Data Package Metadata   View Summary

GENERATION OF CHLA FORECASTS: Detailed methods can be found in Woelmer et al. 20XX, submitted to Ecological Applications in April 2021. Briefly, forecasts of near-surface (~1.0 m depth) chl-a at FCR were produced for ~600 days (from 2 January 2019 to 15 August 2020) using daily, weekly, and fortnightly autoregressive (AR) linear models developed using observational sensor data from FCR. For each forecast, model driver and validation data were collected via automated sensors at the study site and wirelessly uploaded to an online data repository using secure sensor gateways. At each model time step, new data were appended to the historical training dataset and used to re-parameterize the AR models. Forecasted model driver data included shortwave radiation and forecasted discharge to the reservoir from the major inflow. Shortwave forecasts were automatically downloaded from the National Atmospheric and Oceanic Administration Global Ensemble Forecasts System (NOAA GEFS) repository. Forecasts of future discharge were modeled using both sensor discharge measurements at the inflow at FCR and NOAA forecasted precipitation as inputs (see Thomas et al. 2020 for more details). Uncertainty was propagated for five different uncertainty types (process, initial condition, parameter, and driver data, where we propagated both meteorological and discharge driver data uncertainty individually). We generated probabilistic daily forecasts with 441 ensembles which had a 1-day, 2-day, 3-day... up to 14-day time horizon, weekly forecasts which had a 1-week and 2-week (i.e., 7-day and 14-day) time horizon, and fortnightly forecasts which had a 2-week time horizon. To develop and run our forecast models, we used a combination of linear parametric and Bayesian statistical methods. We used parametric linear model selection on historical data to pick model covariates and starting parameter values. To produce forecasts, we applied our model in a Bayesian framework. All forecast output is published in this data product.

NAMING CONVENTION FOR HINDCAST FILES: Within the provided .zip folders, we have included forecast output for all three models under all uncertainty modes (defined below) (forecast_output_EDI), parameter output for all three models under uncertainty mode 1 (parameter_output_EDI), and null persistence forecast output (null_model_EDI). We also provide three example files, one for each file type contained in the three zipped folders. Naming convention for each .csv file is as follows: The forecast_output_EDI folder includes files following the convention: YYYY_MM_DD_chla_TIMESTEP_UNCERTMODE.csv The parameter_output_EDI folder includes files following the convention: YYYY_MM_DD_ensemble_parameters_TIMESTEP_UNCERTMODE.csv The null_model_EDI folder includes files following the convention: YYYY_MM_DD_null_summary_TIMESTEP.csv

TIMESTEP = 1day, 7day, or 14day depending on the timestep of the model used to make the forecast And UNCERTMODES are 1 = all uncertainty 2 = only process uncertainty 3 = only weather driver uncertainty 4 = only initial condition uncertainty 5 = only parameter uncertainty 6 = only discharge driver uncertainty

CITATIONS: Thomas, R.Q, R.J. Figueiredo, V. Daneshmand, B.J. Bookout, L. Puckett, and C.C. Carey. 2020. A near-term iterative forecasting system successfully predicts reservoir hydrodynamics and partitions uncertainty. In press at Water Resources Research. DOI: 10.1029/2019WR026138