Improving national peat depth inventories with terrain-based digital soil mapping: evidence from Norway

Accurate mapping of peat depth is crucial for carbon accounting, areal planning, and land management in peatlands. However, existing maps often lack the resolution and accuracy needed for these purposes, even in countries with rich spatial data sets.

We present a study that evaluated whether digital soil mapping using remotely sensed data could improve existing maps of peat depth across two, >10 km² sites in western and southeastern Norway. We measured peat depth by probing and ground-penetrating radar at 372 and 1878 locations at the two sites, respectively. Then we trained Random Forest models using radiometric and terrain variables, plus the national map of peat depth, to predict peat depth at 10 m resolution.

The two best models achieved mean absolute errors of 60 and 56 cm, explaining one-third of the variation in peat depth. Our remote sensing models had better accuracy than the national map of peat depth, even when we calibrated the national map to the same depth data. Terrain variables were much better predictors than radiometric variables, with elevation and valley bottom flatness showing the strongest relationships to depth. At our coastal site, peats were much deeper above the Holocene marine limit than below, emphasizing the importance of accumulation time in places that have experienced glacial isostatic adjustment. Meanwhile, the national map of peat depth itself carried much more information about peat depth at one of the sites than the other --- likely as a result of uneven historical field sampling. 

Based on these findings, we conclude that digital soil mapping with DTM-derived predictors can improve the existing, national map of peat depth in Norway. Doing so would support national and regional-scale peatland carbon stock assessments and land management policies, as well as specific areal planning decisions at the municipal scale. Since our remote-sensing models relied on predictor--depth relationships that were specific to the sites we mapped, more depth measurements would be needed to expand the spatial coverage of an improved national map. A structured surveying effort coordinated in the manner of, or in conjunction with, the National Forest Inventory would be an efficient way to collect these data. Better data infrastructure for hosting and compiling peatland parameters from opportunistic measurements would also accelerate the accumulation of accuracy improvements. Besides specific accuracy improvements, digital soil mapping of peatland also offers advantages in transparency, reproducibility, and updatability.