integration of connected stems may influence competitive interactions in clonal plant populations (de Kroon, Hara & Kwant 1992), but it was not possible to investigate this in the present study. The positive effect of available space on survival indicates that mortality is at least partially density dependent in Brigalow regrowth and is probably related to local depletion of plant available moisture in crowded neighbourhoods.

Predicting long-term stand dynamics from short-term stem responses

Multi-level models have considerable utility for parameterising IBM (Buckley, Briese & Rees 2003a, b). The varying quadrat coefficients from the growth and mortality models were incorporated to capture the spatial variation in stem–neighbourhood relationships that reflect unexplained sources of heterogeneity throughout the stand. The random stem effect incorporated in the growth component of the IBM represented the small differences in growth performance that were not explained by available space or the basal area of quadrats. This random stem variance was higher in the more severe treatments probably because a greater range of neighbourhoods were generated in these treatments (not shown).

The stand scale dynamics that emerged were based on growth and mortality observed during the first 2 years after thinning. The neighbourhood coefficients from the growth and mortality HBM were effectively kept constant throughout the simulation period except for minor variations included to capture uncertainty in these estimates. Thus, the predictive intervals generated by the IBM (refer to Figs S2–S4 in Supporting Information) represent the unexplained variation (among stems and among quadrats) during the monitoring period and do not incorporate variability in growth and survival because of climatic fluctuations over longer time periods. Climate related variability is likely to be substantial given the irregular rainfall experienced in the study region. Rainfall during the implementation of thinning and the following 2 years was close to average for the period. However, the temporal distribution of rainfall was favourable for growth, with considerable totals recorded during A. harpophylla’s major growth periods (spring, early summer and late summer; Johnson 1964). It is therefore possible that the observed growth response was atypically high. Similarly tree death is a cumulative process involving multiple interacting factors (Franklin, Shugart & Harmon 1987). Rainfall in the 2 years prior to the application of treatments was well below average which probably contributed to mortality of some stems in dense neighbourhoods and may have resulted in an over-estimated mortality rate. In addition, we did not detect an effect of stem size on the mortality rate, but mortality risk probably declines as stems are released from competition and become larger (Kariuki 2008). This size-related phenomenon was captured to some extent in the IBM which reduced mortality risk as available space increased. If mortality was over-estimated by the IBM, then the efficacy of thinning would have also been over-estimated, that is, optimal thinning densities may actually be lower than 6000 stems ha−1. However, in an ecosystem with such sporadic recruitment, a figure of 6000 stems ha−1 is both precautionary and is low enough to stimulate structural development. It is also more likely to maintain canopy cover at levels that minimize grass invasion and hence the risk of fire. Precaution is further warranted given climate change predictions of declining annual rainfall and increasing rainfall variability throughout the Brigalow Belt [CSIRO (Commonwealth Scientific and Industrial Research Organisation) and BOM (Australian Bureau of Meteorology) 2007].

Growth responses to thinning may not be so pronounced in Brigalow stands receiving less rainfall. Nevertheless, stem density has been identified as the major determinant of restoration and carbon sequestration potential in any given region (Dwyer, Fensham & Buckley 2010), so thinning should accelerate structural development in dense stands relative to unthinned stands in the same region. Thinning in younger or older stands is also likely to prove effective because the positive effect of available space on growth was the same for target stems of all sizes [i.e. there was limited evidence of an interaction between stem size and sqrt(available space)].

Recommendations for management at the stand scale

Restoration and carbon sequestration benefits may be achieved in dense Brigalow regrowth stands by thinning to 6000 stems ha−1, while ensuring that canopy cover is maintained at >50%. It may also be beneficial to retain buffers of dense stems along pasture boundaries to further minimize grass encroachment. Manual thinning should be undertaken only when soil moisture is high to reduce the risk of prolific secondary suckering (Johnson 1964). Random stem selection retains a range of sizes in spatial arrangements similar to mature Brigalow forests.

Ringbarking is a time and labour intensive thinning method but was chosen to avoid the risk of chemical transmission of herbicides to nontarget stems and in preference to felling (with a chainsaw) which would have increased the risk of prolific secondary suckering. Pellet herbicide has been successfully applied in restoration thinning experiments in shrub-steppe communities in North America (Olson & Whitson 2002), but research is required to determine safe and effective prescriptions for thinning Brigalow regrowth. Burning may be a low cost option but prescriptive application for selective thinning would be extremely difficult and grass fuel loads are generally low in dense regrowth stands.

The IBM predictions suggest that even without thinning, more than 90% of maximum potential total above-ground biomass can be accrued within 79 years of the last clearing attempt, but restoration of mature forest structure will take considerably longer in unthinned stands.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We thank the Fitzroy Basin Association Inc. for research funding and Graham Lightbody for his support. JMD was funded by an Australian postgraduate award and YMB by an Australian Research Council Australian Research Fellowship (DP077138). The IBM simulations were performed at the Vital-IT (http://www.vital-it.ch) Center for high-performance computing of the Swiss Institute of Bioinformatics with the able assistance of Fred Shutz. Thanks to the Grieve family for their cooperation and Phil and Leanne at SSAQLD for their exceptional hospitality. Thanks also to all field volunteers, Martin Ambrose, Mark Cant, Russel Fairfax, Don Butler, Clive McAlpine and Julien Pottier for assistance and advice and to Antoine Guisan for hosting JMD during the drafting of the manuscript. We also thank Peter Vesk, Elizabeth Crone and three anonymous reviewers for their valuable comments.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Allen, C.D., Savage, M., Falk, D.A., Suckling, K.F., Swetnam, T.W., Schulke, T., Stacey, P.B., Morgan, P., Hoffman, M. & Klingel, J.T. (2002) Ecological restoration of Southwestern ponderosa pine ecosystems: a broad perspective. Ecological Applications, 12, 14181433.
  • Allen, E.B., Violi, H.A., Allen, M.F. & Gomez-Pompa, A. (2003) Restoration of tropical seasonal forest in Quintana Roo. The Lowland Maya Area: Three Millennia at The Human-Wildland Interface (eds A.Gomez-Pompa, M.F.Allen, S.L.Fedick & J.Jimenez-Osornio), pp. 587598. Haworth Press, New York.
  • Bates, D. & Maechler, M. (2009) R Package: LME4 – Linear Mixed-Effects Models Using S4 Classes. Available at: http://www.r-project.org/, Maintained by D.Bates, Last updated July 2009, accessed 20 August 2009.
  • Bolker, B.M., Brooks, M.E., Clark, C.J., Geange, S.W., Poulsen, J.R., Stevens, M.H.H. & White, J.S.S. (2009) Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution, 24, 127135.
  • Brandl, H. (1992) The development of thinning in German forestry. Allgemeine Forst Und Jagdzeitung, 163, 6170.
  • Buckley, Y.M., Briese, D.T. & Rees, M. (2003a) Demography and management of the invasive plant species Hypericum perforatum. I. Using multi-level mixed-effects models for characterizing growth, survival and fecundity in a long-term data set. Journal of Applied Ecology, 40, 481493.
  • Buckley, Y.M., Briese, D.T. & Rees, M. (2003b) Demography and management of the invasive plant species Hypericum perforatum. II. Construction and use of an individual-based model to predict population dynamics and the effects of management strategies. Journal of Applied Ecology, 40, 494507.
  • Bureau of Meteorology (2009a) Climate Averages for Inglewood Tobacco Res (041341). Australian Government, Canberra.
  • Bureau of Meteorology (2009b) Monthly Rainfall Totals for Langley TM Station (41152). Australian Government, Canberra.
  • Butler, D.W. (2009) Planning iterative investment for landscape restoration: choice of biodiversity indicator makes a difference. Biological Conservation, 142, 22022216.
  • Butler, D.W. & Fairfax, R.J. (2003) Buffel grass and fire in a gidgee and brigalow woodland: a case study from central Queensland. Ecological Management & Restoration, 4, 120125.
  • Cahill, J.F. & Casper, B.B. (2000) Investigating the relationship between neighbor root biomass and belowground competition: field evidence for symmetric competition belowground. Oikos, 90, 311320.
  • Chandler, S.T., Buckley, Y.M. & Dwyer, J.M. (2007) Restoration potential of Brigalow regrowth: insights from a cross-sectional study in southern Queensland. Ecological Management and Restoration, 8, 218221.
  • Covington, W.W. & Moore, M.M. (1994) Southwestern ponderosa pine forest structure: changes since Euro-American settlement. Journal of Forestry, 92, 3947.
  • CSIRO (Commonwealth Scientific and Industrial Research Organisation) and BOM (Australian Bureau of Meteorology) (2007) Climate Change in Australia – Technical Report 2007. CSIRO, Canberra.
  • Dwyer, J.M., Fensham, R.J., Butler, D.W. & Buckley, Y.M. (2009) Carbon for conservation: assessing the potential for win-win investment in an extensive Australian regrowth ecosystem. Agriculture Ecosystems & Environment, 134, 17.
  • Dwyer, J.M., Fensham, R.J. & Buckley, Y.M. (2010) Agricultural legacy, climate and soil influence the restoration and carbon potential of an extensive woody regrowth ecosystem in Australia. Ecological Applications, in press.
  • Fisher, D.O. (1999) Behavioural Ecology and Demography of the Bridled Nailtail Wallaby, Onychogalea fraenata. PhD thesis. University of Queensland, Brisbane.
  • Franklin, J.F., Shugart, H.H. & Harmon, M.E. (1987) Tree death as an ecological process. BioScience, 37, 550556.
  • Hart, D.M. (1995) Litterfall and decomposition in the Pilliga State Forests, New South Wales, Australia. Austral Ecology, 20, 266272.
  • Houlder, D.J., Hutchinson, M.F., Nix, H.A. & McMahon, J.P. (2000) ANUCLIM 5.0. Centre for Resource and Environmental Studies, Australian National University, Canberra.
  • Johnson, R.W. (1964) Ecology and Control of Brigalow in Queensland. Queensland Department of Primary Industries, Brisbane.
  • Kariuki, M. (2008) Modelling the impacts of various thinning intensities on tree growth and survival in a mixed species eucalypt forest in central Gippsland, Victoria, Australia. Forest Ecology and Management, 256, 20072017.
  • Kaye, J.P. & Hart, S.C. (1998) Ecological restoration alters nitrogen transformations in a ponderosa pine bunchgrass ecosystem. Ecological Applications, 8, 10521060.
  • De Kroon, H., Hara, T. & Kwant, R. (1992) Size hierarchies of shoots and clones in clonal herb monocultures - Do clonal and nonclonal plants compete differently. Oikos, 63, 410419.
  • Laughlin, D.C., Bakker, J.D., Daniels, M.L., Moore, M.M., Casey, C.A. & Springer, J.D. (2008) Restoring plant species diversity and community composition in a ponderosa pine-bunchgrass ecosystem. Plant Ecology, 197, 139151.
  • Mackensen, J., Bauhus, J. & Webber, E. (2003) Decomposition rates of coarse woody debris; A review with particular emphasis on Australian tree species. Australian Journal of Botany, 51, 2737.
  • Mead, R. (1966) A relationship between individual plant spacing and yield. Annals of Botany, 30, 301309.
  • Moore, A.W., Russell, J.S. & Coaldrake, J.E. (1967) Dry matter and nutrient content of a subtropical semiarid forest of Acacia harpophylla F Muell (brigalow). Australian Journal of Botany, 15, 1124.
  • Nelson, C.R., Halpern, C.B. & Agee, J.K. (2008) Thinning and burning result in low-level invasion by nonnative plants but neutral effects on natives. Ecological Applications, 18, 762770.
  • Olson, R.A. & Whitson, T.D. (2002) Restoring structure in late-successional sagebrush communities by thinning with tebuthiuron. Restoration Ecology, 10, 146155.
  • Pearson, H.L. & Vitousek, P.M. (2001) Stand dynamics, nitrogen accumulation, and symbiotic nitrogen fixation in regenerating stands of Acacia koa. Ecological Applications, 11, 13811394.
  • Rentch, J.S., Schuler, T.M., Ford, W.M. & Nowacki, G.J. (2007) Red spruce stand dynamics, simulations, and restoration opportunities in the central Appalachians. Restoration Ecology, 15, 440452.
  • Richardson, R. (2006) Queensland Brigalow Belt Reptile Recovery Plan 2008–2012. Report to the Department of the Environment, Water, Heritage and the Arts, Canberra. WWF Australia, Brisbane.
  • Scanlan, J.C. (1991) Woody overstorey and herbaceous understorey biomass in Acacia harpophylla (brigalow) woodlands. Australian Journal of Ecology, 16, 521530.
  • Schwinning, S. & Weiner, J. (1998) Mechanisms determining the degree of size asymmetry in competition among plants. Oecologia, 113, 447455.
  • Spieglhalter, D.J., Best, N.G., Carlin, B.P. & Van der Linde, A. (2002) Bayesian Measures of Model Complexity and Fit (with Discussion). Journal of the Royal Statistical, Series B, 64, 583616.
  • Stone, J.E., Kolb, T.E. & Covington, W.J. (1999) Effects of restoration thinning on presettlement Pinus ponderosa in northern Arizona. Restoration Ecology, 7, 172182.
  • The R Foundation for Statistical Computing (2009) The R Environment for Statistical Computing and Graphics, Version 2.9.0. The R Foundation for Statistical Computing, Vienna, Austria, Available at: http://www.r-project.org/, accessed 20 January 2010.
  • Vargas, R., Allen, E.B. & Allen, M.F. (2009) Effects of vegetation thinning on above- and belowground carbon in a seasonally dry tropical forest in Mexico. Biotropica, 41, 302311.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Methods S1. Detailed description of the HBMs and the IBM.

Figure S1. Plot of the 70% stem asymptote function with three scenarios of available space.

Figure S2. Predicted median trajectories from 0 to 50 years after thinning for living and dead above-ground biomass accumulation.

Figure S3. Predicted median trajectories from 0 to 50 years after thinning for total aboveground biomass accumulation and stem density decline.

Figure S4. Predicted median circumference (cm) size-class distributions for each treatment at 2, 24 and 50 years after thinning.

Table S1. Mean (SD, n = 4) structural and floristic characteristics of the adjacent mature reference forest.

Table S2. Model DIC values for each combination of stem- and quadrat-level explanatory variables considered for the growth and mortality HBMs.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

FilenameFormatSizeDescription
JPE_1775_sm_Supmat.pdf615KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.