JOBNAME: No Job Name PAGE: 1 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
Does stand structure influence susceptibility of eucalyptfloodplain forests to dieback?aec_2043 1..11
SHAUN C. CUNNINGHAM,1* J. R. THOMSON,1 J. READ,2 P. J. BAKER1 ANDR. MAC NALLY1
1Australian Centre for Biodiversity, School of Biological Sciences, Monash University, Melbourne,Vic.3800, Australia (Email: [email protected]), and 2School of Biological Sciences,Monash University, Melbourne,Victoria, Australia
Abstract Forest dieback is a worldwide problem that is likely to increase with climate change and increasinghuman demands for resources. Eucalyptus camaldulensis forests are an acute example of forest dieback, with 70% ofthe Victorian Murray River floodplain in some state of dieback. If we are to halt dieback in these floodplain forests,we need to understand what makes stands susceptible to dieback. Forest diebacks are often related to standstructure, with dieback more severe in senescent or high-density stands. We determined whether certain standstructures make these forests more susceptible to dieback.We undertook an extensive survey of 176 stands across100 000 ha of forest, covering the range of stand condition on this floodplain. Large and small trees (20-, 40-, 80-and 120-cm diameter) showed a similar reduction in the probability of being alive with decreasing stand condition.A slight improvement in stand condition was found at higher densities and basal areas, which may reflect the higherproductivity or younger age of these stands. Stand condition was moderately, positively correlated with longitude,with stand condition being higher in the east of the Murray River floodplain where flooding frequencies arecurrently higher. This suggests that dieback of these floodplain forests would be more effectively mitigated byincreased water availability through flooding than by altering stand structure.
Extensive dieback has been reported in many foresttypes since the 1960s (Huettl & Mueller-Dombois1993). Forest dieback is characterized by a progressivereduction in the crowns of individual trees that leads towidespread mortality. Proposed causes of forestdieback include habitat fragmentation (e.g. Landsberget al. 1990), climate change (e.g. Bourque et al. 2005),impacts of introduced animals (e.g. Close et al. 2008),succession (e.g. Mueller-Dombois 2006) and air pol-lution (e.g. McLaughlin & Percy 1999). Forest diebackaffects the quantity and quality of forest and waterresources, the flora and fauna dependent on theseecosystems, and associated terrestrial and aquatic eco-systems (Ballinger & Lake 2006). If we are to mitigatethis dieback under increasing pressure from land-useand climate change, we need to understand whatmakes forests susceptible to dieback.
Riparian forests are an acute example of howresource use and climate change affect forests. Ripar-ian systems throughout the world are used as watersupplies for agricultural, industrial and human con-sumption through pumping, diversions and dams.
Increased regulation of rivers over the last century hasled to dieback of many floodplain forests, particularlyin arid regions (e.g. Busch & Smith 1995). Dieback offloodplain forests is likely to increase in extent underpredicted future climates due to decreasing precipita-tion in many regions and increasing human demandsfor water (IPCC 2007).
The Murray River, in south-eastern Australia, is thecountry’s longest river (2520 km), providing water forextensive irrigation farming, domestic consumptionand power generation. Eucalyptus camaldulensisDehnh. (river red gum) is the dominant floodplain treealong the Murray River and has the widest naturaldistribution of all eucalypts, occupying watercoursesthroughout mainland Australia (Boland et al. 1984).Increasing regulation of the Murray River since the1920s has significantly reduced peak flows, andreduced frequency (35–62% of historical) and dura-tion (40–84% of historical) of extensive floods thatconnect anabranches to the main river (MDBC2005a). Substantial dieback of E. camaldulensis forestshas been observed in the lower Murray River over thepast 20 years (Margules et al. 1990; MDBC 2005b). Aquantitative assessment of the extent of this diebackusing rigorous ground surveys, remote sensing andmodelling found that 70% of E. camaldulensis forestson theVictorian Murray River floodplain were in some
*Corresponding author.Accepted for publication June 2009.
JOBNAME: No Job Name PAGE: 2 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
state of dieback (Cunningham et al. 2009). There wasa downstream deterioration in stand condition, whichis associated with more extreme reductions in floodingfrequency (54% of historical), due to water harvesting,and the drier climate (270 mm year-1) in the LowerMurray compared with the Upper Murray region(38% of historical frequency, 715 mm year-1, MDBC2005a; BOM 2007).
In the future, dieback of E. camaldulensis is likely toexpand upstream along the Murray River as wateravailability decreases in south-eastern Australia dueto rising temperatures, decreasing annual rainfall(CSIRO & Australian Bureau of Meteorology 2007)and the increasing severity of droughts (Nicholls2004).We need management strategies that will bufferthese forests against the effects of decreasing wateravailability. Dieback of E. camaldulensis is patchywithin the landscape, with stands of good conditionadjacent to areas of severe dieback. If we are to reducedieback in these floodplain forests, we need to under-stand what makes some stands less susceptible todieback than others.
Forest diebacks are often related to stand structure,with dieback more severe in senescent (Sprugel 1976)or high-density stands (Greenwood &Weisberg 2008).Stands of E. camaldulensis may be very dense due tomass recruitment after a sequence of advantageousfloods, and are now self-thinning to a sustainabledensity. Alternatively, there may have been no substan-tial recruitment events under reduced flooding andmature trees may be dying due to senescence. Largetrees may be less affected by dieback than small treesbecause they possess deeper root systems that canaccess groundwater during dry periods (Mensforthet al. 1994). However, dieback may be unrelated to thedemography of stands, with both large and small treesdying.
Here, we investigated whether the structure of E.camaldulensis stands influences their susceptibility todieback under dry conditions and, therefore, if man-agement of stand structure could be used to mitigatedieback under continuing reduced water availability.We determined: (i) if tree mortality within stands wassize-dependent; and (ii) whether structural character-istics of stands were associated with increased suscep-tibility to dieback.
METHODS
Study area
The study area included all stands of E. camaldulensison the floodplains of the Murray River from the HumeDam (36°06′S, 147°01′E) to the South Australianborder (34°01′S, 141°00′E), the lower Ovens River
downstream of Wangaratta and the lower GoulburnRiver downstream of Shepparton in Victoria, Australia(103 550 ha of forest, Fig. 1). This included openforests (10–30 m tall, 30–45% projective foliage cover)and woodlands (10–30 m tall, 20–25% projectivefoliage cover (Specht 1981)) with shrubby, sedgyand grassy understoreys (Margules et al. 1990).The climate across this area is temperate and has anorth-westerly increase in mean annual maximumtemperatures (22.1–24.5°C), mean annual minimumtemperatures (8.9–15.2°C) and annual evaporation(120–1790 mm year-1), and a decrease in annual pre-cipitation (270–715 mm year-1, BOM 2007).
Site selection
A stratified-random approach was used to select sitescovering a range of stand conditions. First, sites wererestricted to public land, which still provided anadequate coverage of the study area, to minimizeaccess problems. Second, the area was divided into fivebioregions (Victorian Riverina, Upper Murray Fans,Lower Murray Fans, Robinvale Plains and MurrayScroll Belt), which are defined by differences inclimate, geomorphology, lithology and biodiversity(Environment Australia 2000). This ensured spatialcoverage of the extensive survey area and an ability tostandardize data across the large environmental gradi-ent (e.g. rainfall 270–715 mm year-1). Third, selectionwas restricted within each bioregion to the predomi-nant E. camaldulensis communities that covered thelargest area. Communities were added sequentially,starting with the largest, until �65% of the total forestarea within the bioregion was included. Theseincluded both riverine and floodplain communities,ensuring that stands covering a range of flooding fre-quencies, and potentially stand condition, wereselected. Finally, sites were selected within each com-munity according to a classification of reflectancevalues from Landsat 7 imagery from 2004 to ensure arange of reflectance values and potentially a range ofstand condition were included. Reflectance values (sixreflectance bands of Landsat7 data) were classifiedwithin each bioregion using principal componentsanalysis of 10 000 randomly selected pixels.
Stand assessment
We surveyed 176 stands of E. camaldulensis betweenJune 2006 and April 2007. A 0.25-ha plot was estab-lished at each point location. Most plots were50 ¥ 50 m but rectangular plots (width = 20–40 m)were used to assess narrower stands alongwatercourses. At each plot, current stand conditionand past stand structure were estimated as follows.
JOBNAME: No Job Name PAGE: 3 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
Plant area index, crown vigour and percentage livebasal area were previously found to be reliable andobjective indicators of condition in E. camaldulensisstands (Cunningham et al. 2007). Plant area indexis the area of leaves and stems per unit groundarea without adjustment for clumping of canopycomponents. Plant area index was estimated fromhemispherical photographs of the canopy, which werefirst classified using image analysis (MultiSpec Appli-cation Version 3.1, Purdue University, Indiana), withthe program Winphot 5.00 (ter Steege 1996). Crownvigour is the percentage of the potential crown, whichis determined by the extent of the existing branchingstructure, that contains foliage. Crown vigour was esti-mated using an interval scale (0%, 1–20%, 21–40%,41–60%, 61–80%, 81–100%) from 30 trees represen-tative of the range of tree size and condition within astand. Percentage live basal area is the percentage of astand’s basal area that is contributed by live trees.Plant area index was standardized relative to biore-gional maxima to detrend the natural downstreamreduction in plant area index owing to the reduction inproductivity associated with reduced rainfall andincreased evaporation. A stand condition score (SCS)was calculated from the three condition indicators,with each indicator contributing up to 5 points to atotal score of 15 points.
It was possible to estimate the structure of standsprior to dieback because dead E. camaldulensis trees
remain upright for more than a decade due to the highdensity of the wood. We measured diameter at breastheight (DBH) and status (live/dead) of every treewithin each plot. Tree deaths that could be attributedto causes other than dieback (e.g. ring barking) wereexcluded from the analysis. Trees were considereddead when they had no leaves in their crown. Wecalculated the pre-dieback structure (mean DBH,basal area and tree density) and size distribution ofstands by including both live and dead trees.
Size distributions within a plot were modelled usinga Weibull distribution, with shape (k) and scale (l)parameters (Bailey & Dell 1973). These parameterscan describe a wide range of size distributions, includ-ing distributions that have a positive skew (1 < k <3.6), are approximately normal (k = 3.6) or have anegative skew (k > 3.6). We estimated the shape andscale parameters for each stand independently usingBayesian updating through Markov Chain MonteCarlo (MCMC) inWinBUGS software (Spielgelhalteret al. 2003).We gave k and l uninformative prior prob-ability distributions using a Gamma (0.001, 0.001)distribution.
Modelling
Models were used to determine whether susceptibilityto dieback was related to tree size (model 1) or stand
WangarattaShepparton
HumeDam
50 km
N
34 So
147 Eo
37 So
SO
UT
H A
US
TR
ALIA
VIC
TO
RIA
NEW SOUTH WALES
141 Eo
increasingaridity
Fig. 1. Distribution of Eucalyptus camaldulensis forests on the Murray River, Lower Ovens River and Lower Goulburn Riverfloodplains in Victoria, Australia.
JOBNAME: No Job Name PAGE: 4 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
structure (model 2). Spatial variables (latitude andlongitude) were included in the models to account forthe known downstream decrease in stand condition(Cunningham et al. 2009). In the absence of accurateflooding history data for the individual stands, thedistance to nearest water body (DISTWB) was used asa surrogate for local flooding frequency. Water bodiesincluded permanent and semi-permanent creeks,rivers and wetlands, which are indicative of frequentlyflooded areas.
Tree mortality model
A hierarchical Bayes model (Congdon 2003) was usedto investigate the relationships between tree mortality,stand condition and stem diameter. The full modelwas:
Yi~Bernoulli(pi)
logp
pDBHs i s i i
1−( )⎛⎝⎜
⎞⎠⎟
= +( ) ( )α β
α α α α εs s s sCOND LONG= + + +0 1 2 1Model
β β β β ϕs s s sCOND LONG= + + +0 1 2
Here the probability that an individual tree i in stand sis alive is modelled as a stand-specific, logistic functionof DBHi. The intercept and slope (on the log-oddsscale) for each stand are modelled as linear functionsof stand condition (COND), longitude (LONG) andsite-level random errors, ei and ji.
We used Bayesian model selection to evaluate theevidence that COND and LONG influence the inter-cept and slopes (i.e. should be included in eachsub-model), and to produce model averaged coeffi-cient estimates and model predictions. Bayesian modelselection ranks models according to posterior modelprobabilities, which combine marginal likelihoods(those integrated across prior distributions forparameter values) with independent prior modelprobabilities. Posterior model probabilities can beused to weight parameter estimates and predictionsfrom multiple models to obtained model averaged esti-mates that account for model uncertainty (Wintleet al. 2003). Summing the posterior model probabili-ties of all models that include a specific parameter (e.g.b1) yields the posterior probability that the parameteris non-zero, Pr(b1 � 0|Data), or the probability thatthe corresponding variable is included in the bestmodel.
The linear sub-models for the intercept and slopewere fitted as variable selection models using reversiblejump MCMC (Lunn et al. 2008) with prior probabil-ity of inclusion for each variable equal to 0.5, meaningthat all 24 possible model structures were equally prob-
able a priori (all possible combinations of COND andLONG in one or both sub-models). Posterior modelprobabilities therefore reflect relative marginal likeli-hoods (evidence in the data), which automaticallypenalize model complexity (Kass & Raftery 1995).
We obtained model averaged parameter estimatesand predictions of survival rates for a range of treesizes (20-, 40-, 80- and 120-cm diameter) along con-dition and longitudinal gradients. We also calculatedthe posterior probabilities of non-zero parameters(variable inclusion) and corresponding odds ratios,OR:
ORData
Dataβ β
ββ
β11
1
1
1
01 0
1 00
( ) = ≠( )− ≠( )
× − ≠( )≠( )
PrPr
PrPr
Odds ratios measure the proportional change fromthe prior odds to the posterior odds, and OR > 3 isgenerally considered ‘substantial’ evidence in favour ofone hypothesis over an alternative (Jeffreys 1961).We used vague normal prior distributions for allparameters in model 1, as follows: a0~N(0, 10 000),a1~N(0, σα
2), a2~N(0, σα2), b0~N(0, 10 000), b1~
N(0, σβ2), b2~N(0, σβ
2), ei~N(0, σε2) and ji~N(0, σψ
2).Weused uniform prior distributions on the interval (0, 10)for the corresponding standard deviations sa, sb, se,and sj.
Stand structure model
Bayesian model selection with piecewise linear splineswas used to examine relationships between COND andthree structural variables: Weibull shape (SHAPE),density (DENS) and basal area (BASAL). The modelalso included latitude (LAT), LONG and DISTWB,and allowed interactions between the spatial variablesand each structural variable. We did not includeWeibull scale or mean DBH as candidate predictorsbecause these variables were strongly correlated (non-linearly, R2 > 0.8) with Weibull shape and tree density,respectively. The full model was:
COND Ni i∼ μ σε, 2( )
μ α β εi ji jij
ix= + +=
∑1
6
where β α φ χ θ
γ
ji jl ji jll
k
jm i jmm
k
jn
I x I LONG
I D
j j
= >( ) + −( )
+
= =∑ ∑
1 1
0 1
IISTWBi jnn
k j
−( )=
∑ ϕ1
2
,
(2)
and I (a > b) = 1 if a > b and 0 otherwise.In model 2, the linear coefficient for variable j, bji, is
potentially a (step) function of the values of xi, LONGand DISTWB. The vectors fj, qj and jj give the knot
JOBNAME: No Job Name PAGE: 5 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
positions, or values of xj, LONG and DISTWB atwhich the linear relationship between condition andvariable xj changes. The numbers (k0j, k1j and k2j) andpositions of knots were treated as unknown parameters(i.e. ‘free knot’ splines), except for fj1, which was fixedat the minimum value of xj.The dimension parametersk0j, k1j and k2j determine the complexity of the fittedrelationships and interactions, and if knj = 0 the corre-sponding term is not included in the model. We setsome k1j and k2j values to zero to remove redundantterms (e.g. LONG ¥ LONG interaction). The remain-ing knj were assigned prior distributions that includedzero; k0j~Binomial(3, 0.3), k1j~Binomial(3, 0.15),k2j~Binomial(3, 0.15). We used the posterior distribu-tions of the dimension parameters to identify terms forwhich there was strong support in the data. Specifi-cally, we calculated the posterior probability that eachterm was included in the model, Pr(knj > 0|Data), andthe corresponding odds ratio.
We fitted a final model that included only terms inmodel 2 with OR > 3 (plus the independent term forany variable included in an interaction term).The finalmodel was:
COND Ni i∼ μ σε, 2( )
μ α β φ β φ
β
i l i ll
k
l i ll
k
l
BASAL LONG
DENSI
= + −( ) + −( )
+
= =∑ ∑1 1
12 2
1
3
01 02
TTY
DENSITY LONG
i ll
k
i m i mm
k
i
−( )
+ −( ) +
=
=
∑
∑
φ
χ θ ε
31
1
03
13
(3)
We used exchangeable normal prior distributions foreach set of linear coefficients, and assigned uniformprior distributions to the corresponding standarddeviations. For example, the prior distributions for theas in model 2 were defined as: ajl~Normal(0, sj
a);sj
a~Uniform(0, 10), j = 1, 2, . . . , 6. We used inverseGamma priors (0.001, 0.001) for the residual varianceparameters σε
2 in models 2 and 3.
We fitted all models in Winbugs, version 1.4(Spielgelhalter et al. 2003) with the reversible jumpMCMC add-on (Lunn et al. 2008). All predictorvariables were standardized (mean 0, SD 1) prior tomodel fitting. Parameters were estimated from threeMCMC chains of 100 000 iterations after 50 000iteration burn-in periods. Mixing and convergence ofMCMC were verified by Gelman–Ruben–Brooks sta-tistics, and by visual inspection of chain histories andautocorrelation plots. We fitted all models with arange of upper limits on the standard deviations ofexchangeable normal priors and obtained consistentresults.
Examination of semi-variograms and Moran’s I sta-tistics confirmed the absence of spatial structure in theresiduals from models 1, 2 and 3.
RESULTS
The 176 stands surveyed covered a representativerange of stand conditions and structures along theVictorian Murray River floodplain. Stand conditionranged from completely dead (SCS = 0.5) to healthyforest (SCS = 15.0). The structure of these forestsvaried widely from very low to high basal area(3–110 m2 ha-1), sparse to dense (16–756 trees perhectare), positively through to negatively skewed sizedistributions (k = 0.9–10.3) and by being dominatedby small to large trees (mean DBH = 17–145 cm).
The probability of a tree being alive increased withstand condition (a1) and with tree size (b0, Table 1).There was also evidence that the probability of beingalive increased more rapidly with increasing DBH asstand conditioned increased (b1, Table 1). TheCOND ¥ DBH interaction (b1) was relatively weak, asshown in a comparison of predicted survival probabili-ties for small (DBH = 20 cm) and large trees(DBH = 120 cm, Fig. 2). Small trees had a slightlylower probability (95% credible interval (CI) = 0.87–0.91) of being alive than large trees (CI = 0.95–0.98)
Table 1. Summary of posterior distributions for parameters in the tree mortality model
Pr(� 0) is the posterior probability that the parameter is non-zero. se and sj are the standard deviations of the site-levelrandom intercepts es and slopes js, respectively. CI, credible interval; DBH, diameter at breast height.
JOBNAME: No Job Name PAGE: 6 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
in stands with moderate dieback (SCS ~ 10). In con-trast, all tree diameters had an equally low probabilityof being alive (CI = 0.3–0.6) in degraded stands(SCS ~ 5).
The model selection procedure suggested that lon-gitude, basal area and the interaction between densityand longitude were highly probable (Pr � 0.9) predic-tors of stand condition, whereas the remaining vari-ables and interactions with indicators of floodinghistory (longitude and distance to nearest water body)were not (Table 2). The model that included only theselected variables and interactions (model 3) was ableto explain 33.6% of the variation in stand condition inthe survey data. The model showed that stand condi-tion improved with increasing longitude, basal area
and density (Fig. 3). Longitude had a moderate posi-tive relationship (coefficient of partial determination(CPD) = 0.31), with stand condition being higher inthe upper Murray River floodplain. Basal area hada weak positive relationship with stand condition(CPD = 0.06), which was the strongest at intermediatevalues (10–50 m2 ha-1).There was also a weak positiverelationship (CPD = 0.08) between stand conditionand density, which was steeper for stands at lowlongitudes.
DISCUSSION
Regional dieback of trees has increased in many forestsof the world in recent decades (e.g. Breshears et al.2005). Forests of E. camaldulensis on the floodplains ofthe Murray River in south-eastern Australia are a starkexample of how the interaction of human resource use(river regulation and water extraction) and climatechange (rising temperatures and decreasing rainfall)are affecting forests. Currently, 70% of E. camaldulensisstands on the Victorian Murray River floodplain havesome level of dieback, with a downstream deteriora-tion in stand condition towards semi-arid regions(Cunningham et al. 2009). Dieback of E. camaldulensisis variable within a forest, suggesting differences insusceptibility among stands.
The size of trees most affected by dieback differsamong the types of forest dieback.The reduced vigourof senescent trees is often suggested to make themmore susceptible to dieback through a reduced abilityto replace lost biomass and a lowered capacity toproduce defence compounds (Doak 2004). Senes-cence dieback, such as ‘wave-regeneration’ of Abiesbalsamea (Sprugel 1976), is a natural process thataffects even-aged stands when trees are large andsenescent. Insect outbreaks tend to cause higher mor-tality in large trees (e.g. Bergeron et al. 1995; De Som-viele et al. 2004), although some insects preferentiallyattack small, suppressed trees (Cedervind & Lång-ström 2003). Drought-related dieback can causehigher mortality in small, suppressed trees becauselarge trees have access to groundwater (Bauce & Allen1991; Lloret et al. 2004), but can alternatively causehigher mortality in large trees due to their reducedvigour (Worrall et al. 2008) or it can affect all treesequally, suggesting a physiological threshold has beenreached (Fensham & Holman 1999; MacGregor &O’Connor 2002). Our extensive survey of E. camaldu-lensis stands found that different-sized trees had similarprobabilities of being alive as stand condition deterio-rates (Fig. 2). Small trees (DBH = 20 cm) had aslightly lower probability of being alive than large trees(DBH = 120 cm) in good-condition stands, which isconsistent with self-thinning of suppressed trees withstand development. This provides little evidence for
0 5 10 15
0.5
Pr(
Aliv
e)
Stand condition score
1.0
20 cm
120 cm
Fig. 2. Probability of Eucalyptus camaldulensis trees of dif-ferent diameters being alive in stands of different condition.Curves are predictions of logistic regressions (n = 176stands) with 95% credible intervals indicated by dashedlines. See Methods for calculation of stand condition score.
Table 2. Posterior probabilities from Bayesian modelselection with cubic regression splines for main effects andtheir interactions
JOBNAME: No Job Name PAGE: 7 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
size-related dieback of E. camaldulensis and suggeststhat trees of all sizes are affected by the water stressassociated with this dieback. However, the cause ofwater stress may differ among different-sized trees,with small trees responding to low surface soil mois-ture while large trees are affected by lower availabilityof groundwater.
Forest dieback is often more severe in high-densitystands due to higher levels of competition among treesfor resources. For example, dieback of eucalypt savan-nas was found to be related to lower soil moisture inhigh-density stands (Fensham & Fairfax 2007). Insectoutbreaks are also more common in high-densitystands due to the higher abundance of suppressedtrees and easier dispersal among trees (Greenwood &Weisberg 2008). In contrast, we found that dieback ofE. camaldulensis was negatively related to stand densityand basal area (Fig. 3). The slight improvement instand condition at higher densities and basal areas mayreflect the higher productivity (e.g. water availability)of these sites or the younger age of high-density stands.
Reducing stand density is an effective tool for man-aging some types of forest dieback (e.g. Bauce & Allen1991).The weak positive relationship between densityand basal area and the condition of E. camaldulensisstands (Fig. 3) suggests that active thinning maynot reduce the susceptibility of stands to dieback.Although the relationship between stand conditionand density changed with longitude, which is associ-ated with changes in flooding frequency, it remainedpositive across the floodplain. The lack of other inter-actions between stand structural variables and indica-tors of flooding history (longitude and distance tonearest water body) suggests a negative effect ofdensity is not masked by differences in flooding fre-quency among stands. In contrast, a four-decadeplanting trial of E. camaldulensis in the middle Murrayshowed higher mortality in high-density stands (morethan 1000 trees per hectare) during the drought con-ditions of the last decade (Horner et al. 2009).
However, our extensive survey only found stands ofless than 800 trees per hectare, suggesting high-densitystands are rare along theVictorian Murray River prob-ably due to long-term logging practices and the scar-city of the flooding events necessary to establish densestands.
Longitude had a moderate positive correlation withstand condition, with condition improving in theupstream, easterly direction (Fig. 3c). The predomi-nant abiotic change across this gradient is the largerrelative reduction in flooding frequency in the LowerMurray (from 76 to 35 years per century) comparedwith the Middle Murray (from 92 to 57 years percentury, MDBC 2005a) due predominantly toincreasing regulation of the Murray River. Thisdecrease in water availability has been exacerbated bydecreases in rainfall, with falls since 2001 being thelowest on record for the floodplain (Cai & Cowan2008). The dieback observed across the floodplain islikely to be due to a combination of reduced floodingfrequency over decades, the recent drought and anincreased reliance on groundwater.
The patchy nature of dieback within these forests isalso likely to be due to differences in water availability(Bacon et al. 1993) and not to stand structure per se.The variation in stand condition not explained by themodel (66%) may be due to small scale (<1 km) varia-tion in water availability. Differences in water availabil-ity within a forest could be caused by minordifferences in topography affecting the distribution offlood waters, different water holding capacities of soiltypes and differences in the depth and salinity ofgroundwater.
In conclusion, our study provided little evidence thatstand structure influences susceptibility to dieback inE. camaldulensis forests along the Victorian MurrayRiver floodplain.The probability of a tree being alive asstand condition deteriorates was independent of treesize. At the stand level, the only structural character-istics to show a relationship with dieback were stand
143140 146
Longitude ( E)o
0
5
10
15
Sta
nd c
onditio
n
a) CPD = 0.914
0 40 80 120
Basal area (m ha )2 -1
c) CPD = 0.168
200 400 600 800
Density (trees ha )-1
b) CPD = 0.231
U
L
Fig. 3. Relationship between stand condition predicted by model 3 (n = 176 stands) and (a) longitude, (b) density and (c)basal area. Dashed lines indicate 95% credible intervals. The proportional reductions in variance when a variable was added tothe overall model, coefficients of partial determination (CPD), are given. The different relationships between density and standcondition in the Lower (L) and Upper (U) Murray are presented. See Methods for calculation of stand condition score.
JOBNAME: No Job Name PAGE: 8 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
density and basal area. However, these were negativerelationships, with stand condition improving withhigher tree density and basal area. This suggests thataltering the structure of these stands (e.g. thinning)will not halt dieback under continued reduced wateravailability – infrequent flooding, below-average rain-fall, altered seasonality of rainfall, declining ground-water depth and increasing groundwater salinity.Therefore, dieback of floodplain forests, such as thoseon the Murray River floodplain, may only be mitigatedby increasing the frequency of floods.
ACKNOWLEDGEMENTS
This research was funded by an Australian ResearchCouncil Linkage grant (LP0560518), which was par-tially funded by the Victorian Department of Sustain-ability and Environment (DSE) and four CatchmentManagement Authorities (Mallee CMA, NorthCentral CMA, Goulburn-Broken CMA and NorthEast CMA). We thank Rachael Nolan for assistancewith fieldwork. This is publication No. 177 from theAustralian Centre for Biodiversity, Monash University.
REFERENCES
Bacon P. E., Stone C., Binns D. L., Leslie D. J. & Edwards D.W.(1993) Relationships between water availability and Euca-lyptus camaldulensis growth in a riparian forest. J. Hydrol.150, 541–61.
Bailey R. L. & Dell T. R. (1973) Quantifying diameter distribu-tions with the Weibull function. For. Sci. 19, 97–104.
Ballinger A. & Lake P. S. (2006) Energy and nutrient fluxes fromrivers and streams into terrestrial food webs. Mar.FreshwaterRes. 57, 15–28.
Bauce E. & Allen D. C. (1991) Etiology of a sugar maple decline.Can. J. For. Res. 21, 686–93.
Bergeron Y., Leduc A., Morin H. & Joyal C. (1995) Balsam firmortality following the last spruce budworm outbreak innorthwestern Quebec. Can. J. For. Res. 25, 1375–84.
Boland D. J., Brooker M. I. H., Chippendale G. M. et al. (1984)Forest Trees of Australia. CSIRO Publications, Melbourne.
BOM (2007) Climate data online. Australian Government Bureauof Meteorology. [Cited •• •• ••.] Available from URL: http://www.bom.gov.au/climate/averages/
Bourque C. P. A., Cox R. M., Allen D. J., Arp P. A. & Meng F. R.(2005) Spatial extent of winter thaw events in eastern NorthAmerica: historical weather records in relation to yellowbirch decline. Global Change Biol. 11, 1477–92.
Breshears D. D., Cobb N. S., Rich P. M. et al. (2005) Regionalvegetation die-off in response to global-change-typedrought. Proc. Natl Acad. Sci. USA 102, 15144–8.
Busch D. E. & Smith S. D. (1995) Mechanisms associated withdecline of woody species in riparian ecosystems of thesouthwestern United States. Ecol. Monogr. 65, 347–70.
Cai W. & Cowan T. (2008) Evidence of impacts from risingtemperature on inflows to the Murray-Darling Basin.Geophys. Res. Lett. 35, L07701.
Cedervind J. & Långström B. (2003) Tree mortality, foliagerecovery and top-kill in stands of Scots pine (Pinus sylvestris)subsequent to defoliation by the pine looper (Bupaluspiniaria). Scand. J. For. Res. 18, 505–13.
Congdon P. (2003) Applied Bayesian Modelling. John Wiley &Sons, New York.
CSIRO & Australian Bureau of Meteorology (2007) ClimateChange in Australia. CSIRO, Melbourne.
Cunningham S. C., Read J., Baker P. J. & Mac Nally R. (2007)Quantitative assessment of stand condition and its relation-ship to physiological stress in stands of Eucalyptus camaldu-lensis (Myrtaceae) in southeastern Australia. Aust. J. Bot. 55,692–9.
Cunningham S. C., Mac Nally R., Read J. et al. (2009) A robusttechnique for mapping vegetation condition across a majorriver system. Ecosystems 12, 207–19.
Davidson N. J., Close D. C., Battaglia M. et al. (2007) Eucalypthealth and agricultural land management within bushlandremnants in the Midlands of Tasmania, Australia. Biol.Conserv. 139, 439–46.
De Somviele B., Lyytikainen-Saarenmaa P. & Niemela P. (2004)Sawfly (Hym., Diprionidae) outbreaks on Scots pine: effectof stand structure, site quality and relative tree position ondefoliation intensity. For. Ecol. Manage. 194, 305–17.
Doak P. (2004) The impact of tree and stand characteristics onspruce beetle (Coleoptera: Scolytidae) induced mortality ofwhite spruce in the Copper River Basin, Alaska. Can. J. For.Res. 34, 810–16.
Durrleman S. & Simon R. (1989) Flexible regression modelswith cubic splines. Stat. Med. 8, 551–61.
Environment Australia (2000) Revision of the Interim Biogeo-graphic Regionalisation of Australia (IBRA) and the Develop-ment of Version 5.1 – Summary Report. Department ofEnvironment and Heritage, Canberra.
Fensham R. J. & Fairfax R. J. (2007) Drought-related tree deathof savanna eucalypts: species susceptibility, soil conditionsand root architecture. J.Veg. Sci. 18, 71–80.
Fensham R. J. & Holman J. E. (1999) Temporal and spatialpatterns in drought-related tree dieback in Australiansavanna. J. Appl. Ecol. 36, 1035–50.
Greenwood D. L. & Weisberg P. J. (2008) Density-dependenttree mortality in pinyon-juniper woodlands. For. Ecol.Manage. 255, 2129–37.
Horner G. J., Baker P. J., Mac Nally R., Cunningham S. C. &Thomson J. R. (2009) Mortality of floodplain forests sub-jected to a drying climate and water extraction. GlobalChange Biol. 15, ••–••.
Huettl R. F. & Mueller-Dombois D. (1993) Forest Decline in theAtlantic and Pacific Region. Springer-Verlag, Berlin.
IPCC (2007) Intergovernmental Panel on Climate Change FourthAssessment Report, Climate Change 2007: Synthesis Report.IPCC Secretariat, World Meteorological Organization,Geneva.
Jeffreys H. (1961) Theory of Probability. Clarendon Press, Oxford.Kass R. E. & Raftery A. E. (1995) Bayes factors. J. Am. Stat.
Assoc. 90, 773–95.Landsberg J., Morse J. & Khanna P. (1990) Tree dieback and
insect dynamics in remnants of native woodlands on farms.Proc. Ecol. Soc. Aust. 16, 149–65.
Lloret F., Siscart D. & Dalmases C. (2004) Canopy recoveryafter drought dieback in holm-oak Mediterranean forests ofCatalonia (NE Spain). Global Change Biol. 10, 2092–9.
Lunn D. J., Best N. & Whittaker J. (2008) Generic reversiblejump MCMC using graphical models. Stat. Comp. doi:10.1007/s11222-008-9100-0
JOBNAME: No Job Name PAGE: 9 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
MacGregor S. D. & O’Connor T. G. (2002) Patch dieback ofColophospermum mopane in a dysfunctional semi-arid Africansavanna. Austral Ecol. 27, 385–95.
McLaughlin S. & Percy K. (1999) Forest health in NorthAmerica: some perspectives on actual and potential rolesof climate and air pollution. Water Air Soil Pollut. 116, 151–97.
Margules ••., Partners Pty. Ltd., Smith P., Smith J. & DCFL(1990) Riparian Vegetation of the River Murray. ReportPrepared by Margules and Partners Pty. Ltd., P. & J. SmithEcological Consultants and Department of ConservationForests and Lands. Murray-Darling Basin Commission,Canberra.
MDBC (2005a) The Living Murray Foundation Report onthe Significant Ecological Assets Targeted in the FirstStep Decision. Murray-Darling Basin Commission,Canberra.
MDBC (2005b) Survey of River Red Gum and Black Box Healthalong the River Murray in New SouthWales,Victoria and SouthAustralia – 2004. Murray-Darling Basin Commission,Canberra.
Mensforth L. J., Thorburn P. J., Tyerman S. D. & Walker G. R.(1994) Sources of water used by riparian Eucalyptus cam-aldulensis overlying highly saline groundwater. Oecologia 100,21–8.
Mueller-Dombois D. (2006) Long-term rain forest successionand landscape change in Hawai’i: the ‘Maui Forest Trouble’revisited. J.Veg. Sci. 17, 685–92.
Nicholls N. (2004) The changing Nature of Australian droughts.Clim. Change 63, 323–36.
Specht R. L. (1981) Major vegetation formations in Australia.In: Ecological Biogeography of Australia (ed. A. Keast),pp. 163–298. Junk, The Hague.
Spielgelhalter D., Thomas A. & Best N. (2003) WinBUGSVersion1.4. Bayesian Inference Using Gibbs Sampling. MRC Biosta-tistics Unit, Institute for Public Health, Cambridge.
Sprugel D. G. (1976) Dynamic structure of wave-regeneratedAbies balsamea forest in the north-eastern United States.J. Ecol. 64, 889–911.
ter Steege H. (1996) WINPHOT 5.0: a programme to analyzevegetation indices, light and light quality from hemisphericalphotographs. Tropenbos Guyana Programme, Report 95-2.Tropenbos, Guyana.
Wintle B. A., McCarthy M. A., Volinsky C.T. & Kavanagh R. P.(2003) The use of Bayesian model averaging to better rep-resent uncertainty in ecological models. Conserv. Biol. 17,1579–90.
Worrall J. J., Egeland L., Eager T. et al. (2008) Rapid mortality ofPopulus tremuloides in southwestern Colorado, USA. For.Ecol. Manage. 255, 686–96.
JOBNAME: No Job Name PAGE: 10 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
SNP Best-set Typesetter Ltd.Journal Code: AEC Proofreader: JasonArticle No: 2043 Delivery date: 7 September 2009Page Extent: 9 Copyeditor: Harry
AUTHOR QUERY FORM
Dear Author,During the preparation of your manuscript for publication, the questions listed below have arisen. Please
attend to these matters and return this form with your proof.Many thanks for your assistance.
QueryReferences
Query Remark
1 AUTHOR: Is this short title running head OK? If not, please provide anotherone, ensuring that it is no more than 40 characters including spaces.
2 AUTHOR: Close et al. 2008 has not been included in the Reference List,please supply full publication details.
3 AUTHOR: ‘36° 06′ S 147° 01′’ and ‘34° 01′ S 141° 00′’ have been changed to‘36°06′S, 147°01E′’ and ‘34°01′S, 141°00′E’, respectively. Is this OK?
4 AUTHOR: The definitions of the variables COND, LONG and DISTWB in thisparagraph have been deleted because these variables have been defined earlier.Is this OK?
5 AUTHOR: Spiegelhalter et al. 2003 has been changed to Spielgelhalter et al.2003, so that this citation matches the Reference List. Please confirm that thisis correct.
6 AUTHOR: ‘MCMC mixing and convergence’ has been changed to ‘Mixing andconvergence of MCMC’. Is this OK?
7 AUTHOR: As per journal style, original paper should not include first-levelheading ‘CONCLUSION’. So now the heading ‘CONCLUSION’ has beendeleted, and ‘In conclusion’ has been added at the beginning of the paragraph.Is this OK?
8 AUTHOR: (BOM, 2007) Please provide the date you accessed this articleonline
9 AUTHOR: (BOM, 2007) Please check this website address/URL and confirmthat it is correct (Please note that it is the responsibility of the authors to ensurethat all URLs given in this article are correct and useable.)
10 AUTHOR: Davidson et al. (2007) has not been cited in the text. Please indicatewhere it should be cited; or delete from the Reference List.
11 AUTHOR: Durrleman and Simon (1989) has not been cited in the text. Pleaseindicate where it should be cited; or delete from the Reference List.
12 AUTHOR: Please provide the page range for Horner et al., 2009.
JOBNAME: No Job Name PAGE: 11 SESS: 66 OUTPUT: Mon Sep 7 16:56:12 2009/v2501/blackwell/A_journals/aec_v0_i0_corr_doi/aec_2043
13 AUTHOR: (Lunn et al. 2008) If this reference has now been published in print,please add relevant volume/page/year information
14 AUTHOR: (Margules et al., 1990) Please confirm the author names arecorrect, and please provide Margules’s initial
15 AUTHOR: Can the boldface on 0.865, 3.309, 1.000, •, 0.883 and 12.33 beremoved?
![Carbon dynamics of eucalypt seedlings exposed to progressive drought in elevated [CO2] and elevated temperature](https://static.documents.page/doc/80x56/635bc559a0f1eac29f0b61c5/carbon-dynamics-of-eucalypt-seedlings-exposed-to-progressive-drought-in-elevated.jpg)
