Our study site for the model application is an estuarine mangrove of the Fukido River (Fukido mangrove forest) in Ishigaki Island, Japan (Fig. 1, 24∘20′ S, 124∘15′ E). The site is characterized as a subtropical region. According to the climatological normal data obtained by the Japan Meteorological Agency, the annual-mean air temperature is 24.5 ∘C, with a monthly average of 29.6 ∘C in July and 18.9 ∘C in January (see also Fig. 4). The mean monthly precipitation is 142 mm in July and 135 mm in January. Four small rivers (R1–R4) flow into the Fukido mangrove forest, while the river R2 has two outlets (Fig. 1c). The mean discharge rates of the rivers in October 2012 were less than 0.03 m3s-1 for R1, R3, and R4 and around 0.05 m3s-1 for R2 (Mori et al., unpublished data). The tide is semi-diurnal with the highest and lowest amplitude of 1.8 and 0.8 m, respectively (Egawa et al., 2021).

The site is vegetated by two species, R. stylosa and B. gymnorrhiza. The trees of R. stylosa dominated the seaward zone, especially areas close to the river mouth (Fig. 1c), while B. gymnorrhiza dominated the landward zone. The species R. stylosa is classified as a relatively salt-tolerant species, while B. gymnorrhiza is classified as a less salt-tolerant but shade-tolerant species (Putz and Chan, 1986; Sharma et al. 2012; Reef et al., 2015). According to Ohtsuka et al. (2019), the Fukido mangrove forest is a mature and intact mangrove forest designated as natural protection area by Ishigaki city, where distinct disturbances to the mangroves have not occurred since at least 1977 based on aerial photograph analysis.

We used the tree census data of the Fukido mangrove forest shown in Suwa et al. (2021) to assess model performance. The tree census data were collected from the survey plots established along four transects (T–A, T–B, T–C, and T–D), shown in Fig. 1c. The details of the survey protocol are described in Suwa et al. (2021). The stem biomass of individual trees (MS, g) was estimated from a common mangrove allometric equation proposed by Komiyama et al. (2005), which was validated with various mangrove species: 1MS=70ρDBH/1002H0.931, where ρ is the wood density (gcm-3), DBH is the stem diameter at breast height (m) divided by 100 for the unit conversion from meter to centimeter, and H is the tree height (m). However, tree height data were occasionally absent at some plots, especially along T–C and T–D, and in such cases, the tree height was estimated using a DBH–H allometric relationship (Fig. S1a and b in the Supplement). The AGB at each plot (Mgha-1) was then calculated from the estimated stem biomass.

The crown diameter was also measured for some selected trees, besides the data shown in Suwa et al. (2021). The trees for crown measurement were randomly selected at each transect. The diameters parallel and perpendicular to the transect line were measured for each tree, and the crown diameter (Dcrown, m) was represented by the average of the values from the two directions. In total, crowns of 81 trees of R. stylosa and 103 trees of B. gymnorrhiza were measured (Supplement Fig. S1c and d).

Soil salinity and porewater dissolved inorganic nitrogen concentration (DIN) were also measured at each plot as environmental drivers of mangrove production. Soil samples were collected by inserting a PVC pipe into the soil at each plot, and soil porewater was extracted from the surface 10 cm soil sample. The porewater samples were kept frozen and brought to the laboratory for analysis. Salinity of the porewater (soil salinity) was measured using a salinity meter (PAL-SALT, ATAGO Co., Ltd., Japan), while DIN concentrations were measured using a QuAAtro 2-HR (SEAL Analytical Ltd., Germany, and BLTEC K.K., Japan). These measurements were conducted from August to September 2013. The summary of the environmental and vegetation variables at each plot is provided in Table S1 in the Supplement.

The mangrove growth model was formulated based on an individual-based model, SEIB-DGVM (Sato et al., 2007). The forest dynamics was represented by a 30 m × 30 m computational domain. In this domain, the irradiance distribution, tree establishment, death, and changes in plant morphology subsequent to growth were simulated (Sato et al., 2007). A feature of SEIB-DGVM is that it explicitly solves the effects of shading by neighboring trees on the light acquisition. The SEIB-DGVM thus provides the advantage in describing tree competition for light more than the other types of DVMs such as big-leaf or cohort-based models (Fisher et al., 2018). In SEIB-DGVM, the crown of each tree is represented by a cylindrical-shaped object divided by 0.1 m thick crown layers to account for the within-crown vertical variability in irradiance distribution. It is assumed that leaf biomass is evenly distributed in the crown layers.

Originally, the SEIB-DGVM defines four biomass pools – leaf, trunk, fine root, and stock (non-structural storage pool); the trunk includes both the above-ground stem and the below-ground coarse root (Sato et al., 2007). In this study, we considered the stem and coarse root separately to explicitly consider the role of coarse root turnover in the biomass dynamics (Castaneda-Moya et al., 2011; Adame et al., 2014). Additionally, we added a new biomass pool – the above-ground root, especially for Rhizophora species whose above-ground root, or “prop root”, could account for nearly 60 % of their AGB (Nishino et al., 2015; Vinh et al., 2019).

The original SEIB-DGVM does not have a plant hydraulic module, and the effects of soil water on stomatal conductance were empirically parameterized. It also does not account for plant nutrient uptake; thus, the plant growth depends solely on photosynthesis. The biomass allocation is modeled based on scaling law (Trugman et al., 2019a). In this study, these processes that control plant growth were almost entirely modified to describe mangrove growth under salt stress (Fig. 2). The following sections explain the modification of the SEIB-DGVM for this study related to plant hydraulics. Other modifications to the SEIB-DGVM are summarized in Notes S3 and S4.

The model was applied to the Fukido mangrove forest to test its performance in reproducing the forest structural variables (species composition, mean DBH, and AGB). The model was forced with atmospheric variables (air temperature, relative humidity, atmospheric pressure, wind speed, and cloud fraction) and substrate conditions (soil salinity and porewater DIN). Direct and diffused solar radiation and longwave radiation were calculated in SEIB-DGVM from the given variables such as cloud fraction, air temperature, and latitude (Sato et al., 2007). The atmospheric variables for the Fukido mangrove forest given to the model were derived from a global reanalysis product JRA-55 (Kobayashi et al., 2015). For long-term simulation (i.e., more than 100 years), the yearly atmospheric variation in 2013, a year when the field data collection was conducted, was repeatedly given in the simulation.

Simulations with different soil salinity, or the “salinity gradient simulation”, which varied from 18 ‰ to 36 ‰ with 2 ‰ intervals, were conducted to reproduce the forest structural variables across a soil salinity gradient. For the porewater DIN, a spatially averaged DIN (average of DIN measured at the survey plots: 200 µmolL-1) was given to the model as the representative value of the porewater DIN in this forest. In each simulation, soil salinity and the porewater DIN were set as constant due to lack of data and model on the temporal variations in substrate conditions. We also conducted “plot-wise simulation”, or the simulation for each survey plot, by giving the measured soil salinity and porewater DIN at each plot. Note that the results shown in this paper are from the salinity gradient simulation; the results of the plot-wise simulation are provided in Fig. S5 in the Supplement and discussed later.

The initial condition was set as bare land (no vegetation) for all simulations. Tree establishment occurs at 1 m × 1 m grid cells at a yearly time step according to light condition at the forest floor and a parameter of establishment probability (Pestablish, m-2yr-1) prescribed for each species (Sato et al., 2007). The species that will establish at a grid cell is determined according to a fraction of total biomass of each species in the computational domain such that a species occupying a larger fraction has a higher probability of establishment. On the other hand, it is sometimes randomly determined by a probability Estrandom, where the value of Estrandom was set to 0.05 in this study. This corresponds to Scenario 4 in the tree establishment scheme in SEIB-DGVM (see Sato, 2015, for the details). We followed Sato et al. (2007) for the initial conditions (tree morphology and biomass proportion) of the established trees.

The SEIB-DGVM uses stochastic models for the processes of tree establishment and mortality, and for this reason the result of a simulation varies every time. In this regard, we conducted ensemble simulations (20 runs) for each soil salinity in the salinity gradient simulation and extracted the general trends.

The model parameter settings related to plant hydraulics and productivity are summarized in Table 2. Other minor model parameters are summarized in Table S2 in the Supplement. The parameter values for the two species in the Fukido mangrove forest, R. stylosa and B. gymnorrhiza, were determined based on literature. If the data for a focal species were unavailable from the literature, the data from the genus or family were applied. Some parameter values were adapted from other mangrove genus or terrestrial ecosystems, and in this case, the same value was given to the two species (Table 2). The values of two plant hydraulic trait parameters – Ψlk (critical leaf water potential) and β0 (sensitivity of marginal WUE to leaf water potential in Eq. S21 in the Supplement; see Note S3) – were calibrated to reproduce the AGB and mean DBH of each species across the soil salinity gradient.

The Fukido mangrove forest's age is unknown, which makes the comparison between the model and field data difficult. However, considering that it is an old and mature forest intact at least since 1977 (Ohtsuka et al., 2019), we assumed that the forest structural variables of the Fukido mangrove forest are in steady states. We conducted long-term simulations for 450 years with this assumption, and we extracted the modeled DBH and AGB in steady states (> 300 years) and compared them with the field data.

Lastly, we performed sensitivity analysis of the plant hydraulic trait parameters (ε, P50, Ψlk, and β0) to see the relative importance of each parameter in reproducing the observed pattern of the forest structure, specifically AGB, across the soil salinity gradient. We changed the value of a target parameter of one species (either R. stylosa or B. gymnorrhiza) to the one determined for the other species, which is shown in Table 2, and ran the salinity gradient simulation. Note that to examine the sensitivity to Ψlk, we changed the values of Ψlk and Ψl,min to keep the buffer between the two parameter values. Also, to save on computational cost, we run only one simulation for each sensitivity test instead of the ensemble approach described above. Model sensitivities are shown in Fig. S6 in the Supplement.

Seasonal variations in atmospheric forcing variables and modeled species-specific gross photosynthetic rate (Pg) and transpiration (T) normalized by the leaf area index (LAI) and midday (Ψl,midday) and predawn (Ψl,predawn) leaf water potential are shown in Fig. 4. The modeled variables are from one of the ensemble simulations with soil salinity set as 30 ‰. The model demonstrated strong seasonality in photosynthesis and transpiration primarily due to seasonality in solar radiation and air temperature. The model predicted the peak of Pg/LAI in June, with values ∼ 5.1 and 4.9 gCm-2d-1 for R. stylosa and B. gymnorrhiza, respectively, and the peak of T/LAI in July–September, with values ∼ 1.07 and 0.85 mmd-1 for each species, respectively. The Pg/LAI and T/LAI were predicted to be depressed during winter (December–February), with values ∼ 3.0 gCm-2d-1 for both species and ∼ 0.43 and 0.36 mmd-1 for each species, respectively. We compared the modeled leaf-level Pg with the field-estimated values in the Fukido mangrove forest by Okimoto et al. (2007). Their measurements were conducted in an area where the LAI is 1.55, the same LAI as the one shown in Fig. 4d; thus, the effects of LAI on leaf-level Pg could be eliminated for comparison. Although the modeled Pg/LAI of both species is slightly lower than the one obtained by Okimoto et al. (2007) (∼ 1.0 gCm-2d-1), especially from June to August, overall, the model agreed well with their results.

The midday leaf water potential showed seasonal variations as with photosynthesis and transpiration (Fig. 4f). Due to the partial salt uptake of R. stylosa (as indicated by the lower ε value of this species, Table 2) that alleviates the osmotic potential difference between the soil and plant, the predawn leaf water potential of R. stylosa was constantly higher than that of B. gymnorrhiza (Fig. 4f). Rhizophora stylosa also showed larger magnitude of leaf water potential reduction at midday during summer compared to B. gymnorrhiza and higher leaf-level transpiration rate (Fig. 4e and f). During winter, due to the lowered transpiration rate, the leaf water potential reduction at midday was resultantly alleviated compared to summer.

Diurnal variations of the simulated photosynthesis, transpiration, and leaf water potential of the two species during summer and winter under two different salinity conditions (30 ‰ and 24 ‰) are shown in Fig. 5. Compared to 24 ‰ salinity, both species showed significantly lowered leaf-level transpiration rates under 30 ‰ salinity especially during summer (Fig. 5b and e), suggesting downregulation of stomatal conductance under high-soil-salinity conditions. On the other hand, the decrease in leaf-level photosynthetic rates was not significant (Fig. 5a and d). The leaf water potential during nighttime was lower when soil salinity was 30 ‰ compared to conditions when salinity was 24 ‰, due to the different osmotic potential in soil porewater. The leaf water potential, however, showed almost the same levels at midday during summer, which were close to the values of Ψlk determined for each species (Fig. 5c, Table 2). The reduction in leaf water potential to the level of Ψlk suggests the role of dynamic biomass allocation, which adjusts the whole-tree transpiration demands and hydraulic conductivity, in constraining the leaf water potential dynamics (Fig. 3). In contrast, the diurnal dynamics in leaf water potential during winter showed similar magnitude of reduction of the water potential at midday between the two soil salinity conditions (Fig. 5f).

Figure 6 shows the changes in the forest structures for over 200 years under different soil salinity conditions, 20 ‰, 24 ‰, 30 ‰, and 34 ‰, from one of the ensemble simulations (the present-day average soil salinity of the survey plots is 28 ‰). The time-series results of AGB, LAI, and mean DBH of the two species are shown in Fig. 7. Trees with DBH < 0.05 m were not accounted for in the calculation of the mean DBH because it is sensitive to the presence of small trees. Overall, the model demonstrated the significant influence of soil salinity on species composition and forest structure.

The model predicted that B. gymnorrhiza dominates over R. stylosa when soil salinity is 20 ‰ or 24 ‰ (Fig. 7a and b). Under soil salinity of 20 ‰, the AGB of B. gymnorrhiza exponentially increased up to 200 Mgha-1 after 60 years since the initial condition. It slightly decreased after that and was kept almost constant at 175 Mgha-1 after 150 years. The LAI of this species showed almost the same trend with AGB, while the mean DBH showed fluctuation especially in the first 200 years (Fig. 7a). The sudden decrease in the mean DBH is attributed to the onset of formation of forest gaps resulting from deaths of large B. gymnorrhiza trees that promoted the establishment of small trees (Fig. 6). After the decrease in the mean DBH, it gradually increased again and saturated at 0.17 m (Fig. 6a). Alternatively, the AGB and LAI of R. stylosa were significantly lower than B. gymnorrhiza, with its peak at only 25 Mgha-1 and 1 m2m-2, respectively. This can also be seen in the decreasing number of R. stylosa trees subsequent to forest growth (Fig. 6). In contrast to AGB and LAI, the mean DBH of R. stylosa reached around 0.2 m after 75 years, as large as that of B. gymnorrhiza in steady state (Fig. 7a). This suggest that some R. stylosa trees can grow until mature conditions (see also Fig. 6), while trees of this species with DBH > 0.05 m disappeared in all ensemble simulations after 300 years (Fig. 7a). The trees of R. stylosa sometimes emerge due to the random factor in the establishment process, but most of the trees did not grow more than DBH of 0.05 m in the canopy of B. gymnorrhiza.

The trend in forest growth under 24 ‰ salinity was similar to that of 20 ‰ (Figs. 6 and 7b) but showed a slightly lower and higher peak for B. gymnorrhiza and R. stylosa, respectively, of the AGB, LAI, and mean DBH. This suggests decreased productivity of B. gymnorrhiza compared to 20 ‰ soil salinity and increased productivity of R. stylosa albeit with the increase in soil salinity. The survival rate of R. stylosa was higher than the results for 20 ‰, resulting in the high mean DBH of this species throughout the simulation period (Fig. 7b).

When the soil salinity was 30 ‰, the AGB of B. gymnorrhiza significantly decreased compared to the results for 20 ‰ and 24 ‰ salinities, becoming equivalent to those of R. stylosa (Fig. 7c). The LAI and mean DBH also showed a significant decrease, suggesting significantly lowered productivity of B. gymnorrhiza. The AGB and LAI of R. stylosa significantly increased compared to the results for 20 ‰ and 24 ‰, but the mean DBH significantly decreased.

The model predicted that B. gymnorrhiza cannot grow well at 34 ‰ soil salinity and that R. stylosa dominates under this salinity condition (Figs. 6 and 7d). Despite the further decrease in AGB, LAI, and mean DBH of B. gymnorrhiza, those of R. stylosa showed almost the same level for these parameters at 30 ‰ soil salinity.

Figure 8 shows the field-measured and modeled mean DBH and AGB of R. stylosa and B. gymnorrhiza across the soil salinity gradient. The field data clearly showed the effects of soil salinity on forest structural variables – decrease in mean DBH for both species and decrease in AGB of B. gymnorrhiza but increase in AGB of R. stylosa with increasing soil salinity. The model reproduced well the said patterns across the soil salinity gradient, and the values are within or close to the field data variations (Fig. 8). The change in species composition is also well reproduced, suggesting that the model can reproduce the forest structural variables across the soil salinity gradient.

Figure 9 shows the field-measured and modeled relationship of tree density and mean individual stem biomass. Although there are some discrepancies between the model and field data especially for soil salinity conditions > 30 ‰, the model reproduced the overall pattern of the field data.

Forest growth is influenced by leaf-level and whole-plant CO2, water and nutrient fluxes, and forest-scale tree competition, which are all interconnected. The leaf-level fluxes were simulated using a well-established stomatal optimization scheme with the marginal WUE linked with leaf water potential (Bonan et al., 2014; Xu et al., 2016). The model predicted the distinct seasonal dynamics in photosynthesis and transpiration as well as leaf water potential in the Fukido mangrove forest (Figs. 4 and 5). The modeled seasonal variations in leaf-level photosynthesis (Pg/LAI) agreed well with the one measured by Okimoto et al. (2007) in this forest (Fig. 4d). Although there are no data on the seasonal variations in transpiration in this forest, studies on other subtropical mangrove forests, such as the Everglades National Park, Florida (Barr et al., 2014), and China (Liang et al., 2019), that incorporated the eddy-covariance approach also showed strong seasonality in transpiration, similar to the one predicted for the Fukido mangrove forest in this study (Fig. 4e). The evapotranspiration rate normalized by LAI in the Everglades measured by Barr et al. (2014) was 0.4–1.2 mmd-1, which is close to the variation of the modeled T/LAI in the Fukido mangrove forest (Fig. 4e). These results suggest that the model produced realistic seasonal dynamics for transpiration in the Fukido mangrove forest.

Tree growth was driven by C and N uptake rates in the developed model resulting from the leaf-level and the whole-plant CO2 and water fluxes. The modeled growth rates at a soil salinity condition where B. gymnorrhiza is the dominant species (sal < 28 ‰) showed close values to the ones measured by Ohtsuka et al. (2019) at a B. gymnorrhiza-dominated site in the Fukido mangrove forest (Fig. S4 in the Supplement). This suggests that the model also reasonably predicted the growth rate of each species in addition to the leaf-level CO2 and water fluxes.

This model also showed reasonable reproducibility of the self-thinning process arising from tree competition. This was inferred from the decrease in tree density with the increase in individual tree biomass patterns based on the agreement of the measured and modeled tree density-mean MS relationship, except for those with soil salinity > 30 ‰ (Fig. 9). An exponent value close to -3/2 was obtained, similar to what is observed in the Fukido mangrove forest (Suwa et al., 2021; Fig. 9) and in many mangrove forests as well (e.g., Analuddin et al., 2009; Deshar et al., 2012; Khan et al., 2013; Tabuchi et al., 2013; Azman et al. 2021). This was achieved by implementing the species-specific morphological traits especially the DBH–Dcrown∗ relationship (Fig. 3b; see also Note S1 and Fig. S1). The underestimation trend of modeled tree density at high soil salinity (> 30 ‰) where R. stylosa starts to dominate (Fig. 9) may be attributed to the inaccurate representation of the crown morphological trait of this species, which generally gives larger Dcrown∗ compared to observed values (see Note S1 and Fig. S1c). Basically, the crown diameters of individuals determine the tree accommodation spaces, and therefore the overestimated crown diameter may have resulted in the underestimation of the tree density. Crown size representation could be a factor that drives a large part of the uncertainty in DVMs (Meunier et al., 2021). Nevertheless, the data are remarkably scarce in the case of mangroves. The morphological traits of crown size should be investigated in future studies for a more realistic representation of mangroves' tree competition and forest dynamics in the model. It is also important to note that some variables of model prediction such as LAI, shoot / root biomass ratio, morphological plasticity in accordance to changes in environmental variables (as shown in Fig. S7 in the Supplement), and leaf water potential dynamics have not been validated due to lack of data, and future research is needed to address these aspects.

Overall, this is the first modeling study to introduce detailed physiological and mechanistic representations of the mangrove forest growth controlled by photosynthesis, water and nutrient (represented by DIN) uptake, and tree competition, and it is a realistic or accurate reproduction of mangrove growth processes. The remarkable agreement of modeled forest structures with field data across a soil salinity gradient validated our hypothesis – individual-based DVM incorporating plant hydraulic traits can reasonably predict mangrove growth processes under salt stress without empirical expression of the soil salinity influence on mangrove productivity. However, the model still does not account for the plant-to-soil feedback through water uptake, which has been identified by a mangrove-growth–groundwater-flow coupled model (Bathmann et al., 2021) as an important factor affecting both mangrove and substrate conditions (soil salinity). Alternatively, the said model also demonstrated that the forest structural variable and soil salinity dynamics could reach steady states after some time from the initial condition, a setting that is considered to describe the Fukido mangrove forest (Ohtsuka et al., 2019). Our modeling results, which did not include the plant-to-soil feedback, therefore may be valid only for the steady states and still hold uncertainty in the developmental stage. This further implies that model application may be limited only to mature mangrove forests, and further model improvement is needed for its application to forests during the developmental stage (after plantation) or during the recovery stage (after disturbances such as typhoons and deforestation).

Overall, the model explained that the changes in mean DBH and AGB of the two coexisting species with change in soil salinity are due to the difference in their salt tolerance and interspecific competition (Figs. 7 and 8). While the model predicted the contrasting changes of AGB of the two species, both species showed a decrease in productivity with an increase in soil salinity as seen in the monotonic decrease in DBH (Figs. 7 and 8). This decrease in productivity can be partly explained by the downregulation of stomatal conductance under high-soil-salinity conditions (Fig. 5). In addition, the changes in the biomass allocation pattern that increased the allocation to the stem and roots relative to leaves with an increase in soil salinity have influenced productivity (Fig. S7) – a pattern that reduced the whole-plant photosynthesis and transpiration and increased the carbon (through the stem and root respiration and root turnover) and nitrogen (through the root turnover) cost relative to the unit leaf area. It should be noted that such morphological plasticity with changes in soil salinity predicted by the model qualitatively agrees with the implications by other studies (e.g., Suwa et al., 2008, 2009; Vovides et al., 2014; Nguyen et al., 2015; Chatting et al., 2020), but future studies are needed for quantitative and systematic validation.

The sensitivity analysis of the plant hydraulics trait parameters provided some insights into the different salt tolerance of the two species that shaped the forest structures along the soil salinity gradient (Fig. S6). For example, it showed the substantial contribution of the partial salt uptake of R. stylosa, represented by the lower ε, to the salt tolerance of this species (Fig. S6a) at the possible expense of higher P50 value (Table 2, Fig. S6c and d), which is considered as the coordinated functional traits (Jiang et al., 2017). The model showed the highest sensitivity to the parameters Ψlk and Ψl,min (Fig. S6e and f), suggesting that the mangroves' capacity to reduce the leaf water potential is one of the most important functional traits characterizing their salt tolerance, as suggested by Reef and Lovelock (2015). The response of AGB to changes in Ψlk, a parameter controlling biomass allocation pattern, also indicates the substantial impact of biomass allocation dynamics influenced by salinity on plant productivity. On the other hand, the model showed minimal sensitivity to β0 (Fig. S6g and h). While the higher stomatal conductance of R. stylosa than B. gymnorrhiza (as shown in Figs. 4 and 5) qualitatively agreed with the implications by Clough and Sim (1989) and Reef and Lovelock (2015), the model results suggested that the choice of -0.6 for β0 already leads to efficient stomatal openings for photosynthesis compared to the case of -0.4 for β0 (Table 2). This may explain the small variations in the simulated leaf-level photosynthetic rates between the two species and among the different soil salinity levels (Figs. 4 and 5). Understanding the mangroves' stomatal behavior relative to soil salinity and covariation of leaf water potential and photosynthesis have not been well established from field data (Perri et al., 2019). Further field-based research and data implementation to the model are needed for better and more reliable representation of mangroves' stomatal conductance and associated regulation of photosynthesis under salt stress.

The model specifically predicted that B. gymnorrhiza competes over R. stylosa when soil salinity is favorably low for the growth of B. gymnorrhiza (sal < 28 ‰), an observation that is consistent with our field data and the data from other mangrove forests (Putz and Chan, 1986; Enoki et al., 2014). This result may be attributed to the following model parameter settings based on literature – higher wood density (ρ), smaller specific leaf area (SLA), and higher leaf turnover rate (TOl) of R. stylosa than B. gymnorrhiza (Table 2). Higher ρ indicates the requirement of higher biomass increase for the height or radial growth of the stem. Smaller SLA and higher TOl indicate the higher requirement of C and N to produce new leaf tissues or to keep the same amount of leaves, i.e., the need of R. stylosa for more C and N resources for growth compared to B. gymnorrhiza. The biomass requirement of prop roots, which lowers the biomass allocation to the stem (Fig. S3), and the smaller Dcrown∗ of R. stylosa compared to B. gymnorrhiza (Fig. S1c and d) may also have contributed to the former's lower growth rate. Consequently, B. gymnorrhiza grew faster and suppressed the growth of R. stylosa by severe shading (Figs. 6 and 7). The higher growth rate of B. gymnorrhiza compared to R. stylosa at relatively low-salinity conditions agrees with the study by Jiang et al. (2019).

Interestingly, our model was able to simulate unique conditions not previously reported by other modeling works. For instance, the model predicted that R. stylosa trees could grow until the mature conditions under the canopy of B. gymnorrhiza-dominated forest provided the chance of favorable light conditions, resulting in the high mean DBH but low AGB of this species at relatively low soil salinity (∼ 24 ‰) (Figs. 6 and 7). Simulating this kind of process may only be possible through the individual-based approach with calculations of detailed irradiance distribution as done by the SEIB-DGVM in this study. Alternatively, the model predicted the significantly lowered growth rate of B. gymnorrhiza at high-soil-salinity conditions (sal > 30 ‰) where B. gymnorrhiza cannot grow until mature conditions, which resulted in the low AGB and small mean DBH of this species. This reduced the suppression of B. gymnorrhiza on R. stylosa and generated the Rhizophora stylosa-dominated forest (Figs. 6 and 7). Despite the abundant population of R. stylosa, the sizes of individuals were relatively small due to high salt stress and resulted in the high AGB but small mean DBH of this species.

Besides soil salinity, this study highlighted the importance of atmospheric variables as important drivers controlling mangrove production. This is seen in the photosynthesis–transpiration seasonal dynamics with peak during summer (June–September) and depression during winter (November–March) (Fig. 4) that none of the previous mangrove modeling studies has examined yet. The model predicted winter depression primarily due to low solar radiation and air temperature. Specifically, low air temperature (< 20 ∘C) significantly reduced photosynthetic capacity – the maximum carboxylation rate and the maximum electron transport rate (Aspinwall et al., 2021); this, in turn, decreased the marginal WUE (ΔAn/ΔE), leading to the downregulation of stomatal conductance, a behavior of mangroves' stomata observed under low-temperature conditions (Akaji et al., 2019; Aspinwall et al., 2021). This resulted in the depression of photosynthesis and transpiration during this season. The significance of atmospheric control on stomatal conductance and associated dynamics in winter compared to salinity control is also highlighted in the similar magnitude of reduction of the leaf water potential at midday between the different soil salinity conditions in this season compared to summer (Fig. 5c and f). Such winter depression lowers the production of mangroves in subtropical regions and may be differentiated from tropical mangroves in terms of productivity. This could be a key factor in explaining and predicting the latitudinal gradients in mangroves' structural variables such as canopy height and AGB with the highest values at the equatorial region (Saenger and Snedaker, 1993; Simard et al., 2019; Rovai et al., 2021).

The model gave significantly better prediction of the AGB spatial distribution when the spatially averaged DIN concentrations were applied to the substrate condition compared to plot-wise DIN values (Fig. S5, plot-wise simulation). This suggests that N availability was better represented by the spatially averaged value in this study. Porewater DIN in mangrove forests is highly heterogeneous horizontally (Inoue et al., 2011) and vertically (Kristensen et al., 1998; Lee et al., 2008) even in very small scales such as 10 cm. The DIN measured from one soil core sample might not have captured the representative value at each plot due to such heterogeneity. Differences in the predicted AGB between the two cases highlight nutrient availability in affecting mangrove production and biomass dynamics in this forest. Therefore, an appropriate representation of nutrient availability is critical for accurate prediction of mangrove production. More detailed measurement of porewater nutrient concentrations in space and time is needed for a more reliable model prediction, and future works will account for this aspect. Similarly, future works should consider biogeochemical processes which control nutrient dynamics in the substrate. For example, the porewater of the Fukido mangrove forest is rich in ammonia compared to nitrate (Table S1), contrary to the groundwater flowing into this forest, which is rich in nitrate (Mori et al., unpublished data). This suggests that biogeochemical processes, such as mineralization of organic matter, N fixation, and denitrification (Reef et al., 2010), are important drivers controlling nutrient dynamics in the forest, which ultimately affects soil organic matter dynamics. These factors should therefore be taken into consideration in future works as one of the plant-to-soil feedbacks in addition to water uptake processes.