A new synergistic approach for Sentinel-1 and PALSAR-2 in a machine learning framework to predict aboveground biomass of a dense mangrove forest

Mangroves are well-recognized for their very high carbon sequestration potential. However, studies on their role in global carbon cycling and climate change are hindered due to lack of operational methodologies to model and map their biophysical properties. This study establishes a robust methodological protocol for aboveground biomass (AGB) estimation using i) field measurements, ii) a generic (in the absence of species-specific) allometric equation, iii) multi-sensor/polarization SAR data and derived variables thereof, and iv) machine learning models; that demonstrated high prediction accuracy (R-2 = 0.93) and low uncertainty (mean <= 3% and median <= 1.5%) for a mangrove forest. Following stratified random sampling and on-field accessibility criteria, we laid out 314 elementary sampling points of 0.04 ha each at Bhitarkanika wildlife sanctuary (BWS), India and measured circumference at breast height (CBH) and canopy tree height for 18 species. The estimated AGB range of a generic allometric equation was 9-474 Mg/ha for BWS, with a major representation of 9-347 Mg/ha. We utilized Sentinel-1 and ALOS-2/PALSAR-2 and derived their variables for AGB prediction. Compared to single sensor-based model, we observed higher prediction accuracy for combined sensor data (R-2 = 0.63, 0.87, 0.93; RMSE = 66.75, 39.95, 28.35 Mg/ha; MAE = 52.63, 24.21, 19.55 Mg/ha; and Bias = 3.42, 0.22, 2.15 Mg/ha for C, L and C & L bands respectively using a Generalized Additive Model (GAM) over Random Forest (RF), Gradient Boosting Machines (GBM) and Support Vector Regression (SVR). The higher uncertainty pixels represented seasonal grassland and scrubs in the swampy areas and along the fringes of the creeks that experience diurnal tidal fluctuations. This robust methodology can be replicated for AGB estimates in other mangrove ecosystems to meet the operational carbon accounting requirements of various entities.

Automated summary

Mangroves are important for their ability to sequester carbon, but the lack of operational methodologies to model and map their properties has hindered studies on their role in global carbon cycling and climate change. This study establishes a robust methodological protocol for estimating aboveground biomass using field measurements, allometric equations, SAR data, and machine learning models. The protocol demonstrated high prediction accuracy and low uncertainty for a mangrove forest.