High-resolution water table depth modelling of Danish peatlands for upscaling emission of greenhouse gases

Water table depth (WTD) modulates greenhouse gas (GHG) emissions from drained peatland soils and rewetting peatlands has been identified as a cost-effective mitigation measure to reduce emissions from the agricultural sector. However, detailed knowledge of the spatial variability of WTD is needed to guide the planning of rewetting measures as well as to upscale GHG emissions from peatlands for national inventories. In this study we developed a high-resolution (10 m) map of long-term mean summertime WTD for Danish peatlands (~9,000 km²) using a gradient boosting decision tree algorithm. The machine learning (ML) model was trained against more than 10,000 WTD observations as well as water levels in over 10,000 groundwater connected lakes and rivers. The WTD observations were transformed to better account for the non-linear relationship between WTD and GHG emissions and the limited WTD range (such as 0 – 50 below ground) in which GHG emissions are most sensitive. Over 20 high-resolution explanatory variables, many of which are satellite based, provided diverse information on topography, groundwater, moisture conditions, land-use and geology to the model. Cross validation was applied to evaluate the accuracy of the trained ML model with special focus on the shallow WTD (mean error= -8cm and mean absolute error = 18 cm). The horizontal and vertical distance to the nearest waterbody as well as organic content of the soil and land surface temperature were among the most important explanatory variables of the trained ML model. The WTD map was subsequently applied as input to two recently developed WTD-dependent GHG emission models to upscale GHG emissions from Danish peatlands. For this purpose, the mean summertime WTD map had to be corrected to represent mean annual conditions. Lastly, simple rewetting scenarios, i.e. decrease in WTD, were applied to elucidate the potentials of rewetting as mitigation measure.