Digital mapping of peat thickness and carbon stock of global peatlands

Abstract

Peatlands, occupying merely 5% of the Earth's land surface, are an important carbon sink, storing up to double the carbon of the world's forests. The quantification of global peatlands carbon stock and their spatial distribution, however, poses a significant challenge due to their heterogeneous nature and the complex hydroecological processes that govern their formation. Using the Global Peatland Map (GPM 2.0), this study employed a digital soil mapping approach to predict peat thickness, and multilayer bulk density (BD) and carbon content (CC) globally. We applied the Quantile Random Forest (QRF) algorithm, informed by land surface data (soil, climate, organisms, and topography), to develop regional models for peat thickness and global models for BD and CC. Peat thickness models, based on approximately 27,000 data points, demonstrated good predictive performance, with the highest accuracy observed in African peatlands (validation R2 = 0.61). In contrast, BD (∼19,000 points) and CC (∼9,000 points) models showed more variable performance across different soil layers (average R2 = 0.45 and R2 = 0.22, respectively). Feature importance analysis indicated that elevation and climate were key predictors, particularly in Latin America and South–Southeast Asia. Applying the models to 1 km resolution covariates across the world, our predicted peat thickness map aligned well with existing high-resolution regional maps. By incorporating error propagation rules, we estimated the global peatlands carbon stock to be 942 ± 312 Pg C over an area of 6.75 million km2. Our results, including detailed maps, are available to facilitate further global peatland analyses and modelling endeavours.