Data integration and modeling process for generating operational inflow forecasting in the Mattagami River basin

The Mattagami River, located in northeastern Ontario, hosts a number of hydroelectric power facilities, which are major contributors to Ontario electric grid, flows through the City of Timmins and connect with a number of First Nation Communities. The Mattagami River can exceed inflows at northern areas of up to 5,000 cms and resulted in flooding of communities.

As a control river system, Ontario Power Generation discharges and storage plays an important role water management. OPG, City of Timmins, Mattagami Region Conservation Authority and Ontario Ministry of Natural Resource and Forestry work together to share information on changing river conditions. To improve decision support information an automated hydrologic forecasting system called HydrologiX II has been developed for the Mattagami River basin for the purpose of near real-time operational forecasting of river flows and reservoir inflows in the basin. The HydrologiX II system consists of a distributed hydrological model (WATFLOOD); data processors for streaming and preparation of model input data, including numerical weather prediction (NWP) data, observed hydrometric data, and remote sensing based snow water equivalent data; and a web-based system interface.

The core of the HydrologiX II system, the Mattagami River basin hydrological model, was calibrated and validated using historical daily data for 1980-2014 and is currently being evaluated with respect to forecasting performance during the fall 2014 to spring/summer 2015 period. The secondary purpose of the model is to be used in the general water availability studies (e.g. climate change scenarios).  The present paper describes model setup, calibration and validation, and model performance in both historical and operational forecasting contexts over past year. 

The approach to operational flow forecasting includes a spin-up period (several years) to account for uncertainties in the initial watershed conditions, a hindcast period (several months) to align the model with the current watershed conditions at the start of the forecast, and a 10-day forecast period. The spin-up model run utilizes historical daily precipitation totals from weather stations, 3-hour temperatures from the North American Regional Reanalysis (NARR) dataset, as well as historical streamflow and reservoir releases. The forecast model run is driven by the 3-hourly precipitation and temperature model from the Environment Canada GEM model, specifically the Regional Deterministic Prediction System (RDPS) for forecast days 1-2, and its Global equivalent (GDPS) for forecast days up to 10. For the hindcast period, we use the latest RDPS data and the Canadian Precipitation Analysis (CaPA) data for gridded accumulated precipitation input, and RDPS for temperature. The paper discusses the effects of various forecasting strategies on model performance, including use of RDPS versus CaPA precipitation data, and model nudging (overriding simulated streamflows with observed values) to update watershed state variables at the start of the forecast run.

The forecast information is generated daily and can be used as a complimentary source of decision support information for communities, water control structures, and emergency response.