Quantifying net loss of global mangrove carbon stocks from 20 years of land cover change

The paper addresses how two decades of land cover change have affected the net global mangrove carbon stock. Despite high carbon densities in mangroves and known deforestation, prior global estimates could not quantify net change because they lacked data on gains from forestation and on the proportion of carbon lost after conversion. The authors combine new datasets on mangrove extent (1996–2016), carbon densities, and temporal dynamics of carbon loss after deforestation and accumulation during forestation to estimate the net change in global mangrove carbon stocks. The study is important for accurate carbon accounting and for evaluating the effectiveness of conservation and restoration efforts.

Previous global studies estimated potential emissions by overlaying carbon stock maps with areas of mangrove loss, effectively quantifying carbon at risk (D) but not net changes due to the absence of data on forestation and proportional carbon loss after conversion. Literature shows mangroves have experienced extensive deforestation over recent decades, with mixed outcomes for restoration programs. Mangrove carbon densities are high, but accumulation to mature stocks can take decades. Not all carbon is immediately or fully lost after deforestation; some biomass and soil carbon persists, and losses proceed over different timescales. Recent syntheses provide estimates of proportional loss after land cover change and of carbon accumulation following restoration of blue carbon ecosystems, enabling improved net change estimation. Additional literature documents spatial variability in mangrove change, natural poleward expansion into saltmarshes, and contributions of anthropogenic restoration, as well as broader context on global emissions and comparisons with other ecosystems.

Study period 1996–2016. The authors estimated net changes in soil organic carbon (to 1 m) and above- and below-ground living biomass carbon attributable to mangrove-related land cover change (LCC). They did not estimate sequestration fluxes. Core quantities: D (carbon at risk from deforestation), F (maximum potential carbon stock gain from forestation), r_t (proportion of carbon lost as a function of time since deforestation), and a_t (proportion of reference carbon accumulated as a function of time since forestation). Net change was computed as D_r_t − F_a_t, after accounting for temporal loss and gain dynamics. - Mangrove extent and change: Used Global Mangrove Watch (GMW) products for 1996 and 2016 (with intermediate dates 2007, 2010) to identify patches of mangrove loss (deforestation) and gain (forestation). Areas were computed under an equal-area projection (Eckert VI) for polygons; other analyses used WGS84. Changes reversed within the study window were not captured. - Classification uncertainty: Incorporated false-positive misclassification rates at the pixel level based on GMW 2010 accuracy, simulating whether pixels of recorded gain or loss were true changes. False negatives could not be incorporated; uncertainty in extent for 1996 total stocks was not propagated. As a result, area change statistics are reported as medians without confidence intervals. - Carbon stock density: Extracted per-hectare soil organic carbon (Sanderman et al. 2018) and biomass carbon (Hutchison et al. 2014) at 0.05° where possible, else 0.5°, filling gaps with global means. Uncertainty was modeled as normal with means from maps and SD equal to reported RMSEs (soil 109 Mg ha−1; aboveground biomass 104.1 Mg ha−1; proxy for total biomass uncertainties). Negative simulated densities were truncated at zero. - Temporal carbon loss after deforestation: Modeled r_t separately for soil and biomass using meta-analytic relationships; biomass loss dynamics were proxied from temporal change in tree diameter where biomass-specific meta-analysis was lacking. Dates of deforestation were bracketed using presence/absence across 1996, 2007, 2010, 2016; an exact date was sampled uniformly within the bracketing interval. Regression uncertainty for r_t relationships was propagated. - Temporal carbon accumulation after forestation: Modeled a_t using a meta-analysis for blue carbon ecosystem restoration to estimate whole-ecosystem accumulation toward reference stocks. Dates of forestation were similarly bracketed and sampled uniformly; regression uncertainty was propagated. A sensitivity analysis using mangrove-specific case studies assessed potential impacts. - Simulation and uncertainty propagation: A bootstrap approach with 1000 replications jointly simulated uncertainties in area change (false positives), carbon densities, event dates, and regression uncertainties for r_t and a_t, yielding median estimates and 95% confidence intervals for D, F, D_r_t, F_a_t, and net change per patch and globally. Spatial aggregation for mapping used 2° (global) and 1° (regional) grids by summing patch-level values within cell centroids. - Sensitivity analyses: Four analyses focused on Southeast Asia to evaluate methodological choices, including the use of blue carbon restoration accumulation curves and replacement land-use categorizations where available.

- Mangrove extent: 142,865 km² in 1996 and 136,717 km² in 2016. Total deforested area 1996–2016: 8050.4 km²; forestation: 2243.3 km²; net loss: 5807.2 km² (4.0% of 1996 area). Annualized net loss rate ~0.2% yr−1; deforestation rate ~0.3% yr−1, with highest change rates between 2007–2010. - Carbon stocks: 1996 global mangrove carbon stock estimated at 8627.6 Mt (95% CI 4274.6–13,767.0 Mt). - Carbon at risk from deforestation (D): 462.5 Mt (95% CI 209.7–758.6 Mt). - Potential gain from forestation (F): 118.2 Mt (95% CI 47.1–200.8 Mt). - Net loss assuming instantaneous, full loss/gain (D − F): 344.3 Mt (95% CI 9.0–711.6 Mt). - After accounting for temporal loss and accumulation: Loss from deforestation D_r_t = 232.6 Mt (95% CI 24.0–537.0 Mt); gain from forestation F_a_t = 74.2 Mt (95% CI 11.0–180.8 Mt). - Net change (D_r_t − F_a_t): 158.4 Mt (95% CI −156.8 to 525.9 Mt), equivalent to a 1.8% decline relative to 1996 stocks. The negative lower CI indicates a possible net gain over the period. - Relative to D, the improved net estimate (D_r_t − F_a_t) is 34.2% of D, indicating substantial overestimation if using D alone. - Spatial patterns: Large net losses in the Caribbean and Southeast Asia, especially Borneo and Papua; localized gains or low losses in parts of Africa, South Asia, Central America. Notable net gains in parts of Mexico and Myanmar; low net loss in southern China, potentially reflecting conservation and replanting. - CO2-equivalent context: If net stock loss were fully emitted as CO2, total 1996–2016 emissions would be 580.4 Mt CO2 (95% CI −574.6 to 1927.0 Mt), about 0.6% of emissions from global land cover change and <0.1% of total global CO2 emissions over the same period.

Accounting for both mangrove forestation and the proportional, time-varying nature of carbon loss after deforestation drastically reduces estimated net losses relative to potential loss metrics. Natural drivers (e.g., poleward expansion into saltmarshes, regeneration on abandoned lands, colonization of newly suitable substrates) and anthropogenic restoration both contributed to gains, though the study cannot disentangle their relative shares. Despite mixed restoration success globally, targeted, ecologically informed projects and conservation efforts appear to have mitigated losses in some regions (e.g., southern China). The overall low net change reflects relatively modest deforestation rates during 1996–2016 combined with notable forestation, suggesting partial conservation success. The authors caution against overstating mangrove blue carbon’s climate mitigation role: although carbon densities are high, the global area is small, so even the upper-bound emissions represent a small fraction of global totals. Nevertheless, mangroves’ high per-area carbon value and co-benefits (coastal protection, fisheries support, recreation) strengthen the case for continued conservation and restoration. Considerable uncertainty exists due to data synthesis across global products, but confidence intervals are not large enough to overturn the central finding of low net loss—and potentially no net loss—of mangrove carbon stocks over the study period.

Using globally consistent datasets and a simulation framework that incorporates carbon loss and accumulation dynamics, the study estimates a small net decline (about 1.8%) in global mangrove carbon stocks from 1996 to 2016, with the possibility of no net loss. The results demonstrate that accounting for forestation and partial, temporal loss processes is critical; potential-loss approaches overestimate impacts. Conservation and restoration efforts, combined with natural expansion, likely contributed to muted net losses. Future work should: (1) improve discrimination between natural expansion and anthropogenic restoration contributions; (2) incorporate additional carbon pools (inorganic sediments, dead biomass) and potential biases in carbon density maps; (3) quantify degradation impacts and false-negative mapping errors globally; (4) consider pre-existing carbon stocks and flux changes when estimating net emissions from LCC; (5) refine temporal histories of mangrove change to identify turning points in global deforestation trends and the policies that drove them.

- Carbon pools: Analysis includes soil organic carbon to 1 m and living biomass; excludes inorganic sediment carbon and dead biomass, which may comprise ~25% of total mangrove carbon. - Degradation: Small-scale degradation (e.g., timber extraction, pollution) not quantified globally due to remote sensing limitations. - Data biases: Global carbon density maps may be biased toward high-quality sites; if actual stocks are lower or forests degrade, losses/gains could be misestimated. Soil organic carbon may have been overestimated in past literature due to conversion factors; sensitivity analysis suggests potentially 35% lower net loss estimates in Southeast Asia after correction. - Mapping uncertainty: Only false-positive classification errors from GMW were modeled; false negatives were not incorporated. Confidence intervals for area changes were not reported. Changes reversed within the study period are not captured. - Temporal inference: Dates of LCC events are bracketed by discrete observation years and sampled uniformly within intervals, introducing temporal uncertainty. - Accumulation model: Used a general blue carbon restoration meta-analysis for accumulation (a_t) due to lack of mangrove-specific global meta-analysis; tested via sensitivity analysis with limited case studies. - Scope: Sequestration fluxes were not estimated; pre-mangrove carbon stocks in replaced ecosystems and changes in carbon fluxes post-LCC were not included, so results pertain to stock changes rather than full net emissions. - Regional sensitivity analyses were limited to Southeast Asia due to data availability on replacement land uses.