Predicting catchment-scale methane fluxes with multi-source remote sensing.

Context: Spatial patterns of CH₄ fluxes can be modeled with remotely sensed data representing land cover, soil moisture and topography. Spatially extensive CH₄ flux measurements conducted with portable analyzers have not been previously upscaled with remote sensing. Objectives: How well can the CH₄ fluxes be predicted with plot-based vegetation measures and remote sensing? How does the predictive skill of the model change when using different combinations of predictor variables? Methods: We measured CH₄ fluxes in 279 plots in a 12.4 km² peatland-forest-mosaic landscape in Pallas area, northern Finland in July 2019. We compared 20 different CH₄ flux maps produced with vegetation field data and remote sensing data including Sentinel-1, Sentinel-2 and digital terrain model (DTM). Results: The landscape acted as a net source of CH₄ (253–502 µg m⁻² h⁻¹) and the proportion of source areas varied considerably between maps (12–50%). The amount of explained variance was high in CH₄ regressions (59–76%, nRMSE 8–10%). Regressions including remote sensing predictors had better performance than regressions with plot-based vegetation predictors. The most important remote sensing predictors included VH-polarized Sentinel-1 features together with topographic wetness index and other DTM features. Spatial patterns were most accurately predicted when the landscape was divided into sinks and sources with remote sensing-based classifications, and the fluxes were modeled for sinks and sources separately. Conclusions: CH₄ fluxes can be predicted accurately with multi-source remote sensing in northern boreal peatland landscapes. High spatial resolution remote sensing-based maps constrain uncertainties related to CH₄ fluxes and their spatial patterns. [ABSTRACT FROM AUTHOR]