Restoration thinning promotes resprouting and recruitment in an Australian floodplain forest
Author contributions: EG, LW, RW, DM, SKT, PS, KL conceptualized the study; DM, EG, RW collected and curated the data; LW, EG, SKT, ES designed and conducted the data analyses; EG, LW wrote the manuscript; all authors reviewed and edited the manuscript; KL, DM, PS administered the project and acquired funding.
Abstract
Thickening of woody vegetation has degraded numerous ecosystems globally. In forests, woody thickening often follows gap-creating disturbance that removes mature trees and promotes the dense recruitment of woody resprouts or seedlings. Restoration thinning seeks to reduce tree densities in thickened forests to hasten recovery of stand structure for habitat and other conservation outcomes. However, as restoration thinning involves gap-creating disturbance, it may stimulate further recruitment. We investigated recruitment responses to restoration thinning in thickened river red gum (Eucalyptus camaldulensis Dehnh.) forests on the Murray River floodplain in semiarid south-eastern Australia by implementing mechanical thinning at various intensities. The average distance between retained trees (up to 15 m) was intended to be insufficient to stimulate further recruitment, and herbicide was applied to cut stumps to reduce resprouting. We recorded seedling abundance annually for 5 years and resprout abundance after 5 years. Resprouting occurred at all levels of thinning intensity. On average, almost one third of the trees affected by thinning resprouted with over three resprout stems per resprouting tree. Thinning that reduced tree densities to below about 400 per hectare also increased seedling abundance by up to 7500 additional seedlings per hectare in some plots in the fifth year of the study when establishment conditions were favorable. These results demonstrate that effective recruitment controls must be identified prior to implementing restoration thinning programs. Without effective recruitment controls, restoration thinning may stimulate more stems than were removed by thinning, and therefore be an ineffective conservation intervention.
Implications for Practice
- Widespread recruitment of resprouts and seedlings may result from a single mechanical thinning treatment in which all trees greater than 40 cm diameter are retained and average thinning gaps are ≤15 m.
- Recruitment may be delayed until appropriate climatic or disturbance conditions occur.
- Light intensity thinning in high-density forest stands (that removes 25–33% of 750–1000 trees/ha) is less likely to stimulate extensive recruitment but may fail to promote restoration benefits such as increased tree growth.
- Resprout and seedling mitigation strategies are necessary for restoration thinning programs in forests where resprouting species dominate and/or if mass seedling recruitment may occur.
- Without effective recruitment prevention or removal, restoration thinning may perpetuate woody thickening and be counterproductive to forest restoration goals in regrowth forests.
Introduction
Woody thickening has become a widespread issue affecting various ecosystems globally, including iconic African savannas (Zhou et al. 2021), North American redwood forests (Soland et al. 2021) and eucalypt forests in Australia (McGregor et al. 2016). Thickening typically presents as dense small trees or shrubs of one or a few species within communities that were previously more open or dominated by larger trees (Dwyer et al. 2010). Woody thickening occurs when changes in disturbance regimes remove or reduce impediments to woody species recruitment over large spatial scales (Lunt et al. 2006; Aszalós et al. 2022). Woody thickening is of concern when dense stands become the dominant state in a landscape due to potential negative implications for structural complexity, habitat diversity, and numerous ecosystem functions (Allen et al. 2002; Franklin & VanPelt 2004; Lindenmayer et al. 2019).
In mature forests, existing large trees impede recruitment by occupying space, dominating resource uptake and casting shade (Coates 2002; Oliveira et al. 2019). The death of a large tree can precipitate small-scale recruitment, a process which is integral to forest ecology, heterogeneity, and succession (Watt 1947; Franklin & VanPelt 2004; Simkin & Baker 2008). Anthropogenic disturbances such as widespread logging can initiate widespread woody thickening by substantially reducing large tree abundance over large spatial scales, thereby creating extensive gaps in which mass recruitment can occur (Tappeiner et al. 1997; Rutten et al. 2015). Depending on the availability of seed and the productivity of the site, a pulse of recruitment may occur after gap formation (Grubb 1977) that creates a high-density cohort of trees in a short period of time (1–10 years) (George et al. 2005; Kenny & Moxham 2022). In woody species for which recruitment is triggered by stochastic events such as a fire or flood (episodic or “event-driven” recruitment), mass recruitment may be delayed after gap-creation until an appropriate trigger occurs. These high-density cohorts of trees (Naficy et al. 2010; Rutten et al. 2015) may develop intense competition, slowing growth and delaying progression to a mature forest structure (Tappeiner et al. 1997).
Restoration thinning has been proposed as an appropriate management action for accelerating the development of trees with old-growth features and restoring structural diversity in thickened forests (e.g. Dwyer et al. 2010; Dagley et al. 2018). Both commercial and restoration thinning involve the selective removal of a proportion of trees throughout the forest to reduce tree densities (Moore et al. 1999; Roccaforte et al. 2015) with the aim of increasing growth in retained trees (Latham & Tappeiner 2002; Baker et al. 2008; Brown et al. 2019). In native multi-aged forests, commercial thinning typically prioritizes retention of trees with a straight bole and minimal branching that are suitable for producing sawlogs. In contrast, trees prioritized for retention in restoration thinning typically have a branched form with the potential to develop a wide branching crown and tree hollows via branch abscission (Lindenmayer et al. 1993; Gorrod et al. 2024a). Restoration thinning has been trialed with mixed results in redwood and ponderosa pine forests in North America (Kolb et al. 2007; Plummer et al. 2012; Dagley et al. 2018) and in brigalow and eucalypt forests in Australia (Dwyer et al. 2010; Brown et al. 2019). Thinning approaches include reducing tree densities throughout a stand (stand thinning) or removing trees from directly around retention trees (crown thinning) but successful stimulation of large tree growth has only been documented where thinning has been of a high intensity and created gaps of sufficiently large size to allow large trees to develop (Skov et al. 2005; Puettmann et al. 2016; Hood et al. 2018).
Given that woody thickening in forests can be caused by gap-creating disturbance, restoration thinning in regrowth forests may stimulate a subsequent recruitment response (O'Hara et al. 2010; Harrington & Devine 2011; Oliveira et al. 2021), thereby creating a new cohort of small trees and repeating patterns of dense regrowth and growth stagnation (Naficy et al. 2010). Few studies have considered the potential for restoration thinning to restart widespread recruitment processes and perpetuate woody thickening (Kolb et al. 2007; Harrington & Devine 2011). The two primary modes of woody recruitment following gap-creating disturbance in forests are seed germination and the resprouting of new stems from stumps, roots, or fallen trees (Franklin & VanPelt 2004; Li et al. 2022). Germinants can readily colonize gaps that have sufficient light exposure and lack competition (Baur 1984; Ren et al. 2015), while resprouting can be triggered even by low-severity disturbance (Bellingham & Sparrow 2000; Bond & Midgley 2001). Regeneration from seed generally prevails in moist fertile conditions, while resprouting is common in less productive environments (Bellingham & Sparrow 2000; Clarke et al. 2013). Both seeding and resprouting capacity may be influenced by forest age or successional stage (Vesk 2006; Bowd et al. 2021). Thus, the effects of thinning on seedling and resprout establishment may be highly variable between forest types and conditions.
Guidance is needed for the restoration of vegetation that has undergone widespread woody thickening (Jones et al. 2015; Soland et al. 2021), including for river red gum (Eucalyptus camaldulensis Dehnh.) forests in south-eastern Australia. In these forests, recruitment is inhibited in the vicinity of large mature trees (Dexter 1978; Mac Nally et al. 2011), but E. camaldulensis can resprout following disturbance (Dexter 1978; Nicolle 2006; Chong et al. 2007) and mass seed germination can occur when adequate moisture is available (Dexter 1978; Kube & Price 1986; Jensen et al. 2008). Historical timber extraction caused widespread woody thickening, possibly exacerbated by changes to flooding regimes (Bren 1992; McGregor et al. 2016), prior to approximately 30,000 ha of forest being reserved as a national park. Managers require evidence about whether restoration thinning is effective at restoring stand structural diversity and habitat features, without also stimulating widespread tree recruitment (Gorrod et al. 2017).
We examined the effects of restoration thinning on tree recruitment in river red gum forests and evaluated whether the strength of the response depended on gap size. Thinning was implemented by removing small trees (<40 cm diameter at breast height [DBH]) at a range of thinning intensities (Gorrod et al. 2017). We assessed whether stem resprouting and seedling establishment responses to thinning varied with site quality (spatial water availability) and explored whether the timing of recruitment responses was dependent on other recruitment triggers (temporal water availability). We expected that thinning would induce minimal stem resprouting due to herbicide treatment of cut stumps, and we expected that seedling establishment would be limited due to the restricted spacing of maximum thinning intensities.
Methods
Study Area and Study Ecosystem
The restoration thinning trial was undertaken in semiarid forest within Murray Valley National Park (lat 35°49′03″S, long 144°58′00″E) in New South Wales, Australia (Fig. 1), at an elevation of approximately 100 m above sea level (Dexter 1978). The region experiences hot summers and cool winters (mean daily temperature range 15–32 and 3–15°C, respectively). Average annual rainfall is 415 mm, with highest mean monthly rainfall during Austral winter (35 mm) and lowest during summer (16 mm). Evaporation greatly exceeds rainfall in most months of the year (Doody et al. 2015). The study site is located on the Murray River floodplain in the vicinity of the Barmah Choke, a narrow section of the river that causes flows to spread over the floodplain (Harrington & Hale 2011). The study area experiences spatially and temporally variable inundation due to small topographical differences within the relatively flat floodplain (Dexter 1978) and the volume of flow through the upstream Yarrawonga weir (Bren 1991).
Our study occurred within the largest remaining stand of river red gum forest (32,000 ha), which has a mono-dominant river red gum canopy and an herbaceous understory (Di Stefano 2002). Shrubs of Exocarpos strictus R.Br. are infrequent and Acacia dealbata Link. are rare, and river red gum trees and saplings typically comprise the only woody vegetation in a stand. River red gum seed production is prolific, and germination often occurs synchronously in response to episodic flooding (Kube & Price 1986; Jensen et al. 2008). Large river red gum trees are reported to inhibit seedling recruitment within a radius of 20–40 m of a mature tree, depending on site productivity (Dexter 1978). Established trees have basal and epicormic resprouting capacity in response to disturbance (Dexter 1978; Nicolle 2006; Chong et al. 2007), unlike some other lignotuberous species that resprout continuously (Mesléard & Lepart 1989). Undisturbed old-growth river red gum forests have an open structure maintained by the competitive dominance of large old trees, with small patches of higher density recruitment that developed in response to sporadic small-scale disturbance events such as tree death (Dexter 1978; Mac Nally et al. 2011).
The structure of these forests was transformed by over 150 years of widespread timber extraction and changes to flooding regimes (Gorrod et al. 2017). The removal of most large trees (Donovan 1997) followed by post-logging recruitment resulted in an average nine-fold increase in tree density since the mid-1800s (McGregor et al. 2016). Contemporary stands are therefore characterized by high tree densities, small average tree sizes, and constrained growth typical of thickened regrowth forests (McGregor et al. 2016; Gorrod et al. 2017). Further, river regulation has altered the flooding regime, causing reduced flood frequency, timing, duration, and extent (Dexter et al. 1986; Mac Nally et al. 2011). Reduced flooding disturbance may have increased the survival of recruiting seedlings, exacerbating woody thickening (Bren 1992). Since 2010, the Murray Valley National Park has been managed for the protection and restoration of forest biodiversity and internationally significant wetland ecosystems.
Restoration Thinning Trial
A trial was implemented to assess whether restoration thinning was effective at improving biodiversity and habitat, including hastening the development of large hollow-bearing trees and increasing structural diversity (Gorrod et al. 2017). Thinning treatments were applied to plots within 22 sites (Fig. 1). Each site consisted of three 9 ha square plots at least 100 m apart, with plots randomly assigned to one of three treatments: Control (no action), thinning to 7 m tree spacings, and thinning to 15 m tree spacings. Sites had not been subject to fire or logging disturbance for at least a decade. Pre-thinning tree densities in each 9 ha plot were surveyed by counting all trees greater than 1.37 m in height in ten 20 × 50 m subplots. Plots spanned a spectrum of initial tree densities between 262 and 1856 trees/ha. All trees greater than 40 cm DBH and trees of any size with visible hollows were retained for habitat preservation. Thinning was implemented by selecting trees for retention every 7 or 15 m throughout the plot and subsequently removing all smaller trees with 10–40 cm DBH between selected trees. Depending on the initial tree density, these treatments represented a range of thinning intensities, and between 67 and 977 trees/ha were removed from each plot (mean 377 trees/ha).
Trees were felled and removed using mechanical harvesting equipment, and thinned woody material was removed from the site. Trees less than 10 cm DBH were too small for removal by mechanical harvesters and no prescriptions were given for their treatment, but many were removed or pushed over during thinning operations (Fig. S1). Cut stumps were treated with Glyphosate herbicide using a vehicle-mounted sprayer immediately after cutting in an effort to control resprouting. Thinning was carried out from April 2016 to August 2017, with a temporary cessation of works between October 2016 and January 2017 due to flooding. Tree densities were surveyed post-thinning using the same 10 subplots as the pre-thinning survey, and thinning intensity was calculated as the proportion of trees removed from each plot (ranging from 0 to 0.86).
Spatial and temporal water availability was also quantified for each study site. Sites were stratified by site quality, a previously mapped surrogate for spatial variation in long-term water availability within the study area (Baur 1984). Half the sites were located within taller forest typical of flood-prone sites with near-surface groundwater (site quality 1). The remaining sites were placed within drier landscape settings that support a shorter average stand height (site quality 2). A continuous predictor representing temporal variation in water availability was calculated using the mean daily flow (gigalitres [GL]/day) at the Yarrawonga Weir (Fig. 1). To account for potential differences in flooding conditions among plots due to varying inter-annual monitoring intervals, we calculated this predictor as the average daily flow between survey dates for each plot (Fig. S2).
Field Measurements
Resprout Abundance
Within each of the 9 ha thinning treatment plots, the total number of resprouts was counted in ten 0.1 ha subplots (20 m × 50 m) (Fig. S3). Resprouts were defined as sapling-sized stems (>1.37 m in height) that arose from the cut stumps of trees removed for thinning or from saplings that were pushed over or damaged during thinning operations. Where multiple resprouts arose from a single stump or sapling, all stems were counted. The DBH of all counted resprouts was also recorded. Resprouts were counted once between February and April 2022, approximately 5 years after thinning operations were completed. Resprouts, by definition, did not occur in control plots, which were assigned a value of zero.
Seedling Abundance
Seedlings were counted in all plots within each site prior to thinning (between October 2015 and June 2016) and then annually for 5 years after thinning (between August and April from 2017–2018 to 2021–2022). On each monitoring occasion, the total number of seedlings was counted within three 0.04 ha subplots (20 m × 20 m) within each 9 ha plot (Fig. S3). Seedlings were defined as non-resprout recruits that were under 1.37 m in height and no longer retained cotyledons.
Data Analysis
We used R (v4.2.2, R Core Team 2022) for all data analysis. We ran hierarchical linear mixed models using the glmmTMB package (Brooks et al. 2017). We used the DHARMa package (Hartig & Lohse 2022) to extract simulated residuals from the fitted model that we inspected to check model assumptions (Figs. S4 & S5). We assessed the significance of all predictor terms and thinning interactions by comparing the goodness of fit of the complete model with that of null models via parametric bootstrapped likelihood ratio tests.
Resprout Abundance
The number of resprouts per 0.1 ha subplot was used as the dependent variable for modeling resprout abundance. Site quality was used as a categorical predictor to represent spatial variation in water availability (site quality 1—wetter sites; and site quality 2—drier sites). Thinning intensity (proportion of trees removed) and the initial tree density (trees/ha) of each plot were included as continuous predictors. We fitted a model that included the three-way interaction between thinning intensity, initial tree density and site quality. Where models include an interaction between two continuous variables, unmodeled non-linear relationships can affect the interaction term (Duncan & Kefford 2021). Therefore, we additionally included quadratic terms for all continuous predictors reasoning that our continuous variables lacked the resolution to use additional degrees of freedom to fit more flexible non-linear terms. We included a unique identifier for each 9 ha plot as a random effect to account for the clustering of subplots within plots. More complex random effects structures caused model convergence problems. We used a negative binomial error distribution with variance increasing linearly with the mean (“nbinom1” family) and a log link.
For each plot, we calculated the abundance of resprouting individual trees per hectare relative to the number of individual trees removed per hectare via thinning. The ratio of resprout stems to resprouting individuals was also calculated for each plot to determine the mean number of resprouts per individual tree. We also examined the spread of resprouts across DBH size classes (<5, 5–10, and 10–20 cm) in each plot.
Seedling Abundance
The count of seedlings per 0.04 ha subplot was used as the dependent variable for modeling seedling abundance. Site quality was included as a categorical predictor to account for spatial variation in water availability and mean daily flow (GL/day) at the Yarrawonga Weir for the inter-monitoring period was included for temporal variation in water availability. Thinning intensity (proportion of trees removed) and the initial tree density (trees/ha) of each plot were included as continuous predictors. We also calculated the decimal number of years elapsed since thinning for each subplot on each monitoring occasion and included this as a continuous predictor for time since thinning. The model included all predictor terms along with quadratic terms for all continuous predictors. We included two-way interactions between all continuous predictors and allowed all terms to interact with site quality. Random effects of site-plot (a unique identifier combining site and plot) and site-plot-subplot (a unique identifier combining site, plot, and subplot) were included to account for subplot clustering within plots and for repeated measures within subplots. More complex random effects structures caused model convergence problems. We used a negative binomial error distribution with variance increasing quadratically with the mean (“nbinom2” family) and a log link.
Model Predictions
We used the parameters package (Lüdecke et al. 2020) to produce bootstrapped model estimates with 999 iterations, and the emmeans package (Lenth 2023) to predict estimated marginal means for specific factor combinations. We predicted the abundance of resprout stems and seedlings for various levels of pre-thinning tree density (500, 750, and 1000 trees/ha) at both levels of site quality. For each level of pre-thinning density, we predicted abundance at control plots with no thinning as well as at levels of thinning intensity that resulted in post-thinning tree densities of 250, 500, and 750 trees/ha. In addition, we predicted seedling abundance in year 5 for pre- and post-thinning density values between 200 and 1000 trees/ha to determine the circumstances in which thinning had a significant effect. For the seedling abundance prediction, we included time since thinning and subplot mean flow, and we matched the average annual mean flow rate to each year since thinning. For the pre-thinning year, we predicted for the control treatment only. We predicted response values with bias correction (sigma = the sum of the random effect variances) to obtain back-transformed estimates. We converted predictions to the number of resprouts or seedlings per hectare for plotting the predicted abundance of resprouts and seedlings. We used the contrasts function to compare estimated marginal means for control plots with each level of thinning intensity across all levels of other factors. Where confidence intervals for the predicted differences did not include zero, this was considered to be evidence of a significant effect.
Results
Resprout Abundance
Tree resprouting occurred in all thinned plots (Figs. 2 & 3). On average, the number of individual trees that resprouted per hectare was 28% of the individual trees removed via thinning (Fig. S6). However, in some plots, the number of individual trees resprouting was up to 80% of the original number removed (Fig. S6). The mean ratio of resprout stems per resprouting individual was 3.5 across all plots, although in some plots there were over 6 resprout stems per resprouting individual (Fig. S7). The modeled mean number of resprout stems exceeded the number of trees removed by thinning in most circumstances in site quality 1 (Fig. 4), excepting some very high-density plots (1000 trees/ha) (Table S5). In site quality 2, resprout stems exceeded the number of trees removed when thinning removed at least 500 trees/ha (Table S5; Fig. 4).
The abundance of resprout stems was more strongly positively correlated with thinning intensity in site quality 2 than in site quality 1 (Fig. 2). In both site qualities, the greatest abundance of resprout stems arose in plots with very high pre-thinning densities (1000 trees/ha). In site quality 1, the highest resprout abundance occurred when thinning removed 500 trees/ha, generating a predicted average of 850 resprout stems ha−1 (Table S5). In site quality 2, the greatest densities occurred when 750 trees/ha were removed, generating 1250 resprout stems/ha (Table S5). However, in both site qualities, the fewest resprout stems also arose in very high-density plots (Fig. 2). For site quality 1, the lowest resprout abundance occurred when thinning removed 250 or 750 trees/ha, generating approximately 220 or 400 resprout stems/ha, respectively (Table S5). For site quality 2, the lowest abundances occurred when 250 trees/ha were removed, generating approximately 50–75 resprout stems/ha (Table S5).
The model summary and model comparisons via likelihood ratio tests showed evidence that thinning intensity, initial tree density, and site quality were significant predictors of resprout abundance, and that there were significant interactions among these predictors (Tables S1 & S2). Model predictions and comparisons of estimated marginal means between thinned and unthinned control plots provided evidence that higher resprout abundance in thinned plots was statistically significant (p value ≤0.05) at all levels of thinning intensity, site quality, and initial tree density (Table S5; Fig. 3).
On average, cut stumps of target trees produced 29% of resprout stems while the remainder arose from saplings that were pushed over or damaged during thinning operations. Most resprout stems (94%) had not yet reached 5 cm DBH. However, larger resprouts (5–10 cm DBH) were recorded in most thinned plots (41 out of 44), and resprouts 10–20 cm DBH were recorded in 12 plots.
Seedling Abundance
Before thinning and up to 4 years after thinning, predicted mean seedling abundance was less than 2000 seedlings/ha within all combinations of site quality, initial tree density, and thinning treatment (Table S5; Fig. 5). In site quality 1, thinning tended to decrease seedling abundance relative to control plots from 2 to 4 years post-thinning (Fig. 5). Evidence from marginal means suggested that the decline was statistically significant (p value ≤0.05) in year 2 on plots that had moderate pre-thinning density (500 trees/ha) for which thinning reduced tree density by half (to 250 trees/ha) (Fig. 6). The magnitude of this temporary reduction was approximately 475 fewer seedlings/ha relative to the 822 predicted to occur on control plots (Table S5). In site quality 2 between years 1 and 4, thinning caused a sustained increase in seedling recruitment relative to control plots (Fig. 5). However, this was only statistically significant (at p value ≤0.05) from year 2 when post-thinning tree densities were reduced to 250 trees/ha and pre-thinning tree densities were less than 1000 trees/ha (Table S5; Fig. 6).
In the fifth year after thinning, predicted mean seedling abundance was higher than in previous years across all plot types (Fig. 5), ranging from 2766 to 11,464 seedlings/ha (Table S5). In year 5, mean seedling abundances were consistently higher in thinned plots than in control plots (Fig. 5), although effects were not always statistically significant (at p value ≤0.05) (Table S5; Fig. 6). In site quality 1, marginal means provided evidence that the effect of thinning was statistically significant in plots with very high pre-thinning density (1000 trees/ha) for which moderately heavy thinning had reduced post-thinning density below 400 trees/ha (Figs. 6 & S8). On these thinned site quality 1 plots in year 5, approximately 10,000 seedlings/ha were predicted to occur, in comparison to approximately 2500 seedlings/ha occurring in control plots (Table S6; Fig. 5). In site quality 2, evidence from marginal means indicated that the effect of thinning was significant in plots that had moderate to high pre-thinning densities (500–750 trees/ha) for which thinning of moderate intensity reduced densities below 350 trees/ha (Figs. 6 & S8). Unthinned site quality 2 control plots with 500–750 trees/ha had approximately 3000–4000 seedlings/ha in year 5, and thinned plots had an average of approximately 6000–8000 additional seedlings/ha (Table S6).
The model summary and model comparisons via likelihood ratio tests showed evidence that thinning intensity, time since thinning, mean river flow, and site quality were significant predictors of seedling abundance, and that there were significant interactions between these predictors (Tables S3 & S4). Evidence for the statistical significance of initial tree density was inconclusive, with mixed outcomes in the likelihood ratio tests and model summary (Table S4). Results were temporally and spatially variable, with some thinning intensities affecting seedling abundance in some years, within some combinations of site quality and initial stem density (Table S5; Figs. 5 & 6).
Discussion
Restoration Thinning Can Perpetuate Woody Thickening in Regrowth Forest
Our results suggest that restoration thinning stimulated tree recruitment across much of the study area, consisting of hundreds of resprouts and thousands of additional seedlings per hectare. Thinning caused resprouting to occur in all plots, and in many cases, the modeled average abundance of resprout stems was greater than the number of trees removed by thinning. Thinning temporarily reduced seedling recruitment in some plot types when overall seedling densities were low (2 years post-thinning). However, after a stochastic germination trigger (flooding) occurred 5 years post-thinning, seedling abundance on all thinned plots was at parity with or substantially greater than control plots. In our study, the only conditions for which an extensive recruitment response from either resprouts or seedlings was less likely to occur were some lightly thinned plots (250 trees/ha removed) with high to very high pre-thinning tree densities (750–1000 trees/ha). Low-intensity thinning in these conditions is unlikely to provide restoration benefits such as increased tree growth (Hood et al. 2018; Gorrod et al. 2024b) to hasten the development of large trees.
Both resprouting and seedling establishment may prevent realization of restoration goals relating to stand structural diversity and large tree development, and both may perpetuate the prevalence of high-density forests at a landscape scale. Other studies have similarly reported dense infilling with seedlings and resprouting stems following restoration thinning, including in redwood forests (O'Hara et al. 2010), western redcedar forests (Harrington & Devine 2011) and temperate broadleaved forests in China (Li et al. 2022). In ecosystems that are susceptible to woody thickening, it is therefore critical to understand woody species recruitment strategies and determine effective recruit prevention or removal methods for effective restoration thinning. This includes an understanding of the effects of interacting drivers such as temporal and spatial resource availability, which may determine the timing and magnitude of seedling recruitment responses after thinning.
Mechanical Thinning Stimulated Widespread Resprouting
Although resprouting only occurred in an average of 28% of the trees that were removed, an average of 3.5 stems arose from each resprouting individual, with modeled averages of 50–1260 resprout stems per hectare. The number of resprout stems exceeded the number of trees removed in almost all circumstances when moderate to heavy thinning was implemented (≥500 trees/ha removed). Resprouting stems tended to be most abundant on sites with very high (1000 trees/ha) pre-thinning densities. Resprout abundance was minimal in most circumstances when light intensity thinning was applied (removing 250 trees/ha). However, there was an exception in which substantial resprouting occurred after light intensity thinning on sites with higher water availability (site quality 1) and high pre-thinning densities (750 trees/ha). Other silvicultural studies have also documented abundant resprout development following mechanical tree removal, especially where small diameter trees are removed (Randall et al. 2005; Keyser & Loftis 2015). Therefore, light intensity thinning is not guaranteed to minimize resprouting, and we recommend testing resprouting responses in a range of pre-thinning stand structures prior to broadscale implementation.
Our results indicate that the herbicide treatment applied in this experiment was relatively ineffective at preventing resprouting in target trees, although higher resprouting rates of 80% have been reported where resprouting species received thinning treatments without herbicide application (Lockhart & Chambers 2007). Future research could determine whether more effective methods of resprout prevention for cut stumps are available, such as more targeted herbicide application methods or different herbicide types or strengths (Hutchinson et al. 2016; Monegi et al. 2023). Effective methods are also required for removing trees that are too small for standard harvesting equipment. In our study, almost 70% of resprouts arose from saplings and small trees (<10 cm DBH) that were pushed over without herbicide treatment. More effective resprout prevention techniques may involve higher risk to human life, water quality, or biodiversity values, or be prohibitively time-consuming and costly to implement at large scales. Restoration thinning program success will depend on whether effective methods for preventing resprouts from cut stumps and small trees can be identified.
Follow-up removal of resprouts may be feasible, given that the magnitude of resprouting is limited by tree densities prior to thinning. However, effective follow-up removal of resprouts would involve hand or mechanical removal of individual stems and would therefore need to overcome the same issues of ineffective herbicide application and machinery limitations that affect resprout prevention.
The widespread resprouting response in our study has important implications for post-thinning forest structure and stand dynamics. Despite high initial mortality, resprout survival rates tend to plateau within a few years post-thinning (Lockhart & Chambers 2007), and therefore most are likely to persist as multi-stemmed trees. With established root systems, resprouting trees will take up belowground resources and quickly re-occupy their original ecological space (Bond & Midgley 2001), thereby negating a key goal of restoration thinning to reduce competition and accelerate growth in larger retained trees. Multi-stemmed trees are less likely to develop into large hollow-bearing trees (Rayner et al. 2014), which is a key objective in many conservation and restoration strategies. The replacement of single-stemmed trees with multi-stemmed trees may also further increase stem and canopy density within these once open forests, which may impede foraging activities of some native species such as bats (Blakey et al. 2016).
Seedling Recruitment Depended on Thinning Intensity and Water Availability
Moderate and heavy thinning amplified seedling recruitment by thousands of seedlings per hectare across some of the study area during the fifth year of the study. Seedling abundance in control plots was also highest in the fifth year of the study, suggesting that conditions during this period were favorable for germination and establishment across the study area. This was likely due to the moderately high river flow that occurred, as episodic seedling recruitment in Eucalyptus camaldulensis is triggered by optimal levels of water availability (Dexter 1978; Horner et al. 2016). Our results provide evidence that mass seedling establishment can be delayed for some time after gap creation, particularly in species whose recruitment strategies are dependent on a stochastic environmental event such as flood or fire.
Seedling recruitment responses were dependent on gap size, as thinning effect sizes generally increased with thinning intensity, and increases in seedling abundance were statistically significant (p value ≤0.05) when tree densities were reduced below 400 trees/ha. Mass seedling recruitment did not occur when small gap sizes were created by light intensity thinning on high- and very high-density plots. Other studies have also found that seedling abundance increases with the intensity of tree removal in secondary forests (Oliveira et al. 2021; Li et al. 2022). This result aligns with our predictions that more intense disturbance is likely to produce conditions that favor seedling recruitment by creating large gaps with minimal competition (Lewis & Tanner 2000; Ren et al. 2015). The thinning treatments in this study were designed to limit mass recruitment by minimizing spacings between retained trees to approximately 15 m for the heaviest thinning intensity, which was less than the reported radius of recruitment inhibition for mature river red gum trees (20–40 m; Dexter 1978). However, our results showed that this gap size effectively created newly unoccupied space that could be rapidly exploited by recruits under appropriate hydro-climatic conditions. Thinning may be more likely to stimulate seedling recruitment in smaller gap sizes when the suppressive effect of large spreading mature trees is largely absent, such as in regrowth forests (Dexter 1978; Ashton 2000).
Seedling responses to thinning in our study varied with spatial differences in water availability and initial tree density. Thinning was associated with a sustained increase in seedling abundance in drier sites, whereas thinning caused a temporary decline in seedling abundance in more flood-prone sites. In the fifth year of the study, moderate to high intensity thinning caused additional seedlings to occur in wetter plots with very high initial tree densities and drier plots with moderate to high initial tree densities. Harrington and Devine (2011) suggest that thinning on less productive sites is more likely to promote seedling recruitment, as resources are more limiting and therefore the gap from each tree removed provides a relatively substantial and prolonged competitive release. Overall, our study demonstrates that seedling responses to thinning can vary spatially as well as temporally across treatment areas, meaning that broadscale application of homogenous thinning treatments without prior testing may have mixed or unintended results.
The impact of thinning-induced seedling recruitment on future forest structure depends on seedling survival. River red gum seedlings are most susceptible to mortality in the immediate post-germination period due to prolonged inundation, severe water stress, fire, or grazing (Dexter 1978; Jensen et al. 2008). In the 2 years following our study, river flows were average to very high (MDBA 2024), evaporation was approximately 15% below-average (BOM 2024), fire did not affect any plots, and herbivore abundance was low (Gorrod 2022, unpublished data). Most seedlings can survive several months of inundation (Dexter 1978) and can develop physiological adaptations to tolerate inundation (e.g. development of adventitious roots) and desiccation (e.g. rapid establishment of a deep taproot) (Colloff 2015). Thus, seedling survival is expected to be moderate in the study area. Given the magnitude of increase in seedling abundance resulting from thinning, even moderate survival would be likely to create new even-aged dense stands of small trees over parts of the treatment area.
As E. camaldulensis tends to produce large quantities of viable seed (Jacobs 1955; Grose & Zimmer 1958), seedlings are likely to continue to germinate in gaps when hydro-climatic conditions are suitable, such as heavy Winter rains or flood recession in late Spring (Di Stefano 2002). McGregor et al. (2016) present evidence that it is the presence of large trees that naturally prevents the establishment of dense woody recruitment in river red gum floodplain forests, rather than disturbance-driven seedling mortality. Therefore, any follow-up removal of seedlings would likely require repeated broad-scale application until large trees have established competitive dominance over newly created gaps. Potential treatments that could be repeatedly applied include flooding, fire, grazing, herbicide, or manual removal. Flooding to prevent seedling establishment would depend on long-term climatic conditions that generate sufficient volume to inundate relevant areas and duration (>4 months with gradual recession, Dexter 1978) of river flows for multiple consecutive years to be effective. Fire would be constrained by flooding and low ground biomass in this non-fire prone ecosystem (Colloff 2015), and repeated application of fire may have unintended or negative consequences on other ecosystem attributes such as floristic composition, tree survival, and ground-dwelling fauna. Ungulate or rabbit grazing is likely to have adverse long-term effects on soil health and vegetation structure and function (Eldridge et al. 2017; Travers et al. 2019), and although kangaroo grazing can reduce seedling survival under drought conditions (Dexter 1978), it would not be practical to control kangaroos over large spatial and temporal scales. Broad-scale application of herbicide would also not be appropriate in a flood-prone national park. Manual manipulation of seedling densities may be an acceptable and effective technique to maintain an open stand structure but is likely to be cost prohibitive at broad scales. It is therefore likely to be very challenging to implement follow-up removal of seedlings.
In comparison with resprouts, mass recruitment of seedlings will potentially develop into a much higher density stratum of trees, but over a longer time frame. The time frame for seedlings to establish as saplings depends on soil and hydro-climatic conditions, with a minimum of 10 months (Baur 1984). Once established as saplings, this cohort is likely to compete for resources and potentially prevent accelerated growth in larger retained trees (Gorrod et al. 2024b). Competition within the cohort may be intense, and it may take a very long time for self-thinning to occur (Baur 1984). This cohort of small trees is likely to be difficult to remove without stimulating resprouting, as found in this study. While both resprout and seedling recruitment responses to thinning have important implications for conservation goals and management, seedlings potentially present a problem of a larger magnitude that may arise over a longer time frame.
Acknowledgments
We acknowledge the Bangerang and Yorta Aboriginal people, upon whose country this research was conducted. The thinning trial and associated research were funded by the New South Wales Department of Climate Change, Energy, the Environment, and Water. We thank Evan Curtis, Tim O'Kelly, other NSW National Parks and Wildlife Service staff, and contractors from Charles Sturt University for assistance with data collection. Our paper has benefited from constructive feedback from Ian Oliver and three anonymous reviewers. Open access publishing facilitated by New South Wales Department of Climate Change Energy the Environment and Water, as part of the Wiley - New South Wales Department of Climate Change Energy the Environment and Water agreement via the Council of Australian University Librarians.
