Objectives

A decision is ultimately a direct expression of the values of a decision maker (Keeney, 1992). Consequently, a first step to making an effective decision is to ensure a comprehensive and complete set of values are considered. Fundamental objectives in decision analysis represent values that may be affected by a decision and are what decision makers or the general public are interested in (Gregory and others, 2012; Runge and Walshe, 2014). Other types of objectives, like “means” objectives that are generally only important as intermediate steps or to signify progress toward the fundamental objectives, may distract participants, analysts, and decision makers from addressing objectives important to the decision-making process and lead to poor decisions (Runge and Walshe, 2014; Converse, 2020). Performance metrics, sometimes referred to as measurable attributes, are measures of the performance of consequences of the alternative management scenarios or actions with respect to the fundamental objectives. The fundamental objectives and relative importance of each (objective weights), as defined in this report, are only relevant to this specific decision context and are not meant to be universally applicable or transferrable to other decisions.

To prepare for in-person workshops, we held virtual briefings where individual agency participants were asked to prepare short descriptions of proposed objectives, performance metrics, and alternative management actions that are important to their respective agency or organization to be considered in the decision analysis (app. 1, table 1.1, fig. 1.1). A series of questions were provided to participants to help them identify fundamental objectives, including an example of means versus fundamental objectives. Most participants submitted draft ideas for objectives in writing before the first in-person workshop, including CDWR, NMFS, FWS, SWC, Reclamation, and the CVPC. These objectives were mostly considered means objectives (app. 1), which are a common focus of science-based organizations (for example, an emphasis on ecosystem drivers; Runge and Walshe, 2014). However, incorporating means objectives into a multicriteria decision analysis (MCDA) can lead to a lack of independence between performance metrics and double counting or other violations of analytical assumptions (Gregory and others, 2012; Runge and Walshe, 2014). Means objectives can be incorporated into models that are used to predict consequences. For example, some of the means objectives listed by participants early in the decision analysis related to specific demographic rates such as monthly survival or other linked drivers such as somatic growth, habitat suitability, or food resources for delta smelt (fig. 2). By themselves, the strength of each of these drivers may vary and contribution toward persistence of delta smelt is uncertain. However, when combined into a population model, drivers may be appropriate for predicting the consequences of alternative actions on delta smelt population dynamics (Smith and Nobriga, 2023). Population-level dynamics and persistence are more likely what decision makers ultimately care about (Runge, 2021). Thus, a first step in our analysis was separating means objectives from fundamental objectives and then incorporating them into appropriate predictive models (Runge and Walshe, 2014)—doing this and establishing performance metrics were the focus of our first two workshops.

The preliminary fundamental objectives and performance metrics were refined in a group setting during in-person or hybrid virtual/in-person workshop sessions on February 13, 20, and 21 and March 11 and 18, 2025, in Sacramento, California. We allowed participants an opportunity to share their objectives and openly discuss their values during these meetings. The participants agreed to population-level objectives for delta smelt, such as persistence or population growth, which represented the group’s fundamental objective. We also narrowed down other fundamental objectives related to water supply for the CVP and SWP, avoiding negative effects to other ESA-listed species, costs of nonflow actions, feasibility for 2025, and water quality and human health. Assuming that water could be delivered by the CVP or SWP, water infrastructure was sufficient to meet outflow habitat goals and serve many different customers, and because the CVP and SWP operate under different authorities, we separated water supply objectives for the CVP and SWP. We later removed feasibility and cost as objectives when the decision makers determined they did not expect that feasibility would be a factor when considering tradeoffs between alternative management actions (all alternatives proposed for analysis could be considered for implementation in 2025) and nonflow action management alternatives relating to costs could not be modeled in time to include in the MCDA composite score calculations at the end of the workshops (March 26, 2025).

The SDM analysis team (including authors BH, CP, BM) further refined objectives to ensure best practices for decision analysis were adhered to, to the extent possible (Gregory and others, 2012; Conroy and Peterson, 2013; Runge and others, 2013), while striving to ensure the objectives and associated performance measures met basic assumptions for MCDA (for example, objectives are clearly defined and preferentially independent; Gregory and others, 2012; Converse, 2020). The SDM analysts met with technical specialists from NMFS, Reclamation, CVPC, and the SWC to further refine objectives and performance metrics for salmonids, water quality and public health, and water supply. During a March 11, 2025, meeting with participants, we presented refinements we had made to the objectives and performance metrics and received additional suggestions for the final set of objectives. Feedback from participants involved the need to address water costs and specific water deliveries by type of contract, additional objectives related to operational flexibility for salmonids, and risks to the water supply in future years because of summer-fall flow actions. Some feedback represented specific values participants placed on reservoir storage that would allow them to meet operational objectives for a single or multiple salmonid species. For water conveyance metrics and water cost performance metrics not specifically addressed in our analysis, Reclamation agreed to provide outputs from the water planning model with parameter estimates to enable participants to interpret how the alternative management actions affected their specific interests. Final objectives involved delta smelt and water supply exports and objectives identified to minimize risk to water operations and minimize negative effects to operations to maintain habitat for State and federally listed salmonids (table 1). We describe each of the finalized fundamental objectives and the final performance metrics in detail in later sections (consequence estimation methods are described in the “Consequences” section below).

Alternative Management Actions

The participants largely focused on alternative management actions involving flow alterations because of their interest in having actions that can realistically be implemented in the summer-fall of 2025. The three types of flow actions considered were: (1) one that pertains to a specific target for delta outflow and the amount of freshwater moving from the Sacramento-San Joaquin Delta (delta) downstream into Suisun Bay; (2) one that targets a particular X2 (where near bottom salinity is two practical salinity units measured as kilometer distance from the Golden Gate Bridge in San Francisco, California); and (3) one pertaining to how the Suisun Marsh salinity control gates operates, a tidal gate facility that can bring in freshwater into a brackish tidal marsh region downstream from the delta (for example, Suisun Marsh). A couple of nonflow actions were also brought up during the discussion: treatment or removal of aquatic weeds from parts of the delta and drainage of water from managed tidal wetlands in Suisun Marsh.

Lower salinity and higher outflow conditions in the system can benefit delta smelt in a few ways. First, lower salinity in Suisun Bay and Suisun Marsh may lead to higher numbers of delta smelt in both regions. Suisun Bay and Suisun Marsh are more turbid than the delta, which has experienced major reductions in turbidity affiliated with the proliferation of invasive aquatic plants (Hestir and others, 2016). Turbidity is purported to be an important modulator of delta smelt’s success or failure within the food web (reviewed in Smith and Nobriga, 2023). Suisun Bay is also generally 1–2 °Celsius cooler than the delta, which is often warm enough to limit delta smelt growth. Additionally, for nearly 40 years, a Potamocorbula amurensis (nonnative bivalve, brackish-water corbicula) has depleted the pelagic productivity of low-salinity habitat indexed by X2 (Kimmerer and Thompson, 2014; Kratina and others, 2014), leaving it dependent on zooplankton subsidy from the delta to feed pelagic fishes like delta smelt (Kimmerer and others, 2019). Zooplankton subsidy increases with higher outflow (Hassrick and others, 2023), and thus, may provide added benefit to delta smelt when outflow is increased and/or X2 is lowered.

For flow actions in the summer (June–August), the participants focused on delta outflow targets based on the findings in Polansky and others (2024). The target value of 10,200 cubic feet per second (ft³/s) was chosen based on the 25 million square meters per day summer delta outflow described by Polansky and others (2024) as being associated with a 50 percent probability of population growth. Some alternative summer delta outflow targets used 20,400 ft³/s instead because that value appears to be associated with an about 75 percent probability of population increase (Polansky and others, 2024). Polansky and others’ (2024) life cycle model used overall average summer delta outflow from June to August as a predictor. As such, an evenly distributed flow across the 3 months might not be needed or if the action mimics a more historical flow pattern, where flow in the month of June is higher than flow in the months of July and August. Both types of summer action alternatives were created to evaluate the performance of these two differing hydrologic conditions (table 2).

Actions surrounding delta outflow in the fall months (September–October) followed the “fall X2” action framework, where the water projects would target X2 at a particular location rather than outflow rate. The fall X2 action was originally described in the FWS 2008 Reasonable and Prudent Alternative (FWS, 2008) and was motivated by concerns about abiotic habitat for delta smelt being compressed in the fall months (Feyrer and others, 2011). X2 at 74 km would be roughly aligned with Chipps Island, whereas X2 at 80 km would be about in the confluence between the Sacramento and San Joaquin Rivers. Both locations are downstream from the Sacramento-San Joaquin Delta and are based on the purported mechanism of the flow action, X2 at 74 km may offer more benefit to delta smelt than X2 at 80 km.

Suisun Marsh is one of the largest contiguous tidal marshes in the San Francisco Bay-Delta and thought to be an important rearing area for delta smelt (Hammock and others, 2015). The general seasonal pattern for the Suisun Marsh is for salinity to increase in summer and fall as freshwater inflow decreases. A tidal gate facility towards the upstream end of Suisun Marsh, referred to as Suisun Marsh salinity control gates, is often operated in wintertime to decrease salinity for managed waterfowl habitat. Use of the Suisun Marsh salinity control gates in the summer-fall period may provide delta smelt with access to a more turbid habitat (Sommer and others, 2020). Under current operations, the Suisun Marsh salinity control gates are operated for delta smelt for 60 days from June to October whenever a salinity threshold is exceeded during above normal, below normal, and dry years after wet or above normal years (refer to U.S. Bureau of Reclamation, 2024b). The gates are also operated for 30 days in dry years after below-normal years. The gates can be operated in alternate ways. For example, the gates can be operated continuously for the 60 days, or they can be operated using a 7 days on and 7 days off schedule to extend how long they operate (albeit at the cost of higher salinity overall with this extended period).

For nonflow actions, aquatic weed control may benefit delta smelt by increasing turbidity via the removal of invasive aquatic plants prevalent in the delta (Hestir and others, 2016), though the exact method (for example, mechanical removal versus herbicide treatment) was not discussed or specified by participants during the workshops. Another nonflow action brought forth by participants was the managed wetland flood-and-drain operations in Suisun Marsh, which can promote food export from the managed wetlands to adjacent tidal sloughs. The group of participants had some disagreements about whether these nonflow actions could be implemented in 2025; however, the participants agreed that alternatives, including nonflow actions, should be analyzed and evaluated. We could not fully evaluate the effects of nonflow actions during the workshops because of a lack of information necessary for modeling the consequences of nonflow actions. However, additional data were gathered after the last workshop to allow for a post hoc analysis of the effects of portfolio alternatives with aquatic weed control. We discuss the effects of two nonflow action alternatives in appendix 2. The description of alternatives considered in our decision analysis is displayed in table 2.

Consequences

In decision analysis, consequence estimation involves making predictions about how the alternative management actions perform for each objective, so decision makers can later choose a management action that may most likely lead to a desirable future outcome. Consequences can be estimated using expert elicitation, or quantitative models, if available (Hemming and others, 2018). We used available quantitative models to estimate consequences for each of the 8-performance metrics for the objectives. The temporal scale for consequence predictions differed between performance metrics, but we assumed alternative management actions considered in our analysis would be implemented in an above normal or wet year, and the effects of alternatives we considered in our analysis for water year 2025 may affect reservoir operational flexibility and coldwater pool objectives in 2026. Our modeling team consisted of technical experts from Reclamation, with technical support provided by FWS and CDWR technical experts.

We used the California Simulation Model (CalSim; CDWR and Reclamation, 2002; Draper and others, 2004), which is the water operations model used by Reclamation and CDWR for CVP and SWP water operations planning, to analyze alternative water management actions in our study. For our analysis, we used the most recently updated version of CalSim—CalSim 3 (CDWR, 2022). The primary input to CalSim 3 consists of a 100-year (1922–2021) historical sequence of empirical monthly river flows in California’s Central Valley, which can then be adjusted to represent numerous user-defined water demand and management scenarios from which the model generates hydrologic outputs over multiple years (CDWR, 2022). Some of the user-defined input variables include agricultural, urban, and managed wetland water demands, return flows, and groundwater recharge rates. Based on this array of inputs, CalSim 3 calculates river and stream flows, water diversions and return flows, reservoir storage, deliveries to water users, salinity at Suisun Marsh and its corollary (X2), and delta outflow—these outputs, which include reservoir storage volume exceedance probabilities, provided the basis for additional modeling to predict consequences for delta smelt and winter-run Chinook salmon redd dewatering, as we describe below. We used the 2024 set of environmental regulations as a baseline for our analysis of consequences (refer to U.S. Bureau of Reclamation, 2024a). A detailed description of CalSim 3 is available from the CDWR: (https://data.cnra.ca.gov/dataset/calsim-3-0-release).

Estimation of Consequences for Delta Smelt

Consequences of the alternative management actions to delta smelt minimum abundance were estimated using a published individual-based life cycle model (IBMR, Rose and others, 2013) implemented in R language (R Core Team, 2019). The model output includes an estimate of the population growth rate (λ), which relies on the prior and projected year estimates of delta smelt abundance. To translate hydrologic information into ecological phenomena that could be represented in the IBMR, the CalSim 3 outputs were fed into three submodels, which were used to dynamically adjust (1) the delta smelt spatial distribution, (2) regional salinity, and (3) zooplankton abundance and distribution (fig. 3).

The delta smelt distribution submodel was used to estimate changes in the species’ distribution based on monthly average X2 predicted by CalSim 3. The delta smelt distribution submodel predicted the proportions of delta smelt in the 12 IBMR regions based on a given X2, using historical fish survey data and a nearest neighbor approach. When a change in X2 was predicted, delta smelt spatial distributions were adjusted by substituting the observed spatial distributions with the historical spatial distribution at X2 closest to the predicted X2. Although we extrapolated changes in the distribution of delta smelt based on X2 for every month, X2 (and therefore distributions) did not vary much between alternatives aside from the summer-fall months.

The second submodel was an X2-salinity submodel—a generalized linear model that predicts the monthly mean salinity at the 12 IBMR regions based on X2 location. Changes in salinity, caused by X2 being manipulated for the alternative water management actions, were only expected to happen in the confluence region and farther downstream; therefore, zooplankton variation as a function of change in salinity was only estimated for the confluence and downstream regions. Salinity information produced from the X2-salinity submodel was then fed into the third submodel, which governs the connection between salinity and zooplankton. The salinity-zooplankton submodel consists of a set of generalized additive models that predict zooplankton density for the downstream IBMR regions based on a given salinity and day of the year. Predictions from the salinity-zooplankton submodel were converted to scalars representing the change in these predicted zooplankton densities from the observed densities and were then applied to the original IBMR prey data for each taxonomic group, alternative management action, downstream IBMR region, and month between 1995 and 2014.

For each alternative, IBMR was run for 330 iterations using the relevant data input (for example, zooplankton scalar values, and X2), while assuming the actions included in each alternative were implemented in above normal or wet years during the 20-year simulation time series. For each 20-year time-series iteration (informed by observed data collected between 1995–2014), the smallest adult cohort size was retrieved and then averaged across the 330 iterations for every alternative management action to provide the minimum population size as a measure of persistence.

Consequences of the alternative management actions to delta smelt minimum abundance ranged widely from about 2,559–2.1 million for the 2022 Reclamation median hydrology to 69,929–1.6 million for the CDWR baseline hydrology (table 3). Only two alternatives— “maximize delta smelt even flow” and “maximize delta smelt historic”—led to minimum abundances that would likely allow for sufficient brood stock collection (for example, those above 400,000).

Table 3.    

Consequences of alternative management actions and competing California Simulation Model (version 3 [Calsim 3]) baseline hydrology models on the delta smelt objective of 2025–26.

[The columns include delta smelt minimum abundance estimates predicted for 2025–26 using the Individual Based Model in R (IBMR) and inputs generated by the California Simulation Model (version 3 [Calsim 3]) with two alternative baseline hydrology model parameterizations (California Department of Water Resources baseline hydrology [CDWR, 2023] and the U.S. Bureau of Reclamation (2022) median hydrology). Greater values indicate better performance and are indicated by a darker color of green.]

Table 3.    Consequences of alternative management actions and competing California Simulation Model (version 3 [Calsim 3]) baseline hydrology models on the delta smelt objective of 2025–26.

Alternative	California Department of Water Resources baseline hydrology	U.S. Bureau of Reclamation median 2022 hydrology
Average minimum delta smelt abundance
1. Status quo	163,397	4,927
2. Maximize delta smelt-even flow	1,626,847	1,791,315
3. Maximize delta smelt historic	1,584,764	2,083,525
4. Summer-fall even flow	96,007	12,695
5. Summer-fall historic	69,929	8,665
6. Summer-even flow	192,681	11,038
7. Summer historic	157,976	9,303
8. Summer even flow—Alternative Suisun Marsh salinity control gate action	188,239	12,276
9. June action	81,969	3,131
10. Maximize water supply	111,062	2,559
11. Maximize water supply—No Suisun Marsh salinity control gate action	95,303	3,175

Relations Between Water Supply Cost and Storage Performance Metrics

There was considerable discussion among participants during the workshops about other objectives or metrics we did not carry forward as fundamental objectives. Many of those that we considered and discussed in early workshops represented different water cost or storage metrics. However, some participants were accustomed to considering these metrics during previous water management planning efforts, which they used to interpret the consequences to the fundamental objectives. We assessed the relations between the performance metrics representing the fundamental objectives in our analysis with other potential performance metrics representing water deliveries and water costs using correlation analysis (Pearson’s r statistic) in R (R Core Team, 2019). We considered metrics to be highly correlated if their r statistic was >0.7, and the metrics were highly correlated, as all r statistics were >0.83. We presented correlations between metrics for these means objectives and the metrics we retained in our analysis to represent fundamental objectives in workshop four on March 18, 2025 (fig. 4).

Swing Weighting

We used swing weighting to elicit participants’ objective weights before calculating composite utility scores (Gregory and others, 2012). Swing weighting incorporates the range of consequences across alternative management actions and the fundamental importance of the eight objectives when assigning objective ranks and scores. The SDM analysts presented an overview of (raw) performance metric scores by alternative management action and provided training in swing weighting techniques during workshop 4 on March 18, 2025. During the workshop, we also provided an overview of the interpretation of each performance metric and facilitated discussion to avoid ambiguity in participants’ understanding of the meaning of the metrics before participants completed swing weighting for their agency or organization. For swing weighting, we used the local scale to represent the range of consequences across alternative management actions and hypotheses (assuming equal weights on the hypotheses), which was defined by values for each objective in tables 3–10. A swing weighting form, including the best- and worst-case performance metric scores for each objective, was provided electronically to all participants to elicit objective ranks and weights (fig. 5). To assist in deliberations about the range of performance, the form included hypothetical alternatives consisting of the best-case for one objective, the worst case for all others, and one alternative reflecting the worst-case performance for all objectives. We asked participants to consider if they could “swing” just one objective from worst- to best-case, which would it be? The hypothetical alternative with the best performance for their top-ranked objective would be ranked number one. We then asked them to repeat this ranking for all remaining hypothetical alternatives 2–9 (with the ninth ranked alternative being the worst case for all objectives). Finally, we asked participants to convert their ranks to scores by assigning 100 points to their top-ranked objective and then assigning a score to each subsequent objective according to relative importance. The swing weighting form automatically standardized these scores on a 0–1 scale to represent objective weights for composite score calculations. Participants were given about ~3 days to complete the swing weighting exercise, and the SDM analysts offered to hold one-on-one discussions with each participant, before summarizing results in plenary, to confirm the swing weighting instructions were clear, to minimize ambiguity in performance-metric interpretation, and to ensure results represented the values and expectations of each participating agency or organization.

We met with five of the eight agencies or organizations invited to participate in the SDM process to discuss their objective weights, and four chose to share their weights and alternative composite scores and rankings. Of these, only Reclamation and CDWR shared their weights publicly, and the others were anonymized. Some participants did not feel qualified to represent their organizations’ values or expressed other concerns about the objective weighting process. Weights assigned to the objectives by the participants differed considerably (fig. 6); regardless of whether participants expressed their objective weights for final MCDA composite score calculations and alternative management action ranking, each discussed their rationale for their objective rankings and weights in plenary. None of the participants ranked delta smelt higher than a two-way tie for second, and one ranked the delta smelt objective seventh, despite the wide swing in consequences for delta smelt abundance relative to the other performance metrics (refer to fig. 5). Some participants were skeptical about the range of delta smelt abundance predictions because the model was calibrated with monitoring data for wild fish, and they questioned if the same population dynamics would be expected for hatchery propagated fish. Another participant stated they were more confident in management outcomes for salmonids than for delta smelt and accordingly placed more weight on salmonid-related objectives. Some participants also viewed tradeoffs between the water quality objective, public health objective, and others as “taboo” (for example, Daw and others, 2015), which caused them to place higher weight on the objective despite the worst-case scores being equivalent to or above “acceptable” levels. At least one participant recognized the acceptable range of water quality and human health scores and put lower weight on the objective as a result, and objective weights on water quality/public health were generally lower than for others (average weight =0.058, range 0–0.12; fig. 6). When assigning weights, other participants considered operational flexibility that was not incorporated in our modeling but could be applied in realtime to mitigate negative consequences to salmonids; winter-run Chinook salmon and steelhead objectives were weighted less in this case. At the same time, the risk for negative consequences appeared high for the SWP coldwater pool objective (71 percent swing between the best- and worst-cases), which was ranked higher by another participant. Discussions by the participants about the rationale of weighting reflected the difficult tradeoffs that had to be considered.

Multicriteria Decision Analysis

We calculated composite utility scores (V) using objective weights provided by the four participants and normalized performance metric scores for the eight objectives using a linear value model (Gregory and others, 2012; Converse, 2020):

where x_i,j,k is the normalized performance measure score on objective i for alternative j under hypothesis k (representing alternative baseline hydrology models) and w_i is the weight individual participants place on objective i (where the sum of the objective weights =1; table 12). Given the equal hypothesis weights that were applied to the competing baseline hydrology models, we used the average composite score calculated with performance metric scores under each competing model to rank the alternative management actions (table 13). An advantage of MCDA is its ability to compare the performance of alternatives using performance metrics on different scales; however, before composite scores are calculated, all performance metrics must be normalized on a 0–1 scale, where 1 maps to the maximum or best performance score predicted and 0 to the worst-case or minimum desired score known as local scaling. This local scaling has some drawbacks in that small differences between alternative performance metric scores can be exaggerated leading to nonsensical decisions if participants fail to account for the range of consequences when assigning ranks and scores to objectives (Monat, 2009). To minimize this potential issue, we provided the range of scores on the local scale and articulated how the predicted consequences for each performance metric should be interpreted before swing weighting.

Assuming equal weights on the competing baseline hydrology models and using the objective weights in table 12, we calculated composite utility scores for each alternative, and they are presented in table 13 (Healy and others, 2025). The top-ranked alternative was fairly robust compared to differences in participants’ objective weights. The “maximize water supply with no Suisun Marsh salinity control gate” alternative ranked highest for three of four participants, the “maximize water supply” alternative was the top-ranked alternative using Reclamation’s objective weights, and the “June action” alternative was tied for first for participant B. These three top alternatives for individual participants were also ranked in the top three for all four participants, and differences in composite scores for the top two alternatives were within 0.001–0.002. The lowest ranking alternative was consistent for all four participants (maximize delta smelt-even flow)—the two alternatives designed to maximize delta smelt were dominated by the others in all cases.

We assessed the sensitivity of the alternative rankings to the weight placed on the delta smelt objective and the combined weight placed on the two water export objectives. Each participants’ weights on the delta smelt, CVP exports, and SWP export objectives were similar (table 12). However, the water supply consequences are correlated (fig. 4) because the logic for CalSim 3 for determining the State and Federal contributions to a flow action follow the terms of the Coordinated Operations Agreement between CDWR and Reclamation (CDWR, 2022). The high correlation between the two water export fundamental objectives could lead participants to place more weight than intended on those objectives relative to the other objectives. Figure 7 allows assessment of the sensitivity of each participant’s top-ranking alternative to variations in the combined weight they placed on the export’s objectives relative to the delta smelt objective. We varied the combined weight on exports from zero to one, with the combined export weight being made up of the individual participant’s relative weightings on the CVP and SWP export objectives. When the combined exports and delta smelt objective weights summed to less than one, we distributed the remaining weight to the other five objectives proportional to how each participant weighted those objectives against each other. The relative weight of delta smelt tended to be between one-third and one-half of the weight of the combined exports (fig. 7, grey points). The robustness of each participants’ top ranked alternative can be inferred by either increasing the weight on the delta smelt objective (up on the y-axis) to be more like the combined export weight or reducing the weight on the combined exports (left on the x-axis) to be more similar to the delta smelt weight.

Sensitivity to Baseline Hydrology Models and Value of Information Analysis

We assessed the sensitivity of the alternative rankings to differences in the two baseline hydrology models by using a hypothesis-weighted average score for each alternative (outcome under uncertainty, V_j). Hypothesis weights (probability, p) are the relative belief in each competing hypothesis. The alternative scores are a sum of the product of V_j,k, and the weights applied to the two baseline hydrology hypotheses (p_k)

We also calculated the EVPI, which is the difference between the expected management performance after uncertainty has been resolved and the expected performance under uncertainty as:

The top-ranking management alternative was sensitive to weight on the baseline hydrology hypothesis for all four of the participants’ objective weights (fig. 8). However, no new alternatives were ranked highest—the same three alternatives ranked in the top three compared to when hypothesis weights were considered equal (refer to table 13), including the “maximize water supply” alternative (alternative 10), the “maximize water supply with no Suisun Marsh salinity control gate” action (11), and the “June action” (9).

Little VOI was associated with reducing the uncertainty between baseline hydrology models. EVPI was <1 percent for all participants’ objective weights, with EVPI ranging from 0.07–0.98 percent (mean=0.511 percent), indicating <1 percent increase in management performance would be achieved if the uncertainty in the true baseline hydrology model was resolved before decision making.

Consequences of Alternatives Including Nonflow Actions

After the final SDM workshops and initial analyses as described in previous sections, the SDM analysts were able to obtain data necessary to do a consequences analysis for two additional alternatives that included nonflow actions meant to improve habitat for delta smelt. Aquatic weed control actions were analyzed in detail in a previous decision analysis, which assumed aquatic weed control would lead to enhanced turbidity and therefore benefits to delta smelt (Crawford and Rudd, 2024). The two nonflow action alternatives we analyzed included aquatic weed control for 3,470 acres, assuming 20 percent efficiency in weed removal, which we combined with actions included in the “maximize water supply” (alternative 10) and “maximize delta smelt-even flow” (alternative 2) alternative management actions (table 2.1). Assuming equal weights on the baseline hydrology models, we discuss the effects of aquatic weed control on delta smelt minimum abundance relative to effects for comparable alternatives without nonflow actions (for more details, refer to app. 2).

Using the IBMR model and population projection methods as described in previous sections, we found higher delta smelt minimum abundance for alternatives that included aquatic weed control compared to the “maximize delta smelt even-flow” and “maximize water supply” alternatives without aquatic weed control. The minimum abundance for delta smelt for the “maximize delta smelt-even flow” alternative was 2.62 million (a 53 percent increase, refer to table 11) and 56,810 (306 percent increase) for the “maximize water supply” alternatives with aquatic weed control. There was also considerable difference in minimum delta smelt abundance between baseline hydrology models. The range of minimum delta smelt abundance between the two baseline hydrology models for the “maximize water supply with aquatic weed control” alternative was 17,216–445,016, indicating sufficient broodstock could be collected under the adjusted historic baseline hydrology model (for example, abundance >400,000). For the “maximize delta smelt-even flow with weed control” alternative management action, the range of minimum delta smelt abundance was 2.24–3.01 million between the two hydrology models.

If alternatives that included aquatic weed control had been included as part of the MCDA with participants that provided objective weights, the “maximize water supply with aquatic weed control” alternative management action would be the highest-ranked alternative for the four participants (CDWR, Reclamation, Participant B, and Participant C) (refer to app. 2, table 2.2). Including aquatic weed control in additional portfolio alternatives may have led to optimized outcomes because the nonflow action from the weed control benefitted delta smelt without the need to curtail water exports or release additional water from reservoirs. However, we did not include the monetary cost associated with this nonflow action in the consideration for this decision problem. How the participants would have weighted objectives (and in turn how the alternative rankings would change) if the financial cost associated with the aquatic weed control action had been included in the MCDA is unclear. Based on information from the California Division of Boating and Waterways that manage aquatic weed control in the delta, 3,470 acres of aquatic weed control could cost anywhere between $31,230,000 to $72,870,000 for the first 3 years of the action (Crawford and Rudd, 2024).

The aquatic weed control action modeled in this study had 20 percent efficiency when removing weeds, which represents an optimistic “best-case” scenario (Crawford and Rudd, 2024). However, the efficacy of aquatic weed removal in the tidal system remains uncertain; a recent study in the delta demonstrated that the widely used herbicide fluridone did not lead to significant reductions in aquatic weed abundance (Rasmussen and others, 2022). Furthermore, participants suggested that aquatic weed removal as proposed would not likely be implementable in 2025 because of regulatory and logistical constraints and that it may take years to develop a successful action.

Summary of Effects to Disadvantaged Communities

The CVP and the SWP provide water to numerous municipalities in California, including community water systems that directly serve households. Some of these communities may be classified as disadvantaged communities (DAC) or severely disadvantaged communities (SDAC). DAC/SDAC status is determined based on a percentage of the statewide annual median household income (MHI). California Water Code (Section 79505.5a of the California State Water Code) defines communities as disadvantaged if their annual MHI is 80 percent of the statewide MHI and as severely disadvantaged if their annual MHI is 60 percent of the statewide MHI, and MHI data are obtained from U.S. Census data. A statewide map of DAC and SDAC communities is available through the State Water Resources Control Board (SWRCB; SWRCB, 2023 fig. 10-1). The CVP and SWP deliver water to communities defined as DAC or SDAC of different estimated population sizes (table 14).

Because the DAC/SDAC populations sizes differ substantially between SWP and CVP service areas, there may be different effects on these communities between alternative management actions evaluated by the SDM group. For example, if any of the alternative management actions that include flow actions listed in table 2 rely to a greater extent on SWP-supplied water, there could be a greater effect on DAC/SAC populations than if the water for the action is achieved primarily through the CVP.

The dataset for this analysis was shared with CDWR by the SWRCB upon a CDWR request in August 2021 (at the time of publication, data were not publicly available from SWRCB). CDWR staff provided the estimated list of community public water systems that were contractors to the SWP and CVP and requested that the SWRCB staff query their internal version of the database to identify all water systems that purchase or receive water from those contractors. As requested, the SWRCB’s query output from the Safe Drinking Water Information System provided what public water systems purchase or otherwise import water from this set of project contractors. The query output also includes systems that purchase or otherwise import water from those subcontractors, down to five levels of receiving water. This dataset does not indicate how much each system relies on project water, only that the system relies on project water as one of its water supply sources at the time of the data request (August 2021). DAC/SDAC status is based on an analysis run by the State Water Board staff for exploratory use only. Updated data were received by CDWR on February 27, 2025, from the SWRCB.

Effects to DAC/SAC populations are linked and correlated with the CVP and SWP water supply objectives because the effects on DAC/SAC populations are directly related to project deliveries. Therefore, the effects are briefly summarized in this section and not evaluated through a consequence analysis as they were for the fundamental objectives. Alternative management actions 2–5 in tables 4 and 5 (“maximize delta smelt—even flow,” “maximize delta smelt historic,” “summer-fall even flow,” and “summer-fall historic”) present reductions in the water supply for CVP and SWP compared to “status quo” conditions (alternative 1). Alternative 2 (“maximize delta smelt—even flow”) represents the maximum reduction in water deliveries across CVP and SWP projects at 521 taf (tables 4 and 5), and about 70 percent of this reduction is achieved through reductions in the SWP deliveries. For all other scenarios presenting a reduction in deliveries, the proportion of reduction is the same, with about 70 percent coming through the SWP. This disproportionate reduction in deliveries from SWP indicates that DAC/SAC populations are more likely to be affected (for example, vulnerable to felt effects of water supply reductions). However, the modeled delivery amounts for the SWP across all alternative management actions were assumed to be above minimum health and human safety standards (for example, because of constraints in CalSim 3); therefore, the effects on DAC/SDAC populations are not considered instrumental to decision making or alternative management action rankings based on this analysis. In the future, to evaluate flow management alternatives involving reduced exports to DAC/SDAC populations, this analysis suggests that these parameters can be considered to determine relative contributions from SWP and CVP sources because of their disproportionate levels of service to these populations.

Considerations for Future Decision Analysis

Defining fundamental objectives and eliciting participants’ weights on the objectives was an important focus of our analysis. Future analysts could consider more facilitation with participants to define fundamental objectives specific to future decision contexts and population-level performance metrics sensitive to water management alternatives for listed salmonids, including steelhead and winter-run and spring-run Chinook salmon. A decision-support model could be identified or developed to translate the life stage-specific effects (for example, redd dewatering) of the alternatives to population-level consequences such as persistence probabilities or abundance (for example, Peterson and Duarte, 2020), which more closely aligns with Central Valley Project Improvement Act (CVPIA) management goals. This approach may address concerns raised by participants related to difficulties in interpreting effects, such as redd dewatering, on the overall winter-run Chinook salmon population, for instance. Some participants also represented broad interests that made it difficult for them to reach consensus on objective weights within their organizations. Groups could be further subdivided, if necessary, to allow representation of a broader range of organizational values (for example, objective weights). We also used the local scale, involving the best and worst cases predicted in our decision context, for swing weighting and normalizing performance scores; defining global scales, meaning the inherent range a decision maker might associate with an objective (Monat, 2009), for each performance metric before swing weighting might reduce uncertainty some participants faced when being asked to provide appropriate weights on a metric intrinsically important to them but with a very small range of consequences (for example, water quality/human health).

Predictions of the consequences to elements of complex systems, including the delta and its tributary river systems, are fraught with uncertainty (Peterson and Duarte, 2020; Healy and others, 2024)—we addressed the sensitivity of the decision to only a single source of uncertainty (baseline hydrology assumptions). Nevertheless, participants raised multiple concerns about uncertainty in predictive models and how predictions should be interpreted. Future analyses could more thoroughly address the sensitivity of the decision to uncertainty using value of information (VOI) tools and metrics, including the expected value of partial information (EVPXI; refer to Runge and others, 2011). EVPXI is a measure of the importance of different sources of uncertainty, and this knowledge can then be used to prioritize future monitoring or research. Interest in adaptive management and learning was also expressed by some participants—VOI is an important precursor to formal adaptive management and involves understanding whether uncertainty is important (whether the decision changes depending on which hypothesis is true), and this knowledge can then also inform future research, monitoring, and learning from recurrent decisions related to water management. Newly developed methods for rapid VOI analysis using constructed scales (Runge and others, 2023) may also be useful for decisions about water management in the Central Valley. Because of time constraints, we were unable to fully assess the effects of decisions made in this decision context to future decision making. Dynamic programming may also allow for condition-dependent decision making that considers effects to resources across years (Runge, 2020) and uncertainty in hydrology in a risk assessment framework, which may be a better approach over the long term than treating the decision as a one-off, given the dynamic nature of the hydrology and the use of management in light of the risk of multiyear drought conditions.