Non-perennial streams are among the largest emitters of carbon fluxes per unit land area, yet their global contribution remains uncertain due to the challenges of characterizing their hydrological conditions, including inundation duration, controlling rates of sediment respiration. Observation of intermittent hydrology in narrow channels can benefit from the finer spatial and temporal resolution of recent commercial satellite imagery. Yet, characterizing intermittent headwater hydrology at the catchment scale remains a challenge to benchmark regional land surface models and upscale biogeochemical rates.
Here, we combine commercial satellite imagery and in situ wildlife cameras with clustering and segmentation models to generate near-daily water cover over the intermittent channels in the Satus Creek watershed located within the Yakima River Basin. To construct a training set of water presence, we tasked Umbra satellites to acquire sub-meter X-band Synthetic Aperture Radar (SAR) image resolution over wildlife camera locations targeting seasonal wet-dry transition periods across the watershed. Umbra images and manually interpreted water masks were paired with near-daily multispectral imagery from PlanetScope (3-5m resolution) to estimate water cover fraction per stream reach using the WaterDetect algorithm.
The resulting near-daily water masks over 2022-2025 were compared to channel ponded depth simulated with the Advanced Terrestrial Simulator (ATS) process-based model for individual 100m-long reach. We found variable agreement between the data sources across the catchment, with persistent errors over narrow headwater channels, overhanging vegetation canopy cover, shadows cast over the channel as well as over braided channels. We also evaluated the catchment-wide connectivity of the river network from the model and imagery to characterize how intermittency affects flow of water and matter in the catchment. This work represents an initial assessment of the lower detection limit of meter-scale imagery over narrow stream channels, and further steps will explore methods fusing optical and SAR commercial imagery to refine the water detection.