Freshwater surfaces emit a substantial quantity of methane relative to their small surface area globally. Among the different types of freshwater methane emission fluxes, ebullition (bubble fluxes from organic-rich sediments to the waterbody surface) remains one of the most challenging to predict due to its temporal heterogeneity. As a result, upscaling methane ebullition emission rates from freshwater surfaces is difficult, prompting calls for new approaches to reduce this uncertainty. As a result, numerous models that predict ebullition have been developed but none to the best of our knowledge use a near-term forecast cycle, which enables rapid, iterative model improvements and uncertainty quantification and partitioning. Using a multi-model framework, we developed iterative near-term forecasts of methane ebullition rates 7 days into the future in a small eutrophic reservoir and updated the forecasts weekly via data assimilation. Every week, parameters and initial conditions were updated based on observations. We used a generalizable forecasting system for lakes and reservoirs coupled to an empirical auto-regressive time series model and generated 441 ensemble member ebullition forecasts per week throughout the open-water period.
Results/Conclusions
Our forecasting system successfully predicted reservoir methane ebullition rates for 24 weeks. By the end of the forecasting period, the RMSE of forecasted methane ebullition was 60% lower than a deterministic model and the weekly total forecast uncertainty decreased by 96% after 7 weeks of data assimilation. Aggregated over 24 weeks, forecasted total methane ebullition was within four grams of observed CH4 ebullition. In contrast, the deterministic model overestimated total emissions by 61%, indicating that forecasts coupled to data assimilation can improve overall estimates of methane ebullition emissions. Despite the temporal heterogeneity of methane ebullition in freshwater ecosystems, our project shows that ebullition can be successfully forecasted on weekly timescales, thereby improving seasonal methane ebullition estimates. Thus, near-term iterative forecasting is a promising approach for reducing uncertainty in methane ebullition fluxes from freshwater ecosystems.