Automating the Observation-Model Feedback Loop for River Sediment Respiration at the Continental Scale

Microbial respiration in river sediment is a key indicator of ecosystem functioning and the biogeochemical fluxes across this critical zone link surface and subsurface waters. As such, there is tremendous interest in measuring and mapping sediment respiration rates. However, sediment respiration observations are expensive and labor intensive; there is limited data available to the community. We address this challenge by implementing a workflow that encapsulates an observation-model feedback loop. Observations are used to train a machine learning (ML) model to predict sediment respiration. The ML model then predicts respiration based on a much larger, publicly available data set that spans the contiguous United States. The sites with the greatest predicted uncertainty are then identified as priority sites for the next round of fieldwork whose new observations, in turn, are used to train a new version of the ML model. The data and ML models used in this feedback loop are publicly available in near real time facilitating coordination across a diverse community of stakeholders. We show examples of how the predictions of sediment respiration and priority sampling sites evolve with iterations of this observation-model feedback loop. Finally, major elements of this workflow are modular and cloud-based thus making this implementation a potential template for observation-model feedback loops for other applications.