Abstract:
The Sundarbans mangrove system constitutes a globally consequential blue carbon sink, yet vertically
resolved soil organic carbon (SOC) inventories, particularly at depth, remain under- constrained. This
study deployed a geospatial deep-learning pipeline integrating multi-source Earth observation data
(optical, SAR, and DEM-derived covariates) to spatially upscale SOC stocks to 1m depth across the
Bangladesh Sundarbans. Stratified coring across five sites positioned along the salinity-tidal inundation
gradient to represent four major vegetation zones (n=25; one pit per site × five depth intervals: 0-20,
20-40, 40-60, 60-80, 80-100cm) captured physicochemical and SOC attributions. A Kruskal–Wallis
test confirmed a positive SOC gradient with depth (H=21.7, df = 4, p<0.001); post-hoc Dunn's pairwise
comparisons showed 60–100cm layers differed significantly from all shallower layers
(p<0.01, Bonferroni-adjusted), with the dominant pool residing below 40cm, implying systematic
underestimation by conventional shallow-core (≤30–40 cm) assessments. To partition total ecosystem
carbon and establish soil-pool dominance, above and belowground biomass were parameterized using
species-specific allometric equations, revealing high stem density but comparatively low basal area
consistent with a regenerating stand structure dominated by Heritiera fomes and Sonneratia apetala.
Analysis of a previously published Landsat-derived mangrove record (1988–2022) revealed net
contraction in mangrove extent; loss-agent raster attribution, based on shoreline-proximity buffers,
land-cover transition sequences and cyclone-track coincidence, identified chronic tidal abrasion and
erosion as the primary driver rather than anthropogenic conversion or cyclonic disturbance.
Benchmarking against a previously published global30m machine-learning SOC product revealed a 4.2
fold underestimation of SOC stocks in the Bangladesh Sundarbans, particularly in peat-enriched, anoxic
substrata. Assimilation of deep-profile in situ observations with the pipeline significantly improves
spatial prediction fidelity and establishes a robust baseline for Measurement, Reporting, and
Verification for greenhouse gas inventories, highlighting the necessity of incorporating deep-soil carbon
pools for inventorizing erosion-vulnerable seaward margins.
