Ninth Plan Hydro Modeling Data

Hydro Data for Public Review of GENESYS Model Update 

Published November 2026

During public comment on the Ninth Plan, an error was identified in the GENESYS model set up. More information on this was shared at the Power Committee on October 23, 2026 and the RAAC/SAAC meeting on November 3, 2026. The data sets linked below provide data from both before and after the modeling fix. We welcome review and questions on these data.

Please review the information below before downloading:
Data sets of original results before the issues were fixed and the results after the fix was implemented. 
Note on format: The data provided for the original results includes individual files for each project and datatype. The data for the updated results is aggregated by project with different tabs per datatype.

Data Set Information:

The GENESYS data request specifically covers:

• Coulee, Libby, Hungry Horse, Albeni Falls, Dworshak:
  • Inflows (unregulated), total outflow, turbined outflow, spilled outflow, generation, forebay
• Lower Columbia and Lower Snake projects:
  • Total outflow, turbined outflow, spilled outflow, generation, forebay
• Priest Rapids:
  • Total outflow

The data for the original results includes 450 simulations for each project/datatype corresponding to the 90 climate years (simulations) under the five load trajectories. The data for the updated results includes 90 climate years, just from the High Load trajectory (with future strategy checks conducted for all load trajectories).

As a reminder, per Appendix N, the 2026 Fish & Wildlife Program adopted spill measures based on the operations as defined in the 2023 Resilient Columbia Basin Agreement, but without a specific end date to the August performance spill. For spring, this means 24 hour 125% total dissolved gas (TDG) spill. For summer, since the 2026 Fish and Wildlife Program does not include a measure on summer spill, the Council assumed extended performance spill until August 31 at the associated Lower Snake and Lower Columbia projects, because this is a more conservative approach for testing power system adequacy. In addition, the Council’s Fish and Wildlife Program calls on operators to limit the flexibility (daily ramping) of the Lower Columbia and Lower Snake projects during the migration season of April-August to minimize the impact on migratory fish populations. Each column is one simulation with 8760 hours where Dec 31st is ignored for modeling purposes (so the year can be evenly divided by 52 weeks).

Inflows:
The unregulated inflows are in a separate file. The format sorts the data by simulation/water year, with different tabs by project. Each column represents one simulation, starting Oct 1^st and ending Sep 30^th. The naming convention key for each column is “Scenario # (A/C/G_#), for example “Scenario 1 (A_1)” which means simulation 1, climate model A, first water year in climate model out of 30). The Scenario # ranges from 1-90, and these match the 1-90 in the project output files (elevation, generation, spill, outflow, and turbine).

Scenarios 1-30 represent CanESM2_RCP85_BCSD_VIC_P1 (what we call A, with A_1, A_2 , …, A_30), scenarios 31-60 represent CCSM4_RCP85_BCSD_VIC_P1 (what we call C, with C_1 through C_30), and scenarios 61-90 represent CNRM-CM5_RCP85_MACA_VIC_P3 (what we call G, with G_1 through G_30).

High Level Reminders:

1.  GENESYS is regionally optimizing through plant coordination, and benefits and tradeoffs have been considered and discussed for over ten years with regional operators and planners. Multiple hydro operation reviews took place, and future ones will continue to be conducted to routinely ensure the hydro representation is as accurate and appropriate as possible. GENESYS is not perfect, but with valuable feedback it has been improved, has greater confidence and comfort with the model, has and will continue to be improved through stakeholder engagement.

2.  Reminder that GENESYS is a multi-stage production cost model that co-optimizes energy and reserves while accounting for fuel use, forecast error and system constraints.

• GENESYS is used as a planning level model with operational level constraints. While aiming to capture the dynamics of the regional hydro system it does not contain enough specific operational data to mimic exact operations of every dam. Rather, as an optimization model adhering to complex system obligations, it portrays the operations of the projects under climate-informed stream flows under a high degree of system coordination.
• Part of these operational decision points revolve around satisfying hard and soft constraints, with a range of relative penalties scaled by priority of the constraint to make sure decisions adhere as best as possible to the intention of constraints.
• GENESYS is a multiple stage model, looking at the week-ahead, day-ahead, hour-ahead, and a true-up (“dispatch”). With multiple stages there are multiple seams (between the week and the day-ahead, the day and the hour-ahead and the hour-ahead to the true-up) which, on the one hand introduces more granularity in model decision making to represent operational decision points but also require nuance in understanding and tuning.

3.  With all this, GENESYS may simulate momentary conditions that we would consider outliers. These outliers can be an hour or days in certain sets of future water conditions where the model simulates conditions outside of the allowed operational ranges. These outliers tend to relate to seam issues (and last/first hour or the day) and/or cascading complex conditions that are hard to satisfy in the model in the allowed computational time. In other words, while we always want ALL model results to be within operational constraints, there are model outcomes (since many operations are enforced by soft constraints) that we understand would not happen in reality. When this happens, we flag these to make sure they are not influencing adequacy results or any of our messaging around the operations of the system. For example, the following situations can sometimes occur.

• Example of such outliers can be outflow/spill that are extremely large (and not “physically possible”) but these are very rare (depending on project, ~0.002%-0.01%, and are computational artifacts not representing model operations.
• There can also be smaller magnitude outliers of outflow spiking, but these are also infrequent (0-0.9%) and tend to occur due to seam issues, not broad model operation.

4.  There are a few complex interactions that we do our best to capture given the modeling parameters and focus. These modeling decisions were made with experts throughout the region, including Bonneville.

• One example is the Hells Canyon Complex and how the model operates Brownlee and Dworshak in relation to Lower Granite/Lower Snake operations. We have greatly improved how we have modeled this over time, but it is likely a contributor to the spiking behavior that occasionally happens at the end of days/periods.
• A different example is the overall relationship with the Canadian projects. Where in actual operations the US and Canada have more control over requested within year Canadian flow release changes to support different water conditions, GENESYS does not. This can be challenging, particularly in years with exceptionally high water. Where, in reality, under those conditions there would be dynamic changes in the Canadian projects, the GENESYS constraints do not allow for these real-time dynamic changes. This can cause some strange operations that would not occur.
• Third, it’s possible our hydro reserve allocation is a bit different from actual operations. This has been an ongoing discussion with several folks (in and out of BPA) about how much MW each project is allowed to carry. We’ve made our assumptions public about hydro reserves and can revisit this in the future if we get specific new/updated feedback.
• Lastly, in a nod to points above and perhaps most important, GENESYS is tuned first and foremost to be an adequacy model with a highly granular and complex hydro operation. We have yet to be able to simultaneously perfectly capture every hydro plant behavior and have the model tuned for adequacy. Additionally, as emphasized since the 2021 Power Plan, GENESYS is not showing “how the system is operated right now” but rather “what things could look like if operated in broader regional coordination”.

5.  A different type of violation that is not necessarily an outlier are hydro constraints (obligations) that couldn’t be met, typically due to a combination of constraints and water availability that the model had to choose which constraint to adhere to at that given time. Unlike hard constraints that cannot be violated, soft constraints, with their relative penalties that reflect priorities, can be. One example can be Coulee’s forebay ramp constraint that in most observations isn’t violated, but infrequently can be in the model. This is a point we’re continuing to monitor and try to improve, and check that it doesn’t influence adequacy concerns when they occur. A different type of constraint not being met could be spill that is below the obligation level for that period, which is more associated with water availability, as occurs in real life is some water conditions. When reviewing results we keep these in mind and try to isolate constraint violation causes to determine what needs to be improved.

6.  Please keep in mind that the climate-informed modified flows used in the model show a shifting trend (see 2021 Power Plan Supporting Material), broadly expecting higher fall and winter flows and lower summer flows in the Dalles (which is a proxy for the overall system) due to increased precipitation as rain and lower snowpack during late fall and winter and lower precipitation during the summer in most of the region. With this overall change in water inventory, GENESYS may be operating (elevations, turbined, spill, generation) in a way that can be different from historic observations due to this shifting trend. As you will see in the simulated data, higher total outflows (and therefore higher generation) are especially observed in the main stem projects. However, this is not quite the case with the Lower Snake projects where the distribution of the simulated total outflows are not consistently higher than, and are more overlapping with, historic observations.

7.  There’s a topic we discussed in a previous RAAC/SAAC (Feb 27th, 2025) about GENESYS Modeling Updates that is worth calling out: an updated hydro budget methodology (slide 10). The approach we now use for setting the max and min elevation constraints of the large storage projects (colloquially the Big 8) relies on considering what we call their “historic envelope” of elevations (which we convert to usable storage as the constraint). For each of these projects we take the hourly elevations from 2013-2024 (we now have up to 2025, with different data sources between the US and Canadian projects when available). Then, the new minimum hourly storage constraint for a given year is the minimum across all years for that hour (for example, the min storage of Project X on Oct 1st at midnight is the minimum of all Oct 1st at midnight from 2013-2024. We do this for all hours in the year. And the process for maximum storage is the same. This historic envelope max and min is then used in the future value of hydro phase (which determines the weekly inventory) as well as the hourly phase to capture the appropriate constraints.