Developing tools to model impaired streamflow in streams throughout California

Posted on September 26, 2021 by andrewrypel

by Jeanette Howard, Kirk Klausmeyer, Laura Read, and Julie Zimmerman

Droughts are extreme, but not necessarily extreme events — at least not in the way we humans usually experience events as discrete, episodic occurrences. Droughts are continuous and exhausting; they can come out of nowhere and take us on a rollercoaster of waiting for precipitation to come, measuring when it does, and hoping it will be enough to keep our rivers flowing for human use and healthy ecosystems. Droughts may feel so extreme that they should be a rare occurrence, but they are a natural part of California climate. And they will become even more frequent – climate change predictions show that extreme events, such as droughts and floods, are becoming more common (Swain et al. 2018).

Droughts are also subjective: in a year like this one, there’s no question that all of California is experiencing a drought. However, many rivers and streams in California experience drought-like conditions almost every year because of human demands for water (Zimmerman et al. 2018) such as in the Sacramento-San Joaquin Delta where the estuary experiences drought-like conditions in most years (Reis et al. 2019). Yet in most rivers in the state, we don’t have enough historical data to know how much water should be available. For example, we often don’t know if a river is running dry because it is the river’s natural condition (a stream with naturally intermittent flow), or if it should always be flowing and is running dry because humans are using too much water at the wrong times.

To effectively and reliably manage water, we need to know (1) how much water needs to be in a river to protect species and ecosystems, (2) how much water is actually in the river, and (3) what is the gap between the two? There is currently a group of scientists from universities, agencies, and NGOs that developed a framework and set of tools to answer the first question for all rivers in California – known as the California Environmental Flows Framework (CEFF). But the second question – how much water is actually in the river – has been a tough nut to crack. Predictions of actual flows would greatly enhance the ability to apply the CEFF framework by enabling the assessment of flow alteration and provide quantitative targets for how much additional water is required instream.

Actual flow measurements are only available at a finite set of point locations across the state – stream gages that are installed and operated by USGS, DWR, or other entities. Work by The Nature Conservancy found that approximately 3,600 stream gages have been active on streams and rivers in California at some point, but only about half of those have been active recently largely due to a lack of funding (see https://gagegap.codefornature.org/). Even with this network of gages, 89% of streams are poorly gaged, which means that for 89% of streams in California we have no information on actual flows and no obvious way to estimate the gap between the river flow that is needed and the flow that is available.

Investments are needed to increase our network of gaged rivers and streams. But rather than measure flow at gages in every stream in the state, it would be more efficient, cost effective, and realistic to predict flows, using a robust and scalable approach. With these predictions, we can develop storylines that tell the tales of these rivers, tracing the hydrologic signatures of a river from its historically unimpaired flows, through transitions brought by alteration (e.g., diversions, dams, land use change), to its current state today and likely state in the future. To write these stories from end-to-end, The Nature Conservancy (TNC-CA) and Upstream Tech have partnered to develop a set of tools that can estimate river flow using dynamic satellite data and machine learning methods. We began by building a model to estimate actual river flow and compared those modeled predictions to stream gage data. Our work in 300 gaged basins across California indicates that the model we’ve developed performs well and can also estimate actual river flow in ungaged basins.

What could be the impact of this model for planning and management in dry years? Consider this example from the Yuba River near Marysville (drainage area = 1,339 mi2 [3,450 km2]):

Using our ‘unimpaired’ flows model, which predicts naturalized streamflow similar to a typical rainfall runoff model, the validation period shows a high bias in the model’s prediction during the spring snow-melt period. The observation record dates back to 1987, so we don’t know the flows here before the reservoirs were built upstream in the basin, but we do know that the signature is altered and that this model is not reflective of impaired flows today.

The challenge ahead of us is to capture this altered behavior such that we can make estimates given the current storyline of the river. The extra challenge: do it with public data sources that have information across many basins so that this approach is scalable and not dependent on difficult to obtain and maintain data (e.g. dam-specific operation plans). Using over 1,000 ‘enhanced basin characteristics’ that were collected from StreamCAT, USGS Gages II, and USGS channel alteration datasets, we enhanced our unimpaired flows model to create an actual flows model. Here’s the result at the same Yuba River site:

The model described the shift from a natural snow-melt signal to a longer and flattened spring peak, reflecting the altered current conditions in the watershed. The real kicker of this site and others like it in our study: this was one of our “test” sites, which in machine learning terminology means that it was hidden from the model during training and treated as ungaged. The model never saw any observations from this site, never saw this time period even at other sites, and still learned this behavior shift from the enhanced basin characteristic inputs that we gave it. The Nash-Sutcliffe Efficiency (NSE), a common goodness-of-fit metric in hydrology, improved from -2.7 in the unimpaired model (top plot) to 0.77 in the actual flows model (bottom plot). A perfect NSE is one.

In the next phase, TNC-CA and Upstream Tech are expanding this approach to predict historic daily flows in 350 basins from 2000-2020, further validating modeling of ungaged impaired flows. From there, we’ll work on expanding estimates to ungaged streams across the state.

In dry years better estimates of flow in ungaged streams can provide a lifeline of information for real and near-term operational decisions, curtailments in drought years, assessments of water rights applications, and many other decisions that require information about water availability. For example, this year the operational forecasts largely missed this drought’s severity. A model that can simulate historic daily flows in altered basins across California could be used in the future for forecasting daily and seasonal flows with confidence and be able to answer crucial questions for decision makers such as:

  • “Will we pass critical habitat thresholds next month?” (Environmental flows planning)
  • “What is the likelihood that flows will be between X and Y, yielding sufficient water for all users?” (Allocation decisions from daily to three-months out)
  • “What types of basins need more gages because the hydrology is difficult to model and the data can improve model predictions elsewhere?” (Gage gap analysis and gage prioritization)

As the climate changes, the data and models we use must keep pace. Let’s work together to discover the true stories of our rivers and how they can better shape our future relationship with water as a lifeline for humans and nature.

Jeanette Howard, Ph.D., is the Director of Science for the Water Program for The Nature Conservancy’s California ChapterKirk Klausmeyer is the Director of Data Science for The Nature Conservancy’s California Chapter. Laura Read, Ph.D., is a Product Manager of HydroForecast at Upstream Tech. She co-authored this blog, representing the technical work of Alden Keefe Sampson and Mostafa Elkurdy. Julie Zimmerman, Ph.D., is Lead Scientist for The Nature Conservancy’s California Water Program.

Further Reading:

California Environmental Flows Working Group (CEFWG). 2020. California Environmental Flows Framework. California Water Quality Monitoring Council Technical Report 37 pp.

Grantham, T. E., J. K. H. Zimmerman, J. K. Carah, and J. K. Howard. 2019. Streamflow modeling tools inform environmental water policy in California. California Agriculture 73(1): 33-39.

Reis, G.J., J.K. Howard, and J.A. Rosenfield. 2019, Clarifying effects of environmental protections on freshwater flows to – and water exports from – the San Francisco Bay Estuary. San Francisco Estuary and Watershed Science 17(1): 1-22.

Swain, D. L., B. Langenbrunner, J. D. Neelin, and A. Hall. 2018. Increasing precipitation volatility in twenty-first-century California. Nature Climate Change 8: 427-433.

Yarnell, S. M., E. D. Stein, J. A. Webb, T, Grantham, R. A. Lusardi, J. Zimmerman, R. A. Peek, B. A. Lane, J. Howard, and S. Sandoval Solis. 2020. A functional flows approach to selecting ecologically relevant flow metrics for environmental flow applications. River Research and Applications 36: 318-324.

Zimmerman, J. K. H., D. M. Carlisle, J. T. May, K. R. Klausmeyer, T. E. Grantham, L. R. Brown, and J. K. Howard. 2018. Patterns and magnitude of flow alteration in California, USA. Freshwater Biology 63: 859-873.