Mangrove reforestation provides greater blue carbon benefit than afforestation for mitigating global climate change

Literature search and screening

Our analysis included a systematic literature search and was conducted by following the PRISMA protocol⁵⁵ (Supplementary Fig. 7). We searched through Web of Science and China National Knowledge Infrastructure (CNKI) platforms by using keywords listed in Supplementary Table 3. A total of 3299 potentially relevant articles were found (Mandarin and English). The availability of peer-reviewed datasets associated with these published articles^{11,15,56,57,58,59} and online databases (The Sustainable Wetlands Adaptation and Mitigation Program (SWAMP) database, https://www2.cifor.org/swamp) were also considered. We then removed a significant number of articles through title screening, leaving 551 articles for further inspection.

For these remaining articles, we used a four-step critique process to screen their title, abstract, and full text. We determined that firstly, they must provide carbon density data for at least one of the four mangrove carbon pools (i.e., aboveground biomass, belowground biomass, sediment organic carbon, or total ecosystem carbon). Secondly, articles needed to state the forest age or the starting date of the restoration action. For those studies providing only age intervals (e.g., 10–25 years, >66 years), we excluded them from the analysis. Thirdly, a description of prior land use was required. From these, mangrove restoration could be divided into two categories—reforestation and afforestation—on whether mangroves previously existed in that location. For reforestation, the initial conditions for inclusion were: (1) abandoned agricultural/aquacultural sites built previously by excavating mangrove forests, (2) clear-felled mangrove lands after wars, timber harvest, and silvicultural management, and (3) mangrove forests with mortality due to spraying of defoliants and hydrological alteration caused by the construction of embankments. We compared the carbon densities of reforested mangroves among sites with different causes of degradation/deforestation, and no significant difference is found (Supplementary Fig. 9). For those reforested mangroves, we assumed they would be protected and conserved by local governments and non-government organizations, so that there will not be human-driven degradation or deforestation in the near future. However, we acknowledge that a fraction of mangrove reforestation is managed for wood production, which means logging would happen at a certain interval after reforestation at these sites. For these logging sites, we used their reported measurements after clear-cut, such as 0-, 5-, 10-, 15-, and 25-year post-harvest sites in Sundarbans, Bangladesh⁶⁰. On the other hand, the future occurrence of natural-driven deforestation (e.g., cyclones) is difficult to predict, and thus not considered in our study. For afforestation, the initial condition for inclusion was the presence of non-mangrove habitat immediately before afforestation began, such as mudflats, seagrass, saltmarsh, coral reef, or denuded areas. In most cases, reforestation and afforestation were undertaken through active planting without much re-engineering⁴, but for reforestation, natural regeneration could have, and in many places likely did, augment recruitment⁶¹. Moreover, we only considered mangrove succession that started from near-barren land with an insignificant amount of biomass, and introductions of exotic species to degraded areas with sparse trees were not incorporated. Lastly, if the forest age or prior land use type was not given, the articles needed to specify the location of sampling plots (latitude, longitude). With the coordinates matching, prior land use type and establishment dates were sometimes identifiable through remote sensing (Supplementary Fig. 10). For those articles sharing the same restoration sites but showing different aspects of the data collection, we combined the results and considered the collective work as one source. Based on the space-for-time method, data in the control sites before mangrove restoration actions were also collected as a paired site of restoration (e.g., abandoned ponds before mangrove reforestation; mudflats before mangrove afforestation). In total, we obtained data from 379 mangrove restoration sites described by 106 articles.

Data extraction

We extracted aboveground living biomass carbon (AGC), belowground living biomass carbon (BGC), sediment carbon (SCS), and total ecosystem carbon (TECS) density from the 106 original data sources. In most cases, numeric values were provided. For those data not provided numerically but graphed, we determined values from figures with the application of GetData Graph Digitizer (http://getdata-graph-digitizer.com/).

Among the articles, aboveground and belowground biomass (Mg ha⁻¹) data were obtained using either a harvesting method (empirical) or an allometric method (calculation). Aboveground biomass represented the sum of stem, leaf, and branch dry weight, and we included prop root biomass when Rhizophora spp. were present. For soil coring methods that determined belowground biomass or sediment carbon density, belowground biomass was considered the dry weight of living coarse and fine roots multiplied by the ratio of core area to land surface area⁶². For allometric methods, trunk diameter at breast height (DBH, ~1.3 m) and tree height were used to calculate aboveground and belowground biomass by species-specific or common allometric equations⁶³. These equations were also used to calculate the belowground biomass when articles provided plot information (DBH, height) but not belowground biomass (Supplementary Table 4). Total biomass was calculated as the sum of aboveground and belowground biomass. Deadwood and pneumatophore biomass were not included in our analysis; these data are rarely provided and/or methods of determination are inconsistent among global studies⁶⁴. Some articles provided total biomass and shoot/root biomass ratio (S/R), and in such cases, above- and belowground biomass data were obtained through calculation as follows:

For those articles measuring carbon content, study-specific carbon conversion factors were used to transform biomass to biomass carbon density (Mg C ha⁻¹). If carbon content data were not provided, we converted aboveground and belowground biomass to carbon density by applying a conversion of 0.47 and 0.39, respectively⁶⁵. The aboveground biomass carbon density was divided by its corresponding age to get the average aboveground biomass carbon accumulation rate (Mg C ha⁻¹ yr⁻¹).

For sediment carbon density (SCS, Mg C ha⁻¹), we selected the top 1 m because this depth equated to the most commonly reported depth and could reflect the impact of root mass input in the deeper depth⁶⁶, which is also consistent with recent blue carbon standing stock assessment guidance^64,67. Sediment carbon stock was calculated by multiplying sediment organic carbon content (SOC, %) by bulk density (BD, g cm⁻³), integrated over depth (cm). For studies that reported sediment carbon stock to <1 m depth, we assumed that its organic sediment layer was deeper than 1 m and that the carbon density of the unmeasured depth was the same as that of the deepest measured layer. For example, when the deepest measured sediment layer was 50–60 cm, the carbon density in the 60–100 cm layer was assumed to be the same as that in 50–60 cm, which might overestimate sediment carbon density slightly. Plus, we did not include studies that only measured surface sediment carbon (<20 cm) in our dataset. All biomass carbon data were summed to estimate TECS.

Climate factors (i.e., mean annual temperature, MAT, and mean annual precipitation, MAP) were extracted from the WorldClim2.0 dataset⁶⁸ (spatial resolution: 30 s, https://www.worldclim.org/data/worldclim21.html) using the longitude and latitude of each restoration site, and averaged within a 1 km buffer of each restoration site.

Influence of restoration pathways and climate factors on carbon accumulation

To examine the influence of restoration type on the different mangrove carbon density pools over time, a linear mixed model was used with restoration duration (age) and restoration type (reforestation, afforestation), and their interaction was denoted as a fixed factor as well as their restoration region as a random factor using the lme4 (version 1.1.29) and emmeans (version 1.7.5) packages of R (version 4.0.4, http://www.r-project.org/). Potential nonlinear growth patterns were modeled with a log transformation of age in comparison to no transformation using Akaike Information Criterion (AIC). Models with lower AIC showed better goodness of fit. The AIC results indicated that log transformation worked better to tease out carbon pool dynamics (Supplementary Tables 1, 2). Analysis of variance (ANOVA) was used to evaluate the significance of each variable at a significant level of 0.05.

Moreover, to assess the difference in recovery trajectories with more detail, we compared the carbon density between mangrove reforestation and afforestation at each stage (age) of mangrove stand development. Ages of sampling plots were divided into 5-year age classes, but were grouped for 20–40 years due to the limited sample size. In contrast to the biomass carbon pool, the sediment carbon pool has various sizes before restoration. To eliminate the influence of initial sediment carbon density, sediment carbon density increments after restoration were also used to compare the carbon accumulation trajectories between these two restoration pathways. Using the space-for-time approach, the difference in sediment carbon densities between the restoration site and the paired control site (without restoration) was considered as the carbon change induced by mangrove restoration. We acknowledge that some uncertainties exist in the space-for-time method because it is difficult to find a perfect control site, which may partly explain the varying, even negative, carbon stock increments in some age groups (Supplementary Fig. 2). To minimize the uncertainties, we corrected the negative values if the carbon accumulation rate was measured simultaneously by ²¹⁰Pb⁶⁰ and recalculated the carbon density increments via multiplying the carbon accumulation rate by its restoration duration. These carbon density increments were then compared between mangrove reforestation and afforestation among age classes.

Before comparison, normality and homogeneity of model residuals were tested using the Shapiro–Wilk and Levene tests. As assumptions were not met in some groups, the significance of differences was assessed with non-parametric tests. The Wilcoxon two-sided tests were used to compare the difference in carbon density or its increments (mean and standard error) between reforestation and afforestation action for each age group. Comparisons of climate factors and sediment properties were analyzed similarly to carbon density data using the Wilcoxon test. The Kruskal–Wallis test combined with the Bonferroni adjusted post hoc Dunn test was used to compare the differences of more than two groups (i.e., aboveground biomass carbon accumulation rates among the five age groups during mangrove reforestation; carbon densities among mangroves at sites with different causes of degradation/deforestation). All the differences were considered to be significant at a level of P ≤ 0.05. Finally, relationships among climate factors, sediment properties, and AGC as well as BGC/AGC were determined through the use of ordinary least squares (OLS) regression.

Plant and sediment carbon accumulation model

The plant growth rate is hypothesized to decrease with stand age and reach an equilibrium (or steady-state) during the latter stages of stand development^22,69. Therefore, plant biomass would exhibit a nonlinear increase in many cases. Thus, to determine the recovery trajectories of mangrove biomass carbon pools, we fit three nonlinear growth models (i.e., Von Bertalanffy model, Gompertz growth model, and Chapman–Richards model)⁷⁰ as follows:

Where Asym defines the maximum carbon density that a mangrove forest could theoretically reach along an age chronosequence, b and c determine the position and shape of the curve before reaching an asymptote.

When mimicking the sediment carbon pool and total ecosystem carbon pool, carbon density at initial condition was introduced in the growth model as a non-zero baseline value for year 0. Therefore, Chapman–Richards model was unsuitable for these two carbon pools as it defines zero carbon density at the starting year. Combining the model performances in all of our four carbon pools, we used the Gompertz growth model to mimic the carbon accumulation trajectories on both mangrove reforestation and afforestation (Supplementary Table 5 and Supplementary Fig. 11).

All model fitting, comparison, confidence interval calculations, bootstrapping, and integration were conducted using nlme (version 3.1.157), nlstools (version 2.0.0), car (version 3.0.13), stats (version 4.2.0), FSA (version 0.9.3) and rcompanion (version 2.4.18) packages, and visualization procedures were determined with ggplot2 (version 3.3.6) packages in R.

Global mangrove carbon sequestration potential from restoration action

Since mangrove reforestation action occurs where a mangrove community previously existed, we assumed that any mangrove area loss since 1996 provided same-area reforestation viability. We used the Global Mangrove Watch dataset to define the mangrove deforestation area between 1996 and 2016³⁸. Mangrove deforestation was mainly derived from coastal erosion, transformation into settlements, commodities production (e.g., aquacultural/agricultural plots), mangrove clear-cutting operations, or mortality from extreme climate events⁴⁰. The proportion of each deforestation caused by a country and its corresponding exclusive economic zone (EEZ)⁷¹ was calculated as the average estimate between 2000 and 2015 by ref. ⁴⁰. For those EEZs that had detectable mangrove loss but were not associated with an associated driver of loss, we used the average proportion of loss drivers in the corresponding geographical zone (Supplementary Table 6). For geographical zones lacking corresponding loss-driver data (i.e., East Asia including China and Japan), we used the average proportion of mangrove loss drivers in representative areas of China to represent East Asia^72,73. The overall feasible reforestation extent was then calculated as the sum of each kind of deforestation area in each EEZ. As a biophysical constraint, regions experiencing coastal erosion and settlement development were excluded from our analysis because no restoration possibility exists for those areas³⁹.

The CO₂-eq sequestration potential (Seq, Mg CO₂-eq ha⁻¹) under mangrove reforestation was calculated as the sum of carbon density increments of AGC, BGC, and SCS from the initial baseline (when age was 0), which are predicted by their corresponding growth models and confidence intervals. We used a period of 40 years (i.e., up to 2060) to assess carbon sequestration potential.

On the other hand, four scenarios with different recovery rates were used: (1) 1-year-completed restoration scenario: restoring all these deforested regions as quickly as possible so that all the restored mangroves could fix carbon for 40 years (Eq. (7)); (2) 5-year-averaged restoration scenario: following some short-term targets that countries have pledged, like Indonesia (rehabilitate all of the damaged mangroves (about 600,000 ha) during 2020–2024)⁷⁴ and China (restoring 18,800 ha during 2020–2025)⁷⁵. All of these mangrove area losses globally would be restored within 5 years. Restoration effort and project implementation rate in each country was assumed to be the same; therefore, mangrove restored in the first year could fix carbon for 40 years while those restored in the fifth year could fix carbon for 36 years (Eq. (8)); (3) 10-year-averaged restoration scenario: following median-term goals promoted by international forums and organizations, like COP 26 (halt and reverse forest loss and land degradation by 2030, https://ukcop26.org/) and Global Mangrove Alliance (increase global mangrove area by 20% by 2030, https://www.mangrovealliance.org/). All of these mangrove area losses globally would be restored within 10 years. The assumption used for restoring areas each year by country was similar to 5-year-averaged restoration scenario (Eq. (9)); (4) varying rates across countries scenario: assuming countries pledging or agreeing to support mangrove replanting projects on COP 26 would restore their deforested mangrove area within 10 years (i.e., by 2030), except for countries like Indonesia and China assumed to finish in 5 years with their officially-promulgated policies^74,75. Other countries were assumed to reforest their harvested or damaged mangroves within 20 years (Eq. (10)). We multiplied the CO₂-eq sequestration potential in certain time intervals by viable restoration area to indicate the maximum climate change-mitigative carbon storage benefit under each mangrove restoration scenario and pathway.

Where Area_2016-1996 represents the total feasible reforestation extent from 1996 to 2016; f_AGC, f_BGC, f_SCS represent the best-fit carbon accumulation model of aboveground biomass carbon, belowground biomass carbon, and sediment carbon pools for mangrove reforestation simulated above, respectively. L, M, and N were the number of countries with 5-year, 10-year, and 20-year restoration periods under assumptions listed in scenario 4. We also calculated the CO₂-eq sequestration potential for afforestation by assuming the same potential afforestation area as for reforestation with carbon accumulation models. All geoinformation processing was executed using QGIS (version 3.18.2, https://www.qgis.org/) and packages in R (dplyr (version 1.0.9), tidyr (version 1.2.0), rstatix (version 0.7.0), ggpmisc (version 0.4.6), ggpubr (version 0.4.0), ggrepel (version 0.9.1), ggalluvial (version 0.12.3), reshape2 (version 1.4.4), introdataviz (version 0.0.0.9003), sf (version 1.0.7), rnaturalearth (version 0.1.0), rnaturalearthdata (version 0.1.0), nlraa (version 1.2), and agricolae (version 1.3.5)).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Source: Ecology - nature.com