Sensor-based peat thickness mapping of a cultivated bog in Denmark

Draining peatlands for agriculture transforms them into significant carbon (C) sources. Restoring drained peatlands is increasingly recognized as a climate action strategy to reduce terrestrial greenhouse gas emissions. Restoration efforts often require accurate inputs, like peat thickness (PT), for C-stock estimation and monitoring; however, these are often lacking or available at suboptimal accuracy levels. In this study, apparent electrical conductivity (EC_a) from proximal electromagnetic induction (EMI) surveys and topographic variables derived from a LiDAR-based digital elevation model were assessed as covariates for PT mapping of an agricultural bog, separately and combined, using the quantile random forest algorithm. Local models were trained separately for the large (308 ha) and small (42 ha) EMI surveyed areas, while global models combined data from both areas for a full site analysis. The subsurface was characterized based on resistivity variations in inverted towed transient electromagnetic (tTEM) data. The results indicated that combining topographic and EC_a covariates yielded the best PT prediction accuracy for the global model, with a coefficient of determination of 0.61 and a normalized root mean square error (NRMSE) of 0.10. The best large area local model was less accurate than the former (NRMSE of 0.18), while the best small area local model (NRMSE of 0.11) was superior to the best global model. Models trained with only topographic or EC_a covariates were the least accurate, especially for the EC_a-only model. The tTEM results revealed a heterogenous site characterized by a thin, resistive peat layer overlying stratified postglacial deposits of clay, sand, and saline chalk. Our findings show that covariates characterizing surface and subsurface properties are essential for accurate PT mapping and can inform tailored land use planning and restoration initiatives for degraded peatlands.