Developing High Spatiotemporal Resolution Inundation Maps to Detect Rapid Changes in Surface Hydrology and Methane Emissions across Ecosystem Gradients in Alaska
Abstract
High latitude wetlands contribute a considerable amount of methane emissions to the global greenhouse gas budget. In tundra and boreal wetlands, ponds (open water area < 0.01 km2) can sustain higher methane production than larger lakes and dry, upland landscapes. Fine scale (< 10 m) delineations of surface water are of great importance for identifying potential hotspots and drivers of methane fluxes. To date, most water mapping efforts have relied on information from Landsat (30 m), MODIS (250 - 500 m), and Sentinel-2 (10 m). The resolutions of these imagery products are too coarse to identify a large fraction of smaller water bodies that tend to be higher emitters of methane, and the satellite platforms often lack the overpass frequency to track meaningful changes in surface water across an individual season. This research leverages a time series of PlanetScope Dove-R and SuperDove imagery (3 m) to track changes in lake and pond extent throughout the snow-free season over four locations in Alaska's boreal interior and coastal Yukon-Kuskokwim Delta. We used a convolutional neural network optimized for image segmentation (U-Net) to map open, standing water in the PlanetScope imagery. The model was trained on a dataset of 11,000 hand-delineated water bodies. Development of the training dataset relied heavily on Maxar WorldView imagery (2 m) to reference precise outlines of open water bodies and very small water features (< 100 m2). The trained U-Net model produces maps of lakes and ponds with high accuracy (95% F1-score, 91% IOU) and was used to generate water masks from 8,062 PlanetScope images. Using these water masks, we developed monthly composite water maps that represent a conservative estimate of seasonal variability in water body extent for 2019 - 2021. We compared the composite water maps with MODIS and Landsat-based surface water products and found that the coarse-resolution products considerably under-estimate total water body area. In the PlanetScope water maps, ponds made up 5 - 25% of total water body area and accounted for 18 - 60% of total surface water variability in the four regions of interest. Through a simple upscaling calculation, we found that these ponds could contribute a non-negligible portion of methane to the total flux from water bodies in these regions, emphasizing the need for high-resolution water maps.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMIN43A..01M