In the Arctic, waterbodies are abundant, and rapid thaw of permafrost is destabilizing the carbon cycle and changing hydrology. It is particularly important to quantify and accurately scale aquatic carbon emissions in arctic ecosystems. Recently available high-resolution remote sensing datasets capture the physical characteristics of arctic landscapes at unprecedented spatial resolution. We demonstrate how machine learning models can capitalize on these spatial datasets to greatly improve accuracy when scaling waterbody CO₂ and CH₄ fluxes across the Yukon-Kuskokwim (YK) Delta of south-west AK. These datasets include carbon dioxide and methane dissolved concentrations and diffusive fluxes from a research watershed in the central YK Delta. 

These maps are the results of training boosted regression tree models on observation datasets of waterbody dissolved CO₂ and CH₄ concentrations (https://arcticdata.io/catalog/view/doi%3A10.18739%2FA2N29P731 and https://arcticdata.io/catalog/view/doi%3A10.18739%2FA23775V7T). We characterized waterbody size, shape, and multispectral surface reflectance (Sentinel-2) using an object-based imagery analysis on a high-resolution map of waterbodies in the region. We split the study region into non-nested hydrologic units (sub-basin) in three sets of sizes based on flow accumulation and a high-resolution DEM. For each waterbody in the region, we calculated sub-basin averages of remote sensing imagery, including NDVI, NDWI, distance to nearest water, slope, multispectral surface reflectances (Sentinel-2), seasonal composites of C-band SAR (Sentinel-1). We use the waterbody and sub-basin variables as potential drivers to train scaling models, then applied the models to the research watershed to create maps of dissolved CO₂ and CH₄ in waterbodies. These were converted to fluxes using the average piston velocity from observations in the region (k600= 0.33 m/d).