Abstract
Greenhouse gas (GHG) emissions originating from rewetted peat extraction sites are largely unknown. While the emissions can be quantified with chamber measurements, spatial upscaling is needed to quantify the emissions over the whole rewetted areas. Therefore, we developed a method to upscale chamber-measured GHG emissions with drone and satellite imagery. We measured GHG fluxes in five different surface types within three rewetted peatlands in northern Finland across two growing seasons. We spatially classified surface types within drone mapped areas with random forest classifier and further upscaled the surface type %-covers outside the drone areas with multitemporal Sentinel-2 imagery using extreme gradient boosting regression. With predicted surface type %-covers, we calculated total and area-normalised GHG flux sums for each measurement moment. The drone-based surface type classifiers performed well (class specific F-scores 0.28–0.97, overall accuracies 0.76–0.85) as did the satellite-based surface type %-cover models (R 2 = 0.38–0.99). Predicted surface type covers appeared realistic and followed topographical gradient observed from the high-resolution classification. Upscaled and area-normalised GHG fluxes ranged from 0.24 to 5.91 g CO2-C m –2 d –1 , 0.01 to 0.34 g CH4-C m –2 d –1 and -98 to 61 μg N2O-N m –2 d –1 depending on site and time. High water and reed %-cover reduced overall emissions due to higher emission factors in drier surface types but due to lack of gross primary production estimations, no conclusion could be drawn about net ecosystem carbon balances. Our results indicate that combination of field measurements, drone flights, and satellite imagery can be used to upscale surface area estimates and GHG fluxes.