Near-term, iterative forecasts highlight the relative importance of two drivers for dynamic oxygen concentrations in a drinking water reservoir

Changes in climate and land use are intensifying seasonal oxygen depletion in the bottom waters of many lakes and reservoirs, leading to wide-scale degradation of water quality. However, current projections of how bottom-water (hypolimnetic) oxygen demand will change over time are primarily based on small-scale laboratory incubations or observational surveys, hindering our ability to understand how drivers interact to alter oxygen demand on a whole-ecosystem scale. In this study, we used near-term, iterative ecological forecasting and whole-ecosystem oxygenation experiments to disentangle the interactive effects of temperature and oxygen on oxygen demand. The forecasting system simultaneously sheds light on fundamental biogeochemical questions and provides a useful tool for preventive water management.

Our forecasting system predicts hypolimnetic oxygen concentrations in Falling Creek Reservoir (Vinton, VA, USA) 14 days in the future. It iteratively assimilates new observations and partitions uncertainty between four components: initial conditions, drivers, parameters, and model process. We used data from 2013-2016 for training, then tested the predictive ability of four alternative forecast models with different drivers (temperature only, oxygen only, both temperature and oxygen, and a null persistence model) using data from 2018 and 2019.

The forecasting system successfully predicted patterns in oxygen concentrations 14 days in advance, and forecasts are available in an interactive application online. In all years, the model with sensitivity to both temperature and oxygen performed better than the null persistence model (0.4-1.1 mg/L RMSE improvement during test period). The temperature-only model performed better than the oxygen-only model (0.1-0.6 mg/L RMSE improvement), indicating that temperature, beyond its control over solubility, may play a more important role in regulating the rate of change in hypolimnetic oxygen concentrations than the initial concentration of oxygen. Results from this study reveal that oxygen concentrations can be accurately forecasted using twice-weekly monitoring data, changing temperatures may substantially alter oxygen conditions, and near-term forecasts can be effectively used to assess the predictive power of ecological drivers.