Application of a novel near-term, iterative water quality forecasting workflow to NEON lakes

Link To Share This Poster: https://cdmcd.co/Rrm93b
Live Discussion Link: https://cdmcd.co/4dqGXA

Ryan McClure, Cayelan Carey and Tadhg Moore, Biological Sciences, Virginia Tech, Blacksburg, VA, Vahid Daneshmand and Renato J. Figueiredo, Advanced Computing and Information Systems Laboratory, University of Florida, Gainesville, FL, Quinn Thomas, Forest Resources and Environmental Conservation, Virginia Tech, Blacksburg, VA

Background/Question/Methods
Lakes and reservoirs globally are increasingly threatened as a result of rapidly changing land use and climate. In response, we applied a novel forecasting workflow (Forecasting Lake and Reservoir Ecosystems, FLARE) to predict future water temperatures at time scales relevant for stakeholders to make management decisions. FLARE is an open-source forecasting system composed of water quality and meteorology sensors deployed on a lake or reservoir, cyberinfrastructure that wirelessly and securely transfers data from the field to the cloud, and an ensemble‐based forecasting and data assimilation algorithm. This produces forecasts of future lake conditions using freely available NOAA weather forecast ensembles and quantifies and partitions the different sources of forecast uncertainty using a 1-D hydrodynamic model. To date, FLARE has been successfully implemented in one freshwater reservoir, so applying the system to other lakes and reservoirs with different sensor networks and data availability is needed to build its robustness and scalability. As a result, we expanded FLARE to more freshwater lakes and reservoirs with water quality and meteorological observations collected in near-real time and are open source, including seven focal lakes from the National Ecological Observatory Network (NEON).
Results/Conclusions
Here, we focus our analysis on Crampton Lake, Wisconsin as an exemplar of applying FLARE to a NEON lake. Crampton Lake is a small clearwater lake with a surface area of 0.25 km2 and maximum depth of 19 m. We modified our existing FLARE workflow for forecasting water quality in NEON lakes by using a new R package for downloading new NEON data (neonstore). We used three years of historical water temperature and meteorological data to calibrate the 1-D hydrodynamic model for Crampton using parameter optimization techniques and successfully recreated water temperature dynamics with an RMSE of 1.2 °C. The calibrated model was then integrated into FLARE to predict future water temperatures with meteorological forecasts as driver data. Because NEON has a 1.5 month data latency, we evaluated our forecasts using hindcasts. Comprehensive water quality forecasting is in process, but to date, testing for water temperature shows that hindcast prediction intervals encompassed the out-of-sample water temperatures in Crampton Lake. Our application of FLARE to NEON lakes demonstrates that open-source NEON data can be successfully integrated into an open-source forecasting workflow to generate near-term, iterative forecasts. We envision that the FLARE forecasting system will serve as a model for forecasting of other NEON variables and water quality in other lakes and reservoirs.