Getting to know HydroForecast through the CEATI forecasting competition
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.
Taylor River, Colorado
Serving as important headwaters to the Colorado River, this upper picturesque stretch of the Taylor River receives snowmelt and flow from the western side of the Continental Divide (of the Rocky Mountains). From the Taylor Park reservoir, a view of the Collegiate Peaks range towers above and gives a great visual of the snowmelt and rainfall as it travels down the mountains and into the lake. In this post, we'll explore how forecasting can aid managers and operators in water stressed systems like the Taylor - Colorado River as they work to allocate water during dry years.
The forecast basin for the CEATI Rodeo competition is defined by the USGS gauge 09107000 Taylor River at Taylor Park, just upstream of the Taylor Park Reservoir created by the Taylor Park Dam (see map below). Downstream of the reservoir, the Taylor River joins with East River and flows southwest into Blue Mesa Reservoir. There it joins with the Gunnison, and serves as a primary tributary to the Upper Colorado River.
This section of the river may seem far removed from the main stem Colorado River and the management of Lake Powell. However, the current historically dry year has triggered the federal government to mandate releases from Blue Mesa and other upper Colorado River basin reservoirs to help meet compact agreements downstream of Lake Powell. This demonstrates how, in such a highly interconnected and controlled system, downstream requirements can impact upstream availability, even thousands of miles away. As droughts in the West intensify, forecasts are even more critical.
These headwaters are attractive to fishers and recreaters and critical for feeding the Taylor Park reservoir. While it serves many uses, the reservoir primarily provides water supply for irrigation. Taylor Park reservoir is part of a much larger water engineering system, the Uncompahgre Project, that diverts water to irrigate 76,000 acres of land.
A Quick History and Basin Facts
- What’s in a name. The namesake of the region actually refers to two different people: Jim Taylor, who discovered gold and for which Taylor River and the town Taylor Park are named after, and U.S. Representative Edward Taylor, for which the reservoir and dam are named.
- Land of the Utes. Originally, the lands of the Uncompahgre Project belonged to the Ute Indians. From ~1860-1900 white settlers arrived in the area and acre-by-acre took over lands via multiple “agreements.” This ultimately forced cession and resettlement of the Utes, mainly to Utah.
- On the frontlines of climate change. The Utes still remain in Southwestern Colorado, and a large number of them are farmers and ranchers. Historic drought conditions this year threaten their livelihoods as rivers and soils dry up, resulting in limited production and, consequently, layoffs. This is yet another example of how climate change is impacting Native communities.
- Hydropower generation on the table. The Bureau of Reclamation solicited proposals in 2020 to select a lessee, and contract for hydroelectric power development for Taylor Park Dam. At least one application has been received, which could generate power for 487 homes to local residents. A retrofit of the 1937 dam would be required, but the basic infrastructure needed for the project is already in place. This type of project, where hydropower is added to an existing non-powered dam, is a major focus of a recent joint statement of collaboration between the hydropower industry and environmental organizations.
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 North Fork Shoshone river basin that we highlighted last month, the active period in the Taylor is the snowmelt-driven pulse in the spring. This sets flow expectations into the reservoir and beyond through the basin for the remainder of the year. As this spring period of the live CEATI competition comes to an end, we reflect on how this year compares with others and how well HydroForecast was able to predict the timing and magnitude of this peak. To understand the performance of the model over the recent years, we evaluated it from March 2019 to June 2021.
The hydrographs below highlight March 2019 to May 2021. They show the 24-hour (top), 48-hour (middle), and 96-hour (bottom) ahead mean predictions (orange), 50% and 90% confidence intervals, long term median (purple), and USGS observations (black).
We notice a few key patterns from these plots:
- The 2019 spring melt peak was much higher and occurred later than the long term median. HydroForecast picked up on the distinct double peak in 2019 that can occur as temperatures oscillate around freezing and precipitation can be mixed.
- Confidence intervals in 2021 are wider than 2020 (though the magnitudes are very similar), likely due to less certain weather conditions.
What do the statistics say?
While the full CEATI competition runs until October 2021, we computed the statistics so far from the start of our forecasts in March 2019 to June 28, 2021. Note that perfect scores for NSE and KGE are equal to one, while an ideal bias score is zero. As a comparison point, the long-term median (i.e. climatology) has NSE = 0.46 and KGE = 0.40, a much lower predictive skill than HydroForecast.
HydroForecast holds high predictive power even five days ahead, maintaining a very reasonable bias (< 2%) and the ability to model high and low flows (illustrated in the hydrographs above).
Comparison to alternative forecasts
Over the life of the CEATI competition thus far (10/1/2020 to 7/26/2021), HydroForecast is the most accurate forecast available for this portion of the upper Taylor River. HydroForecast is in first place in all four of the metrics tracked by the competition (NSE, normalized root mean squared error, Correlation Coefficient and bias). This holds across various splices of the lead time windows: 1-5 days, 6-10, and 1-10. HydroForecast especially shines at the longer lead time windows (e.g. 5+ days) compared to other forecasts in the competition.
The complexity of western water allocation
Starting in 1860, Colorado began following prior appropriation water right laws. Basically, this law is first come, first serve. The most senior water right receives water first, then on down the list of decreasing seniority until either the water is fully allocated or there’s none left.
This water rights system results in a tightly controlled reservoir level at Taylor Park. Essentially, every drop is allocated for and used multiple times as it travels from the Continental Divide to the Gulf of California. Yet as climate change disturbs the reliability of weather and climate patterns that govern the hydrologic cycle, inflow timing and magnitude driven by snowpack is increasingly uncertain. Climate change in the Colorado River basin may impact annual runoff totals by up to 30% according to a National Academies study, creating a sense of urgency in advancing forecast capabilities. When historical trends no longer hold, managers need new tools that can capture these dynamics.
Accurate forecasts on daily to seasonal horizons can aid planning efforts to make sure no water is lost or wasted due to missing opportunities for storage and/or release. The Western Water Assessment funded by NOAA offers hydrologic outlook summaries in a Dashboard for the Inter Mountain West. We are building HydroForecast with the goal of integrating our short-term and seasonal services into regional and agency systems.
Taylor Park's main demands are irrigation and recreation, which require detailed accounting of the water balance and meeting water level targets at specific times of the year. Over the 2021 spring melt season, our 10-day forecasts tracked the diurnal melt pattern quite well, picking up on the accelerated melt during the day and slower melt at night due temperature fluctuation. Because HydroForecast’s predictions include such robust input data and predict hourly flow values, it is able to capture these diurnal patterns uniquely well. The hydrograph below shows an example of forecasts at two critical times - the rising and recession limbs of the melt.
See our June blog post for more details on why this diurnal pulsing is important and how its changing because of rising night time temperatures.
We are pleased by HydroForecast’s spring snowmelt season performance in the Taylor River. In a basin that plays such a crucial role for supporting communities throughout the Western United States, and one where accurately tracking every drop matters, HydroForecast’s strong performance is encouraging.This case study offers two lessons:
- As basins like the Colorado River experience more frequent abnormal weather and hydrological conditions, forecasting operationally on short-term and long-term planning horizons can help close the widening gap in reliability.
- What happens in Lake Powell doesn’t stay in Lake Powell. The interconnectedness of our hydrologic system, especially in the West where water rights are governed through prior appropriation, creates a ripple effect. We may see more instances of politics stepping in during extreme years to resolve allocation issues.