A near-term iterative forecasting system successfully predicts reservoir hydrodynamics and partitions uncertainty in real time

Abstract

We created a near-term iterative lake water temperature forecasting system that uses 13 sensors, data assimilation, and hydrodynamic modeling 14 15 • FLARE quantifies the uncertainty in each daily forecast and provides an open-source, 16 generalizable system for water quality forecasting 17 18• 16-day forecasted temperatures were within 0.91℃ over 100 days in a reservoir case Abstract 22 Freshwater ecosystems are experiencing greater variability due to human activities, necessitating 23 new tools to anticipate future water quality. In response, we developed and operationalized a 24 near-term iterative water temperature forecasting system (FLARE -Forecasting Lake And 25 Reservoir Ecosystems) that is generalizable for lakes and reservoirs. FLARE is composed of: 26 water quality and meteorology sensors that wirelessly stream data, a data assimilation algorithm 27 that uses sensor observations to update predictions from a hydrodynamic model and calibrate 28 model parameters, and an ensemble-based forecasting algorithm to generate forecasts that 29 include uncertainty. Importantly, FLARE quantifies the contribution of different sources of 30 uncertainty (parameters, driver data, initial conditions, and process) to each daily forecast of 31 water temperature at multiple depths. We applied FLARE to a temperate reservoir during a 100-32 day period that encompassed stratified and mixed thermal conditions and found that daily 33 forecasted water temperatures were on average within 0.91℃ at all depths of the reservoir over a 34 16-day forecast horizon. FLARE successfully predicted the onset of fall turnover eight days in 35 advance, and identified meteorology driver data and downscaling as the dominant sources of 36 forecast uncertainty. Overall, FLARE provides an open-source and easily-generalizable system 37 for water quality forecasting for lakes and reservoirs to improve management. 39Lake Model, Water temperature 41 45 freshwater ecosystems, which have been more degraded than any other ecosystem on the planet 46 [Millennium Ecosystem Assessment 2005], are seeking new tools to anticipate future change and 47 ensure clean water for drinking, fisheries, irrigation, industry, and recreation [Brookes et al. 48 2014]. 49 In response to this need, near-term iterative ecological forecasting has emerged as a 50 solution to provide stakeholders, managers, and policy-makers crucial information about future 51 ecosystem conditions [Clark et al. 2001, Dietze et al. 2018, Luo et al. 2011. Here, we define a 52 near-term iterative forecast as a projection of future ecosystem states with fully-specified 53 uncertainties, generated from predictive models that can be constantly updated with new data as 54 they become available [Clark et al. 2001]. Importantly, a near-term iterative forecast is not 55 created from merely one ecosystem simulation, but an ensemble of simulations that enable 56 quantification of the uncertainty in the forecast contributed by different sources [i.e., parameters, 57 driver data, initi...

Advancing lake and reservoir water quality management with near-term, iterative ecological forecasting

“…The daily data assimilation for FLARE includes 5 major steps that are triggered each morning ( Fig. 1; described in detail by Thomas et al 2020a). First, FCR water temperature and DO concentrations are simulated by GLM-AED for the preceding 24 h on an hourly time step, using observed meteorology and inflow data as model inputs.…”

“…The ensemble output is also archived for future analysis in a GitHub repository for versioning control. We refer interested readers to Thomas et al (2020a) for detailed information on FLARE setup and performance in forecasting thermal structure; here, we focus on DO forecasting and its application for management.…”

“…We used the FLARE system to develop scenario-based DO forecasts that inform HOx system activation and operation. These DO forecasts build on our existing infrastructure for water temperature forecasting detailed by Thomas et al (2020a) by simulating a depth profile of DO and including 2 additional GLM-AED parameters in the oxygen module that are calibrated by the ensemble Kalman filter (see Supplemental Material for detailed DO forecasting methods, oxygen module configuration, parameters, and DO forecast performance evaluation).…”

Influence of climate forecasts, data assimilation, and uncertainty propagation on the performance of near-term phenology forecasts

“…While near-term iterative ecological forecasting is becoming more prevalent (Van Doren and Horton, 2018;Welch et al, 2019;White et al, 2019;Thomas et al, 2020;Pearlstine et al, 2020) there has been limited work evaluating these forecasts and attempting to determine how different aspects of the iterative system influence their performance. Quantifying the proficiency of a forecast and how it is influenced by different aspects of data and modeling aids end users in forecast interpretation, allows inter-comparison of different models and methods, and can point the way toward model improvement.…”

“…Integrating climate forecasts and assimilating climate data into ecological forecast systems both come with development and computational costs (Taylor and White, 2020;Thomas et al, 2020;Welch et al, 2019), and they should be rigorously tested to justify inclusion over simpler methods. Comparison against a baseline model, such as one based on the long term climatological average, allows the assessment of "forecast skill" (Jolliffe and Stephenson, 2003;Harris et al, 2018) which indicates the improvements to the model by including detailed forecasts and data on climate.…”