The Bureau of Reclamation is currently preparing an Environmental Impact Statement (EIS) to establish new operational guidelines for Lake Powell and Lake Mead beyond 2026. As part of this effort, they have developed updated hydrology datasets to evaluate how different management alternatives might perform. These new datasets differ significantly from those used to create the 2007 guidelines and the 2019 Drought Contingency Plan. During the August meeting of the Metropolitan Water District’s Imported Water Subcommittee, Senior Engineer Laura Lamdin provided an in-depth explanation of how the Bureau of Reclamation is integrating climate change considerations into its planning for post-2026 Colorado River operations.
The Index Sequential Method
Planning for the future is no easy task. Historically, particularly in the 20th century, many people approached planning by assuming that the future would closely resemble the past. In water resource planning, this assumption was operationalized through methods like the index sequential method, which used historical data to generate a range of potential future scenarios.
The index sequential method works by creating “traces” from historical records. For example, imagine you’re planning for the next 30 years and have a dataset spanning from 1901 to 2000. Using this method, your first potential future (or trace) would align the hydrology of the first year of your analysis with the first year of the dataset, 1901. The second year of your analysis would then correspond to 1902, and so on, stepping through the dataset year by year until you’ve constructed a 30-year sequence. This sequence becomes one potential future, or trace.
The second trace would shift the starting point forward by one year. In this case, the first year of your analysis would align with 1902, the second year with 1903, and so on, again creating a 30-year sequence. This process continues, shifting the starting year forward each time, until every year in the dataset has been used as the starting point for a potential future. If you reach the end of the dataset before completing a 30-year sequence, the method loops back to the beginning of the dataset. For instance, in this example, the year following 2000 would wrap back around to 1901.
With a 100-year dataset, this approach would generate 100 potential futures, or traces, each representing a plausible hydrological scenario. This method has been widely used in the past, including by the Bureau of Reclamation.
Relying on historical data is no longer sufficient. What now?
In the 21st century, it became clear that relying solely on historical data was no longer sufficient. The past holds more variability than what has been directly observed, and the impacts of climate change introduce entirely new challenges. Recognizing this, Metropolitan Water District took a more advanced approach during its last Integrated Resource Plan process, moving beyond the traditional Index Sequential Method. They adjusted the traces to incorporate the effects of climate change. Similarly, planning efforts on the Colorado River have also evolved to address these growing complexities.
The graph below illustrates the natural flow of the Colorado River at Lees Ferry. Natural flow, similar to unimpaired flow, represents the amount of water that would have been in the river without human intervention, such as water diversions for agricultural, municipal, or industrial use. Lees Ferry serves as the dividing point between the Upper and Lower Basins of the Colorado River. Therefore, the natural flow at Lees Ferry reflects how much water would have flowed from the Upper Basin into the Lower Basin in the absence of human activity.
Historically, water planning relied on methods like the Index Sequential Method, which used the full hydrologic record to simulate potential futures. For example, the 2007 Interim Guidelines were developed using the entire historical dataset available at the time. Back then, the average natural flow at Lees Ferry was estimated at 14.8 million acre-feet per year. However, with the prolonged dry years experienced since, that average has dropped to 14.6 million acre-feet per year.
By 2019, the Drought Contingency Plan (DCP) introduced a more targeted approach, using a subset of the hydrologic record known as the “stress test” period. This period, beginning in 1988, has been notably drier than the overall historical record, with an average natural flow of just 13 million acre-feet per year. The stress test was explicitly chosen to evaluate and plan for the challenges of sustained dry conditions, reflecting the evolving realities of water management in the Colorado River Basin.
“Even within the stress test, the last 24 years have been drier than that- an average of only 12.4 million acre feet, representing another decline,” said Ms. Lamdin. “These smaller subsets, while they do test for drier conditions, contain a much smaller amount of variability than the full record. So while we have seen this drying trend, we do expect that the Colorado River will still have wet years and dry years, as well as droughts and wet periods, and all of that variability within the future. And so we don’t want to lose that variability in our planning efforts.”
A new method: the “super ensemble”
So the Bureau of Reclamation has been exploring a variety of methods to improve future water planning, each with its own strengths and limitations:
- Observed History: This approach relies on precise, measured records collected from various locations. While it provides accurate data, it lacks the full range of historical variability and does not account for the impacts of climate change.
- Paleo Hydrology: By analyzing tree ring data, this method reconstructs hydrology from periods predating the observed record. It captures greater variability, including more extreme highs, lows, and prolonged droughts. However, it remains tied to historical conditions and does not incorporate climate change. Additionally, translating tree ring data into runoff estimates introduces a level of imprecision.
- Global Climate Models (GCMs): These models simulate temperature and precipitation on large, gridded scales, which must then be downscaled and processed through land surface models to estimate runoff. While GCMs can incorporate future climate scenarios, the downscaling and modeling processes can introduce biases, particularly in regions with complex topography like the Colorado River Basin.
- Statistical Approaches: Observed historical data or paleo hydrology can be adjusted using statistical relationships between temperature and runoff to account for climate change. While this method can capture some climate-driven changes, it oversimplifies the complex physical processes involved, potentially missing non-linear interactions that could significantly impact outcomes.
Each of these methods offers valuable insights, but they also highlight the challenges of accurately predicting future hydrology in a changing climate.
Since no single method stands out as the definitive solution, the Bureau of Reclamation (USBR) is adopting a comprehensive approach for the post-2026 analysis by creating a “super ensemble.” This super ensemble combines elements from all four hydrology methods to provide a robust framework for evaluating future scenarios. The goal of the super ensemble is not to predict the future or account for every possible outcome, but rather to identify the conditions that could lead to undesirable results and inform better decision-making.
The super ensemble will consist of 400 traces, each representing a possible 30-year future. It incorporates a diverse mix of data sources, including:
- Observed Historical Data: Using the stress test period to reflect recent dry conditions.
- Post-Pluvial Temperature-Adjusted Data: This includes 100 traces where USBR has statistically adjusted the observed historical record (1920 to the present) to account for climate change impacts, similar to the approach used in Metropolitan’s Integrated Resource Plan (IRP).
- Global Climate Model Data: Traces are drawn from the CMIP3 generation of global climate models to incorporate future climate projections.
- Paleo Hydrology: Tree ring data is included to capture the variability of hydrology from pre-observed periods.
The selection of these 400 traces was carefully curated to ensure each trace provides unique insights. A variety of metrics were used to guide the selection process, including annual average runoff, annual minimum and maximum flows, the minimum, maximum, and median of 2- to 20-year running averages, and 30-year trends. This diverse and methodical approach ensures the super ensemble captures a wide range of potential hydrological conditions, offering a valuable tool for planning and risk assessment in the face of uncertainty.
“This way, you can get traces that explore a wide variety of sequencing of wet and dry conditions over a variety of average annual flow conditions,” said Ms. Lamdin. “It lets you consider droughts with different durations and intensities than we’ve seen in the historical record. And because of this, they don’t represent probability; they just represent examples of what could happen.”
The super ensemble is highly comprehensive. The graphic shows the average annual runoff over the 30-year period for each trace.
“The shape of all of the dots is very long and skinny, so that shows that it is representing a wide variety of average annual conditions,” said Ms. Lamdin. “In some futures, the average runoff is almost 20 million acre feet a year; if that were to happen, no reductions would be needed. In other futures, the average annual flow is all the way down at around 8.5 MAF. Since 2000, the average flow has been about 12.5 million acre feet, so dropping down to 8.5 MAF does represent incredibly dry conditions being considered. If that were to happen, reductions in excess of the 4 million acre feet that former Commissioner Touton mentioned back in 2022 would be necessary. So USBR is considering very dry conditions.”
Evaluating alternatives not by probability, but by robustness and vulnerability
Ms. Lamdin emphasized that the super ensemble is not designed to predict future outcomes with any specific probability. In other words, it’s not appropriate to interpret the results as, “If 10% of traces show X, Y, or Z happening, then there’s a 10% chance of that occurring in the future.” Instead, the 400 traces are not meant to simulate every possible future but to provide a broad enough range of scenarios to identify the conditions that lead to undesirable outcomes.
This raises the question: if probability isn’t the lens through which these results are evaluated, how will the alternatives be assessed? The answer lies in two key concepts: robustness and vulnerability.
- Robustness measures how well an alternative performs across a wide range of possible futures. A robustness threshold is set to define what constitutes a desirable outcome, and this metric evaluates how many traces meet that threshold. It provides a high-level view of how consistently an alternative achieves acceptable results across many scenarios.
- Vulnerability, on the other hand, focuses on the situations where outcomes fall below the robustness threshold. It helps identify the conditions that lead to unacceptable results and provides insights into the specific weaknesses of an alternative. Vulnerability metrics are critical for understanding how severe the consequences could be when things go wrong.
Together, these metrics offer a more nuanced way to evaluate alternatives, moving beyond probabilities to focus on resilience and risk under a wide range of potential future conditions. This approach ensures that decision-making is informed by both the likelihood of success and the potential for failure.
Ms. Lamdin provided an example regarding the elevation of Lake Mead staying above elevation 1050 more than 80% of the time over a 20-year period. “Let’s say we’re going to evaluate alternative one. You look at the first trace, and Lake Mead stays above elevation 1050 the whole time. That’s an acceptable performance. You look at trace two, and Lake Mead stays above elevation 1050 81% of the time over 20 years, so that would also be an acceptable performance. If you have a third trace where Lake Mead only stays above elevation 1050 40% of the time, that would be an unacceptable trace. For each alternative, you would step through all 400 traces to determine the percentage of desirable, acceptable outcomes. So in this completely made-up example, you get 90% of the 400 traces for alternative one having desirable outcomes that is higher than the other alternatives. So it would be considered a robust strategy in terms of Lake Mead elevation.”
Vulnerability is the flip side to robustness. It is a similar process, but for unacceptable outcomes. “So an unacceptable performance could be if Lake Mead is below elevation 1000 more than 10% of the time over the next 20 years. So you’d step through and do that same analysis for each trace, and then you could graph it and compare the alternatives.
There are countless ways to analyze and visualize the concepts of robustness and vulnerability, and the slide below highlights three potential approaches.
One method under consideration by Reclamation staff is the use of box plots. Box plots are an effective way to display the distribution of data across a range of outcomes, providing a clear visual summary of variability, medians, and extremes. These box plots will likely be categorized, or “binned,“ by hydrological conditions. For example, Lake Mead’s elevation results could be presented as separate box plots for wet, average, and dry scenarios. This approach allows for a straightforward comparison of how different conditions impact outcomes, offering valuable insights into the performance of alternatives under varying hydrological circumstances.
Ms. Lamdin noted that, given all the changes to the NEPA rules mentioned at previous meetings, we really don’t know what it will look like.
In summary …
To summarize, the USBR is evaluating alternatives against hydrologies differently than before, including consideration of drier conditions and more intense and pervasive droughts. The hydrologic conditions will be used to assess alternatives under consideration, but that assessment will not be based on probability; instead, it will be based on the concepts of robustness and vulnerability.
For more on the concepts of robustness and vulnerability, check out these Reclamation slide decks:
- Hydrology: https://www.usbr.gov/ColoradoRiverBasin/post2026/ITEW%20Session%203%20Hydrology_WebPosting_508.pdf
- Metrics: https://www.usbr.gov/ColoradoRiverBasin/documents/post2026/itew/ITEWSession5_MORDM_WebPosting_508.pdf
DISCUSSION
During the discussion period, Chair Mark Gold noted that this approach is a much more management-friendly approach. “It’s saying these are the various management outcomes that are going to trigger us to take an action, as opposed to, what’s the percent chance that you’re going to have this flow in the river by this date mid-century. .. So, presenting it in the form of management outcomes that decision-makers can understand, but probability feeds into that completely, and there is still some subjectivity at that level. Are you comfortable with 1050, and being below that number 70% of the time? It’s still going to fall on us to some degree, but they’re narrowing it down and doing it in a way that I think is more comfortable for people to handle, rather than taking on an abstract, acceptable risk that becomes really difficult. I have never met any elected official who’s been comfortable with the acceptable risk discussion; it’s a very, very tough thing to do. In toxicity, we discuss it frequently. Are you comfortable with a one in 1000 cancer risk versus a one in 10,000 cancer risk? That’s not an easy decision for people like us to make, which is why this change is really interesting. It enables us to handle it better by putting it in the format of decisions we’re used to making.”
Shanti Rosset, Colorado River Program Manager, noted that the super ensemble includes much wetter and much drier traces than were used in previous modeling for the 2007 interim guidelines and for the 2019 DCP. “One of the results of that is it’s going to make the results of the alternatives look worse. So, everything will appear less robust than in previous iterations, because there are going to be many wetter and drier conditions, and there’s going to be more frequent vulnerability shown for both Lake Powell and Lake Mead. The upper basin has a very conservative view. They do not want Lake Powell to fall below certain elevations under any conditions. They consider that an unacceptable risk, and their risk tolerance is much lower than the risk tolerance in the lower basin generally. And so if anything, this makes it harder to reach a consensus with them.”
