Environmental predictors of forest structure, tree growth and wood production for Xylocarpus granatum in mixed-species mangrove forests

Study areas

We chose four rivers in tropical, far north Queensland, Australia, and one in Western Province, Papua New Guinea, representing a range of catchment sizes, rainfall, seasonality and annual discharge. Each river had been surveyed extensively in work on the within-estuary distribution of mangroves, had a very high diversity of mangrove species and contained stands of mixed forests in which X. granatum was a numerical dominant (Bunt et al. 1982, 1985; Robertson et al. 1991).

The Fly River (8°31′36.71″S, 143°48′40.78″E) ranks in the top 40 of the world’s rivers, with a mean annual discharge of ⁓2.35 × 10⁵ GL. The Star Mountains region (3700 m) at the top of the Fly River catchment (76,000 km²) is one of the wettest places on Earth with a mean annual rainfall of ⁓12 m year⁻¹. Mean annual rainfall at Tabubil, 434 km from the mouth of the river, is ⁓6000 mm. The mean annual maximum and minimum air temperatures at the Fly delta are 27.8 and 25.2°C respectively, and vary little throughout the year. The catchment is mainly intact, pristine rainforest with extensive and mostly intact floodplain wetland habitats. In the driest season, estuarine conditions exist for 100°km upstream from the mouth of the river and there are massive mangrove forests (874°km²) in the delta, containing at least 29 mangrove species (Robertson et al. 1991).

The Pascoe River on Cape York in northern Australia (12°29′43.01″S, 143°16′19.18″E) has a catchment area of 2124 km². Gauging station records are sparse and mean annual discharge is likely to be ⁓1300 GL. There are cattle stations in the upper catchment, whereas in the lower reaches, vegetation is in mainly natural condition. The river is 119 km long and drops 243 m from its source in the Great Dividing Range before discharging into the inshore waters of the Great Barrier Reef. Annual mean catchment rainfall is 1646 mm. The mean maximum and minimum air temperatures at the coast are 30.4 and 22.7°C. There are 33 km² of riverine wetlands in the catchment, dominated by mangroves. At the end of the dry season, the river can be saline up to 10 km from mouth (Bunt et al. 1982). Extensive surveys of mangroves in the estuary of the Pascoe have recorded 23 species of mangroves (Bunt et al. 1982).

Just to the south of the Pascoe lies the Claudie River (12°50′11.67″S, 143°21′26.57″E). It is a much smaller river (length ⁓35 km) with a catchment of 160 km². Vegetation in the catchment is in a near-natural state. The mean annual rainfall in the catchment is 1729 mm. Mean annual discharge for the river may be ⁓70 GL. Mean annual maximum and minimum air temperatures at the coast are 30.6 and 22.7°C. The Claudie is saline up to 5 km from its mouth and its estuary supports a total of 2.30 km² of intertidal forests containing 23 species of mangroves (Bunt et al. 1982).

The McIvor River (15°08′20.73″S, 145°14′05.47″E) is ⁓65 km long with a catchment area of 588 km². Cattle stations and associated fodder farming are present in the upper catchment and the lowlands have mainly intact native vegetation. Mean annual rainfall is the lowest among the five rivers (1422 mm). The mean annual discharge for the McIvor is 114 GL. There may also be significant groundwater input to the estuarine section of the river from surrounding dune fields (Holliday et al. 2007). Mean annual maximum and minimum air temperatures at the coast are 29.9 and 22.9°C. The McIvor can be saline up to 9 km upriver and has 4.10 km² of intertidal forest supporting 28 species of mangroves (Bunt et al. 1982).

The Daintree River (16°17′16.63″S, 145°27′12.76″E) has a large catchment (2125 km²) in the north of the World Heritage Wet Tropics region of far north Queensland. Mean annual rainfall in the upper catchment is 2689 mm. The river descends 1270 m over its 127-km course. Mean annual riverine discharge is ⁓1000 GL. Mean annual maximum and minimum air temperatures at the coast are 29.8 and 22.6°C. There are intact rainforest communities in its upper catchment. Clearing for agriculture in the lower catchment represents <5% of the total catchment area. The estuary of the Daintree can be slightly saline up to 18 km from its mouth and supports 28 species of mangroves in its 33 km² of intertidal forests.

Visits to the rivers were aboard a research vessel, which provided both accommodation and small boats to access research sites within each estuary. Cruises to each Australian river occurred at a frequency of approximately every 6 months over 3.2 years. For the Fly River, there were two cruises separated by 1 year.

Surveys of forest structure and environmental variables

We did not intend that this study would provide a forest inventory (sensuKauffman and Donato 2012) based on a fully randomised or stratified sampling design (e.g. Trettin et al. 2016). Rather we focused on assessing Xylocarpus forest structure and growth across as wide a set of forest and environmental settings as possible, so that we could address the main questions outlined in the Introduction.

In each river, we surveyed forest structure at eight sites distributed along the estuarine salinity gradient. Initially, we explored positions along each river and walked transect lines perpendicular to the river channel until X. granatum was encountered. Site choice on the salinity gradient ensured that there was at least one X. granatum tree present in each site. This was a necessary approach only at the extreme low- or high-salinity regions of each estuary, because X. granatum is more common in the mid-salinity zones of these rivers (Bunt et al. 1982; Robertson et al. 1991).

Relative to the Australian estuaries, logistics in the Fly River estuary are challenging. The delta region is ⁓100 km across the river mouth, estuarine conditions can extend ~100 km upstream from the front of the delta and there are 874 km² of mangrove forests. Hence, we constrained our choice of sites to within a ⁓30-km length of the estuary in a region where the low water salinities in the late drier season (September) ranged from 4 to 18.

At each site in each of the five estuaries, forest structure was assessed using a relascope (Spiegel Relaskop) and employing the angle-count cruising method (Cintron and Novelli 1984). The DBH was recorded for each tree of each species included in each 360° swing of the relascope and tree densities, and basal areas for X. granatum and all other species were calculated following English et al. (1997). Saplings of X. granatum were considered to have a DBH of <5 cm (Trettin et al. 2016). Maximum tree height for X. granatum in a stand was estimated using an inclinometer and a tape measure to record distance from the tree base and the height at the eye-line of the investigator.

The allometric relationship between DBH and total tree mass was used to estimate the above-ground biomass for X. granatum (AGB_X) (Clough and Scott 1989). With the exception of the palm Nypa fruticans (see below), for all other species encountered in the field surveys, AGB was determined using a general equation for mangroves (Komiyama et al. 2008):

where p is wood density (g cm⁻³) and D is DBH (cm). Species-specific values of p were obtained from Zanne et al. (2009). Where there was more than one value of p per species, we used Australian data. If species-specific wood densities were not available, we used data for a congener. For Nypa fruticans, woody petioles of erect leaves were considered as individuals in relascope swings, and to estimate AGB, we used the allometric equation relating dry mass of leaves to petiole DBH from Robertson et al. (2020).

At each site, we measured or took samples near the centre of each site to assess physical, chemical and biological variables that are known to influence forest structure and growth in mangroves in general and X. granatum specifically (Table 1). Methods for each variable are provided in Table 1, with reference to details described in previous papers by us and others. We also had available data on climate and hydrological variables for each river (Table 1; and see Study areas, above).

Tree growth and wood production

We established a total of 18 plots across the region to measure the growth of X. granatum over a range of environmental settings (Table 2). There were three plots in each of the four Australian estuaries and six in the Fly estuary, all arranged to cover the salinity gradient of each estuary. Plots were positioned immediately adjacent to a survey site and their sizes varied (range of areas 150–496 m²) depending on the densities of individuals of X. granatum at each site. We included at least 20 individuals of X. granatum in each plot (range for all plots, 20–41 individuals).

At the beginning of the study, in each estuary, we marked off each plot with star pickets and all individual trees of X. granatum were tagged with a stainless steel, numbered marker attached to the tree trunk loosely with stainless-steel wire. A small mark was made in the bark of each tree at 1.2 m from the ground and the DBH was recorded using a DBH tape 10 cm above the mark (i.e. at 1.3 m). The DBH of all trees that had survived was recorded in the same way after between 3.15 and 3.18 years in Australian estuaries. For the Fly River, measurements were taken before and after 1.0 year. The data were used to calculate the mean annual increment in DBH for X. granatum trees in each plot (MAI_X, cm year⁻¹).

The growth plots contained sufficient trees to estimate the MAI and to estimate P_W at the plot level (Table 2). Overall, we tagged and measured 455 trees in all rivers at the start of the study. Over the time period that elapsed between the first and second measurements of trees in plots, 22 trees died. Time gaps of 1 to 3 years between measurements are far too short a time to estimate rates of tree mortality (Lewis et al. 2004), so we excluded these trees when estimating stand-level P_W (see below).

The mass of wood in each X. granatum tree before and after the period of growth was calculated from the allometric relationship between DBH and trunk wood mass (Clough and Scott 1989). Wood-mass increment (kg) was normalised to 1 year for all trees in a plot. P_W in a plot was obtained by summing mass increments for all trees in the plot and normalising plot data to 10,000 m² to obtain annual stand-level trunk P_W (Mg ha⁻¹ year⁻¹).

Management of potential error and bias

There are four main potential sources of error when estimating tree biomass and growth in forests, viz. (1) errors in measurement of DBH, (2) choice of allometric relationship between DBH and mass, (3) the representativeness of small plots across large forest landscapes, and (4) sampling uncertainty, related to the size of the study plot (Chave et al. 2004).

We attempted to minimise uncertainty in tree-biomass estimates for X. granatum by using an allometric relationship between DBH and total tree mass and stem mass developed using trees from far north Queensland, Australia (Clough and Scott 1989). However, the maximum tree DBH included in the Clough and Scott (1989) model was 17 cm, and extrapolation to trees with DBH between 20 and 40 cm, commonly encountered in the present study, was another possible source of error. To test the direction and the possible degree of error arising from extrapolation, we used the Clough and Scott (1989) model and the general model of Komiyama et al. (2008) to compare the total tree-mass increment for X. granatum in the six forest growth plots in the Fly River, where the range of individual sizes was from 1.0 to 81.9 cm DBH. Relative to the general model, the Clough and Scott (1989) model underestimated stand mass increase by an average of 9%, an indication that our conversions of DBH to mass are conservative.

In this study, all trees were measured by one of three investigators. We assessed measurement error by allocating five trees of different DBH to each investigator and each tree was measured at five random times during a 1-h period while all other trees in a plot were being measured. The grand mean measurement error per tree was 0.06 cm, with no difference between investigators. The grand MAI across all 18 growth plots in this study was 0.25 cm (range 0.03–0.55), and only 1 of 18 plots exhibited a mean annual increment that was at or below measurement error.

We were not aiming to provide a full forest inventory, but to investigate the possible causes of variation in forest structure, growth and wood production of one species in relation to a range of environmental variables. Hence, the issue of representativeness of small plots across large forest landscapes and subsequent sampling uncertainty related to the size of study plots (Chave et al. 2004) were not sources of error in this study.

There are many ways that data collected from forest plots and used for estimates of MAI may be biased (Phillips et al. 2002). Many relate to the choice of plot positioning and have been dealt with above under errors. The greatest bias in data management relates to the non-recording of negative growth increments (Phillips et al. 2002). In this study, we took a very conservative approach to data inclusion. Overall, a total of 37 trees across plots exhibited a decrease in DBH in the periods between measurements (Table 2). In our calculations of MAI and P_W, we included all trees that did not grow more than measuring error (Table 2) and, with the exception of two trees, all trees with negative growth increments (Table 2). The two trees that were excluded were very large trees (DBH, 48.6 cm and 81.9 cm) from separate plots in the Fly estuary. Errors in DBH measurement or real shrinkage in DBH in such large, old trees would have a disproportionate effect on estimates of MIA and P_W.

Data analyses

For each river, the distribution of AGB_X at sites along the estuarine gradient was examined using a Kolmogorov–Smirnov one-sample test assuming an even distribution of AGB across sites.

Boosted regression tree (BRT) modelling was employed to investigate the environmental predictors of elements of Xylocarpus forest stand structure, including AGB_X and maximum tree height (H_max), growth (MAI_X) and trunk P_W. The approach has been used successfully in predicting the drivers of growth and yield in other forests (e.g. Aertsen et al. 2012; Gu et al. 2019). BRT modelling is a powerful approach for building predictive models because it is insensitive to outliers and multicollinearity, can fit complex non-linear relationships, automatically handles the interactive effects between predictors and assumes that relationships between variables are complex and unknown (Breiman 2001; De’ath 2007; Elith et al. 2008).

The BRT method combines large numbers of simple regression-tree models in an adaptive fitting process. Boosting is a forward-learning, sequential ensemble technique in which the model is improved using the information from previously grown weaker models. This process is continued for multiple iterations, reducing error iteratively until a final model is built following pruning (Elith et al. 2008). BRT modelling was performed using the gbm package in R (ver. 2.1.8.1, see https://cran.r-project.org/package=gbm), with a squared error-loss function, a shrinkage (‘learning rate’ in Elith et al. 2008) value of 0.001 and the maximum interaction depth (‘individual tree complexity’) set to 9. BRT models were based on 10,000 trees, and the optimum number of trees for each target variable was selected by five-fold cross-validation of a 0.5 out-of-bag (OOB) fraction. The partial relationships were monotonically constrained where associative polarity was evident (based on simple univariate explorations) to help mitigate artifacts resulting from sparse data. The above parameters were selected by a tuning process aimed at yielding an optimum number of trees of at least 1000 (ensuring adequate learning granularity), yet less than the full complement of trees (ensuring all major features had been learned). The associative profile of each predictor to the predictions was assessed by a pseudo-R².

For AGB_X, H_max, MAI and P_W we used the full suite of environmental variables (Table 1) to produce ‘ecosystem models’. For the MAI and P_W ecosystem models, we used environmental data from survey sites immediately adjacent to the growth plots. For MAI, we also ran a ‘sediment model’ to further investigate the predictive role of physical and chemical variables (Table 1). It should be noted that in the modelling of AGB_X and H_max there were only six survey sites in the Fly River, covering a narrower range of salinity. This is unlikely to have a major impact on model outputs given the robust nature of the BRT approach (see above).

When modelling was completed, the relative influences (percentage contribution) of each environmental variables on AGB_X, H_max, MAI and P_W were calculated as the proportion of the total tree splits involving each variable. The relative-importance values were ranked and those that exceeded the threshold of 100/p (where p is the number of environmental variables) were considered to be useful predictors. Individual partial-dependence plots were developed for each of the useful predictor variables to understand how the response variable changed on the basis of model predictions. This was illustrated with partial-dependence plots for the ecosystem model for MAI. We followed Aertsen et al. (2012) in summarising the main outcomes of the modelling.

Patterns in the structure of forest stands

Xylocarpus granatum was never encountered in pure stands and the mixed forests we sampled had a mean richness per site ranging from 4.8 to 6.3 species, with some sites having up to 10 tree species (Table 3). The mean numerical dominance of X. granatum ranged from 18 to 33%, and in four of the five rivers there was at least one of the eight survey sites where X. granatum was the dominant species. Other abundant mangroves in the mixed forests we sampled were Nypa fruticans, Heritiera littoralis, Bruguiera gymnorhiza and Camptostemon schultzii in the Fly estuary, and Excoecaria agallocha, B. gymnorhiza, species of Ceriops, H. littoralis and Rhizophora apiculata in the Australian estuaries (Table 3).

The DBH size distribution for X. granatum varied across estuaries (Fig. 1). In the Fly and Pascoe estuaries, there was a broad range of tree sizes. No saplings (<5 cm DBH) were included in the relascope surveys in these two estuaries (Fig. 1). However, some saplings were present in all the plots established to measure tree growth in both estuaries. In the Claudie estuary, there was a more truncated distribution of tree sizes and, although no saplings occurred in the relascope swings (Fig. 1), some did occur in all three plots established to measure tree growth. The forests in the McIvor and Daintree estuaries were dominated by saplings and small trees, but a few larger trees (>30 cm DBH) were present in the McIvor (Fig. 1). With the exception of the Daintree estuary, survey sites contained many trees of <40 cm DBH and trees of <60 cm DBH were encountered in the Pascoe and Fly estuaries. The largest tree (81.9 cm DBH) encountered in all surveys was in the Fly River.

Fig. 1.

Diameter at breast height (DBH) percentage frequency distributions of X. granatum derived from data pooled over the eight survey sites in each of the five estuaries.

The AGB_X ranged from 16 to >200 Mg ha⁻¹ across the 40 sites we surveyed, with AGB_X being greatest at sites in the Pascoe (232 Mg ha⁻¹) and Fly (151 Mg ha⁻¹) rivers. The lowest AGB_X (16 Mg ha⁻¹) was recorded in the Daintree River (Fig. 2). AGB_X was not evenly distributed across survey sites in all five estuaries (Kolmogorov–Smirnov one-sample tests for all estuaries, D_max < D_crit 0.001). In the Australian rivers (Pascoe, Claudie, McIvor, Daintree), AGB_X was usually greatest at sites in the mid- to upper reaches of each estuary (Fig. 2). In the Fly estuary, AGB_X was not evenly distributed but there was no evidence of an upstream–downstream gradient (Fig. 2) because stations did not cover the full spectrum of estuarine salinity, owing to logistic constraints (see Methods). X. granatum grew in mixed forests at all sampling sites and, with the exception of six sites in the Fly, Pascoe and Claudie rivers, where the AGB_X was similar to or greater than that of the total AGB of other species, at most sites AGB_X was less than one-quarter of that of the AGB of other species (Fig. 2).

Fig. 2.

Above-ground biomass (AGB) of X. granatum (solid bars), AGB of all other tree species (dotted bars) and the maximum tree height (H_max, closed dots) at eight sites on upstream–downstream salinity gradients in the five estuaries. Note that the upstream–downstream gradient in the Fly estuary did not cover the full spectrum of estuarine salinity, owing to logistic constraints (see Methods).

Trees of ≥18 m tall were recorded in all estuaries (Fig. 2). The range of H_max across all 40 survey sites was 4.4 m in the most upstream site in the McIvor estuary and 28.8 m at a site in the Fly River (Fig. 2, Table 3). In the Australian rivers (Pascoe, Claudie, McIvor and Daintree), H_max was greater in the upstream section of each estuary than closer to the sea (Fig. 2). In the Fly estuary, sampling stations did not cover the full spectrum of estuarine salinity, owing to logistic constraints (see Methods). There was no clear pattern of H_max across sites in the Fly estuary (Fig. 2).

BRT modelling showed that five site-level predictor variables had the most influence on AGB_X across the study area. AGB_X decreased steadily with increases in the density and basal area of other mangrove species at a site and decreased when the AGB of other species reached a threshold of 100 Mg ha⁻¹ (Table 4). AGB_X increased gradually with increasing soil redox potentials and was greater when the density of crab holes was less than 7 holes m⁻² (Table 4).

Five site-level edaphic predictor variables had the most influence on H_max (Table 4). Maximum tree heights were greater when soil redox potentials were >100 mV and when groundwater salinity was <18. H_max was greater where sediments were >40% silt, phosphorus concentration was >0.065% and decreased gradually when sediment sand content increased from 10 to 55% (Table 4).

Tree growth and wood production

Growth of X. granatum individuals in growth plots, measured as the MAI, varied over 1 order of magnitude (from 0.03 to 0.55 cm year⁻¹) across the study area (Fig. 3). The highest MAIs were recorded in some Fly River plots. The lowest growth rate was recorded from a plot in the Daintree River (Fig. 3). For data pooled across plots within rivers, there were positive correlations between MAI and DBH of individual trees in the Fly (r₁₁₆ = 0.44, P < 0.01), Pascoe (r₈₈ = 0.24, P < 0.05) and Claudie (r₇₈ = 0.36, P < 0.01), but not in the McIvor (r₆₆ = 0.21, n.s.) or the Daintree (r₇₇ = 0.01, n.s.).

Fig. 3.

Mean mean annual increment in DBH for X. granatum (MAI_X) (+1 s.e.) in all growth plots in the five estuaries.

The ecosystem BRT model for MAI showed that a mix of regional-scale climate-predictor variables and site-scale edaphic, predictor variables contributed most to variation in growth (Table 4, Fig. 4). MAI was greatest in estuaries where mean annual catchment rainfall was >4000 mm and MAI increased slightly, and was less variable, where soil redox potential was greater than 100 mV (Table 4, Fig. 4). There was a gradual increase in MAI where sediment was between 5 and 75% silt and mean minimum air temperatures were 22.6–25.2°C (Table 4, Fig. 4). The sediment BRT model for MAI also showed that MAI increased steadily with sediment phosphorus between 0.044 and 0.064% P and remained steady thereafter (Table 4, Fig. 4).

Fig. 4.

Partial-dependence plots for predictor variable with highest relative importance from the ecosystem BRT model of the mean annual increment in DBH (MAI). (a) Sediment redox potential, (b) mean annual catchment rainfall, (c) sediment percentage silt, and (d) mean annual minimum air temperature. Dotted lines are 95% confidence limits. The pseudo-R² is shown for each predictor.

The growth of individual trees from all the growth plots was also expressed as the annual wood-mass increment (kg) scaled with DBH (Fig. 5). There was a large scatter in the data, reflecting differences in MAI among rivers and plots within rivers (Fig. 3). However, annual wood-mass increment per tree generally increased with DBH and comparison with the model fitted to a global data set derived from multispecies tree growth in natural mangrove forests indicated that many individuals of X. granatum exhibited a growth rate significantly greater than for other mangrove species (Fig. 5).

Fig. 5.

Annual wood-mass increment per tree (kg) plotted against initial diameter at breast height (DBH, cm) for all individual trees of X. granatum in the 18 growth plots used in the study (n = 429). The dashed line shows the relationship between annual wood-growth increments and DBH from a recent global review for trees in natural mangrove forests (Xiong et al. 2019a).

X. granatum P_W in growth plots ranged from 0.20 to 9.86 Mg ha⁻¹ year⁻¹ across the five estuaries, with the lowest P_W recorded in the Daintree estuary and the highest in the Fly estuary (Fig. 6). Results of the ecosystem BRT model for P_W showed that P_W decreased steadily as the densities of other trees of all species increased over the range of 400–12,000 stems ha⁻¹ and increased steadily as the percentage of silt in the sediment increased over the range from 5 to 90% (Table 4). When soil redox potential was >180 mV and the density of crab holes was less than 4 holes m⁻², P_W increased (Table 4).

Fig. 6.

Annual wood production (P_W) for X. granatum in all growth plots in the five estuaries.

Variation in forest stand structure

Across its range in the Indo West Pacific region, X. granatum usually occurs in diverse mixed forests, grows in most regions of estuaries and can be a dominant species in habitats with freshwater input on well-drained sediments (Bunt et al. 1982; Smith 1992; Cole et al. 1999; Ashton and Macintosh 2002; Trettin et al. 2016; Nur-Fatehah et al. 2022). Across the five estuaries in this study, X. granatum was a numerical (5 of 40 sites) and biomass (4 of 40 sites) dominant at some sites and the species co-occurred with between 3 and 10 other species per site.

Comparison of the DBH size distributions of X. granatum in the five estuaries showed that there were many large (>30 cm DBH) individuals in the Fly and Pascoe rivers, whereas other estuaries were dominated by much smaller size classes. The diameter distribution of the trees in the McIvor and Daintree estuaries exhibited an inverse J-shape curve, which is characteristic of classic uneven-aged stands (Trettin et al. 2016). The frequency of cyclones is much greater in the McIvor and Daintree River regions than in the Pascoe and Fly River regions to the north of our study region (Bureau of Meteorology, see http://www.bom.gov.au/cyclone/tropical-cyclone-knowledge-centre/history/) and the variation in size structure may reflect the frequency and severity of historical disturbance by cyclones (Krauss and Osland 2020). The age structures of the forests in the region are unknown. Some recent advances have shown that X. granatum has annual growth rings related to river flows in the Sundarbans of Bangladesh (Chowdhury et al. 2023). However, X. granatum from the Claudie River has more than one ring per year (Hancock et al. 2006) and it appears that estuary-specific dendrological work on X. granatum will be required before the age structures of forests in this region can be determined (see Roberts et al. 2011).

More large individuals (>30 cm DBH) of X. granatum trees were encountered in four of our estuaries than in forests in Mozambique, Peninsular Malaysia and Sarawak, Malaysia (Shah et al. 2016; Trettin et al. 2016; Nur-Fatehah et al. 2022). Much larger individuals (>120 cm DBH; 31 m in height) have been recorded in Micronesia, probably reflecting the high annual rainfall (>5000 mm) in that region and time for forest tree growth since the last major cyclonic disturbance in 1905 (Cole et al. 1999).

At a few sites in our study, AGB_X was similar to mean AGBs for similar geomorphic settings across the Indo West Pacific region. AGB_X was surprisingly high in the four sites (three in the Pascoe estuary, one in the McIvor estuary) where X. granatum was a biomass dominant, with a maximum estimate of 232 Mg ha⁻¹ dry mass. When converted to a carbon equivalent (0.475×), this maximum estimate compares well with the mean for multispecies mangrove forests growing in small deltas and tidal systems (180 Mg C ha⁻¹) and large deltas (85 Mg C ha⁻¹) elsewhere in the Indo West Pacific region (Rovai et al. 2021).

The density of trees, basal area and AGB of other species in the mixed-species forests in our study region were the most important predictors of AGB_X. This is likely to reflect spatial variation in competition for light during forest establishment and development and on-going competition for sediment nutrient and lower-salinity groundwater resources (e.g. Analuddin et al. 2009; Peters et al. 2020). The BRT model for AGB_X also identified higher redox potential having a role in predicting AGB_X and AGB_X declining with the density of crab holes below a certain threshold value. We included crab-hole density as a physical measure of sediment porosity in mangroves and a priori we expected that because sesarmid crab holes are important locations for water exchange (Stieglitz et al. 2000) that their impact as a predictor of AGB_X would be positive. However, sesarmid crabs can be important predators of the seeds of X. granatum and may have helped shape forest composition well prior to our surveys (e.g. Allen et al. 2003).

Globally, mangrove tree height is a function of rainfall, air temperature and cyclone frequency (Simard et al. 2019). There was a wide range of maximum tree heights at our survey sites (4.4–28.8 m). The tallest X. granatum occurred as emergent individuals above a 10-m canopy (Robertson et al. 2020) of Nypa fruticans in the Fly estuary. BRT modelling showed that low groundwater salinity (with a threshold at 18) and greater concentrations of silt (above a threshold of 40%) and phosphorus (above a threshold of 0.065%) in the sediment were important predictors of tree height within our region.

Growth and wood production

We measured a wide range of MAI_X across the 18 forest plots in this study (range, 0.03–0.55 cm year⁻¹). MAI_X scaled with DBH in the three most northern estuaries and not the two most southern estuaries in our study. In terrestrial forests, MAI generally scales with DBH but interactions between climate and tree physiology modify responses of MAI to increases in DBH (Anderson-Teixeira et al. 2015). Relationships between DBH and MAI do not appear to be common in mangroves (Xiong et al. 2019a) and growth is complex, being influenced by climate, the duration of inundation and sediment factors such as redox potential, groundwater salinity and nutrient availability (Clough 1992; Krauss et al. 2006).

BRT models showed that MAIs increased substantially above mean annual catchment rainfall of >4000 mm and increased with increasing mean minimum air temperatures. MAI also increased with an increasing percentage of silt and when phosphorus concentrations in sediments were between 0.044 and 0.064% P. Well-drained sediments with redox potentials greater than 100 mV also predicted higher MAIs.

The four other existing estimates of MAI for X. granatum are 0.39 cm year⁻¹ in southern Thailand (United Nations Development Programme and United Nations Educational, Scientific and Cultural Organization 1991), 0.31 cm year⁻¹ in the Federated States of Micronesia (Devoe and Cole 1998), 0.24 cm year⁻¹ (based on two trees) in Kenya (Robert et al. 2011) and 0.50 cm year⁻¹ in Tanzania (Njana 2020). The highest MAIs we measured for X. granatum were in the Fly estuary (from 0.32 to 0.55 cm year⁻¹) and were higher than the median (0.24 cm year⁻¹) for mangrove species globally (data from, Xiong et al. 2019b and Njana 2020).

The calculation of annual trunk mass increments per tree incorporates tree size (DBH) within forest plots, and when scaled with tree DBH, represents an excellent way to compare mangrove tree growth across species and regions (Xiong et al. 2019a). It was clear from the results of our study that many Xylocarpus granatum trees have annual trunk mass increments that are greater than the global average for mangrove species.

Plot-level wood production by X. granatum in natural mixed-species forests ranks highly with other tropical terrestrial and mangrove forests. The highest P_W for X. granatum recorded in our study was 9.86 Mg ha⁻¹ year⁻¹ in a plot in the Fly estuary. For comparison, mean P_W for plots in tropical terrestrial forests in Borneo and the Amazon, that include some of the most productive terrestrial forest sites, were 9.73 Mg ha⁻¹ year⁻¹ and 6.53 Mg h⁻¹ year⁻¹ respectively (Banin et al. 2014). The maximum P_W for X. granatum in the Fly estuary was well above the median value of stem wood production globally of 5.75 Mg h⁻¹ year⁻¹ for species of mangroves growing in mainly natural, mixed-species forests (Xiong et al. 2019b). The only previous estimate of wood production for X. granatum is 7.38 Mg h⁻¹ year⁻¹, the mean of 2 years of data from a secondary mangrove forest in one of the wettest provinces in Thailand (Poungparn et al. 2012).