Landcover map for the central region of the Yukon-Kuskokwim Delta, Alaska

Climate change is causing an intensification in tundra fires across the Arctic, including the unprecedented 2015 fires in the Yukon-Kuskokwim (YK) Delta. The YK Delta contains extensive surface waters (∼33% cover) and significant quantities of organic carbon, much of which is stored in vulnerable permafrost. Inland aquatic ecosystems act as hot-spots for landscape CO2 and CH4 emissions and likely represent a significant component of the Arctic carbon balance, yet aquatic fluxes of CO2 and CH4 are also some of the most uncertain. We measured dissolved CH4 and CO2 concentrations (n = 364), in surface waters from different types of waterbodies during summers from 2016 to 2019. We used Sentinel-2 multispectral imagery to classify landcover types and area burned in contributing watersheds. We develop a model using machine learning to assess how waterbody properties (size, shape, and landscape properties), environmental conditions (O2, temperature), and surface water chemistry (dissolved organic carbon composition, nutrient concentrations) help predict in situ observations of CH4 and CO2 concentrations across deltaic waterbodies. CO2 concentrations were negatively related to waterbody size and positively related to waterbody edge effects. CH4 concentrations were primarily related to organic matter quantity and composition. Waterbodies in burned watersheds appeared to be less carbon limited and had longer soil water residence times than in unburned watersheds. Our results illustrate the importance of small lakes for regional carbon emissions and demonstrate the need for a mechanistic understanding of the drivers of greenhouse gasses in small waterbodies.

Methods
This landcover classification was created for the purposes of determining watershed landcover as potential drivers of downstream waterbody CH4 and CO2 concentrations. The region of interest is a watershed in the central portion of the Yukon-Kuskokwim Delta, Alaska, where field observations were based. The landcover map has been clipped to the watershed extent, and included as a shapefile.

We created a 10-m resolution landcover map for the region of interest to determine the presence and abundance of various terrestrial, wetland, surface waterbodies, and disturbed areas in sample watersheds. We used an unsupervised k-means algorithm (Google Earth Engine, “wekaKMeans”) with the surface reflectance raw bands, derived bands (NDWI, NDVI), slope, and elevation as inputs for the classification. The Alaska Interagency Coordination Center historical wildfire database was used for wildfire delineations. Wildfires in the region of interest included fire scars from the 1970s, 1990s, and early 2000s, collectively designated as “old fires,” and fire scars from the large area burned in 2015. First, the region of interest was divided into unburned, old fire scars, and 2015 fire scars, and the classification algorithm was run separately for each. We used an initial number of classes “k” higher than the number of known landcover types in order to capture the variability in the driving layers, then later grouped similar classes produced by the k-means algorithm.

Landcover accuracy was assessed using 350 randomly stratified points from the region of interest. The classifications at these points were compared to higher resolution (Worldview-2) imagery using Google Earth Engine and reclassified using expert assessment. We used a confusion matrix to assess the balanced accuracy of each classification, which ranged from 0.75 to 0.99 (Figure S2 in Supporting Information S1 from Ludwig et al. 2022 (the article associated with this dataset)).