We begin our 161 kilometer (100 mile) journey down the Rouge River at its headwaters lake, known in this French-speaking region as Lac de la Fougère.
The Streamflow Forecast Rodeo is a competition hosted through a partnership between the Bureau of Reclamation and the Centre for Energy Advancement through Technological Innovation (CEATI)'s Hydropower Operations and Planning Interest Group. The Rodeo website summarizes: "[This] challenge seeks to improve the skill of short-term streamflow forecasts (10 days) via a year-long competition." Check out the map below to see the 19 hard-to-forecast sites selected for the competition.
Each month we’ll shine a spotlight on a different site, review HydroForecast’s overall performance, and take a look at interesting events.
We begin our 161 kilometer (100 mile) journey down the Rouge River at its headwaters lake, known in this French-speaking region as Lac de la Fougère. Traveling downstream we are captivated by red tinted riverbanks from exposed iron as we weave on and off the Canadian Shield. The scenery offers the surrounding Laurentian mountains and views of the vast protected wildlife reserve lands. As we approach the confluence with the Ottawa river, townships dot the landscape and river rafters take on thrilling waves. When the two rivers meet, one can imagine looking upstream at Ottawa city and downstream toward Montreal as the Rouge connects in and joins the mighty Ottawa river.
HydroForecast provides state-of-the-art, accurate streamflow forecasts using a hybrid approach that combines physical science with artificial intelligence. HydroForecast offers a range of advantages over existing forecasting techniques, and we've joined the CEATI competition in order to exhibit, live, these strengths. Under the hood, every forecast is created by an ensemble of neural networks that are provided different members of meteorological forecast ensembles. HydroForecast is rapid to deploy in a new basin and resilient to basin and climatic changes.
Like the Dworshak basin that we highlighted last month, the most interesting and highest inflow period of the year in the Rouge is the spring freshet (snowmelt). That period of the live competition is yet to come so we’ve evaluated our model on previous years to understand its performance. To do this, we produce reforecasts and then quantify the model’s performance against observations that it has not used or seen.
The hydrographs below highlight the active spring melt period from March through July. Each plot shows the 24-hr, 48-hr and 72-hr mean prediction (green), 50% and 90% confidence intervals, and observations (black) over the April 2019 event.
In late April 2019, the Rouge River at Bell Falls reached dangerous water levels never before seen due to heavy spring rains mixed with melting of an elevated snowpack. Just how big? Hydro-Québec reported that by April 1 soil moisture levels from snowpack were 177 percent of normal and rainfall was 210 percent of normal. This extreme weather caused unprecedented high flows, swelling the river to a one thousand year event. A state of emergency was declared. Residents below the dam were evacuated out of an abundance of caution and, though the dam did not fail, insured flood damages downstream on the Ottawa River were widespread and estimated upwards of $200 million. Following the event, Hydro-Québec announced plans to increase flow volume capacity through modification of turbines in order to help prevent future risk from similar events.
Coincidentally, the April 2019 event fell into our model validation period. As noted from the hydrographs over this period, HydroForecast captured the timing of the rising limb and recession of this significant event quite well, and also forecasted the magnitude for the duration of the event and the entire freshet period within the 50% confidence intervals.
While the full CEATI competition continues to run until October 2021, the statistics from our one year reforecast validation period provide confidence that we’ll capture the behavior in the Rouge River watershed well. Note that perfect scores for NSE, Correlation and KGE are one, while an ideal bias score is zero.
Over the life of the CEATI competition thus far [10/1/2020 to 3/15/2021], the ‘by lead time’ statistics indicate HydroForecast is in first place in all four metrics over the 1-10 day and 1-7 day lead-time windows. In the 1-3 day lead-time window, we lead in three of four metrics. Since the competition began, our 1-10 day NSE is 0.83 and the Correlation Coefficient is 0.91.
One effect of climate change induced warming is that we are less certain that precipitation over the winter will always come as snowfall and accumulate as it has in the past. In mountainous and cold regions across the globe, one study found some evidence that rain on snow events are increasing in Eurasia, Canada and the United States. The authors describe that this phenomenon is tricky to measure because while warmer temperatures mean more rain instead of snow, it also means the snow is melting the existing snow cover.
Since the CEATI rodeo began, we have seen a couple of streamflow pulses above the relatively steady, low winter flows. The hydrograph below illustrates these, around 12/05/2020 and then again around Christmas day, 12/25/2020.
First, we notice that the observations (black line) goes out for a period, 12/16/2020 - 01/01/2021 and miss the initial, larger peak of this event. Any time a gauge goes out of service it can lead to much strife for operational systems as well as any models that rely on the observations. HydroForecast’s architecture is set up such that observations are used when available, but if they are not, the system will not be disrupted or result in abrupt shifting.
The second piece in this investigation is to take a deeper look into how the model knew to produce streamflow during this winter period. The plots below show time series from two of our important model inputs, precipitation and temperature. The black arrows point to precipitation events that led to the streamflow pulses we see in the hydrograph above. If we check the temperature during these black arrow precipitation events, we see that the model picked up on the transition from below freezing (when precipitation accumulates), to above freezing temperatures (when runoff and melt occurs) and correctly predicted runoff. For a different case a couple of weeks later, the purple arrows show a precipitation event which happened when temperatures were below freezing. The model correctly kept the streamflow low here, knowing it should accumulate snow. These events are quite difficult and usually reserved for the spring transition period, though as we see more climate variability they may become more frequent and important for tracking winter flows.
Spring is beginning and we are excited to see how HydroForecast tracks snowmelt through the freshet season. Our near perfect NSEs during validation and events during CEATI so far are promising for the coming months.