Super-Resolution Surface Water Mapping on the Canadian Shield Using Planet CubeSat Images and a Generative Adversarial Network

摘要

The Canadian Shield, the world’s largest exposure of glaciated crystalline bedrock, is themost lake-rich region on Earth. Recent studies using high-resolution CubeSat satelliteimagery have revealed its surface water hydrology to be surprisingly dynamic at fine spatialscales. Here we test whether super-resolution (SR), the resampling of coarse imagery to afiner-than-native resolution, can detect such changes. We degrade high-resolution PlanetCubeSat images of the Shield, then resample the coarsened imagery back to its native resolutionusing both traditional cubic resampling and a generative adversarial network, a typeof neural network often used for SR. To test classification accuracy from the generated SRimagery, we apply the same water classification to both resampling methods and find similarperformance based on confusion matrices with the control case of high-resolutionimagery. Next, we compare fine-scale shoreline mapping in SR imagery, cubic resampling,and in-situ field surveys. SR shorelines outperform those from cubic resampling, with anincrease in the modified kappa coefficient from -0.070 to 0.073. Potential applicationsinclude improved mapping of Shield lakes and retroactive application of SR to coarser-resolutionsatellite datasets to infer historical changes in fine-scale surface water dynamics.