Resilient water infrastructure partnerships in institutionally complex systems face challenging supply and financial risk tradeoffs

The study addresses how to design resilient and equitable collaborative infrastructure partnerships in complex, interconnected water systems facing rising scarcity, fiscal constraints, and regulatory pressures. Traditional planning relies on ad hoc coalitions and expected-value benefit–cost analyses using aggregated models, which fail to capture local-scale tradeoffs, partner-specific risks, and network interactions. The authors ask how multiobjective intelligent search, coupled with high-fidelity ensemble water system modeling, can discover partnership structures that balance conflicting goals: maximizing partner water supply gains, managing partner count/scale, minimizing worst-partner financial risk, and limiting negative impacts on non-partners. Focusing on California’s San Joaquin Valley (Friant-Kern Canal rehabilitation and a new groundwater bank), the purpose is to quantify tradeoffs, uncertainties, and heterogeneity in benefits/risks, and to compare optimized designs against a real-world status quo partnership.

The paper situates the problem within growing global water scarcity and declining higher-level funding, increasing local reliance on municipal debt and elevating financial risk for drought-exposed providers. Literature highlights potential benefits of collaboration and regionalization (economies of scale, diversification, access to capital), but also substantial risks due to hydrologic variability, infrastructural interdependencies, and complex institutions. Existing planning tools—aggregated models, expected-value benefit–cost analyses, and game-theoretic cost allocation—are limited when facing many potential partners, interactive network effects, deep uncertainty, and multiple conflicting objectives. Prior work often cannot resolve local-scale outcomes or partner-specific tradeoffs, and partnerships typically form via pre-existing relationships, risking regret. The study builds on advances in detailed water-resource simulation (e.g., CALFEWS), exploratory/robust decision frameworks, and multiobjective evolutionary search, aiming to fill gaps in partner-level assessment of benefits and financial risks under realistic hydro-institutional dynamics.

Case context: Tulare Lake Basin (San Joaquin Valley, CA), focusing on (1) rehabilitation/expansion of the Friant-Kern Canal (FKC) and (2) a hypothetical new groundwater bank near the FKC and Tule River. Local partner costs assumed at ~$50M for each project (FKC rehab part of ~$500M total with external subsidies; only $50M local share modeled). The groundwater bank capacity parameters in CALFEWS are set to ~25% of Kern Water Bank: infiltration capacity ~0.74 GL/day, pond storage ~0.37 GL, and recovery pumping ~0.25 GL/day. Financing assumes 30-year revenue bonds at 3% interest with equal annual payments: $2.55M/year per project, or $5.1M/year if both are built; O&M and water procurement costs are excluded (optimistic). Partnership decision variables: (i) project choice (canal, bank, or both), (ii) subset selection from 40 candidate water providers, and (iii) ownership shares (1–100%, sum to 100%), which determine priority access to capacity and proportional debt obligations. Modeling: CALFEWS simulates daily operations at water-provider resolution, including reservoirs, conveyance, interbasin transfers, water rights, and conjunctive surface–groundwater management and banking. Hydroclimate: a multi-site, two-state Gaussian Hidden Markov Model generates synthetic daily full natural flows for 15 watersheds, trained on a 110-year dataset, capturing wet/dry persistence and spatial correlation. Search phase uses 21 sampled 30-year sequences; reevaluation phase uses 79 additional sequences (total 100 sequences per partnership). Robustness check includes 20 downscaled CMIP5/VIC hydrologic sequences for 2021–2050 on two example partnerships. Objectives (tradeoff metrics): (1) number of partners; (2) captured water gain for partners (mean across scenarios); (3) captured water gain for non-partners (mean); (4) worst-partner cost of gains, defined as annual debt service divided by that partner’s captured water gain, assessed as the 90th percentile across scenarios and then max over partners. Optimization: Master–worker Borg MOEA on PSC Bridges-2 (four independent seeds, 96 h/seed, 4032 cores total), testing ~300,000 candidate partnerships. Non-dominated sets from each seed are combined and reevaluated on 79 additional sequences to yield a final Pareto set of 270 unique partnerships. Baseline comparison: a Status Quo Partnership modeled on the ongoing Friant Contractors’ FKC rehabilitation (16 partners, canal-only), with ownership shares proportional to historical Friant deliveries.

- Optimal partnership diversity and design: 270 Pareto-optimal partnerships spanning 11–24 partners (most common 16–19). Ownership shares are concentrated (few large, many small). 83% invest in both canal expansion and groundwater banking; 17% canal-only; none bank-only, indicating synergy between conveyance and storage. Providers near the FKC/Tule River frequently participate with larger shares. - Water supply gains: Partner captured water gains range from 48 to 85 GL/year (mean across scenarios), representing meaningful local-scale improvements that can support groundwater sustainability efforts. - Financial risk heterogeneity: Worst-partner cost of gains (90th percentile across hydrologic scenarios) ranges from $162/ML to >$1000/ML across partnerships. Larger partnerships tend to elevate worst-partner risks; all partnerships where every partner pays under $200/ML have 16 or fewer partners, and over half of 24+ partner designs have a worst-off partner >$1000/ML. - External impacts: Non-partner impacts vary widely, from +31 to −10 GL/year (mean). Partnerships with the largest partner gains often reduce non-partner deliveries, implying both surplus capture and intra-network reallocations. Excluding partnerships that reduce non-partner deliveries lowers achievable partner gains by about 6%. - Uncertainty: The Compromise Partnership (16 partners; canal+bank) exhibits wide single-scenario variability despite an expected 76 GL/year gain: 43–109 GL/year for partners; non-partner impacts from −81 to +26 GL/year; worst-partner costs from $105/ML to $407/ML (vs a 90th percentile metric of ~$180/ML). Benefits/risks are uneven across partners; some face worst-case costs >$300/ML, others < $100/ML, reflecting imperfect alignment between ownership shares and realized gains. - Baseline regret: The Status Quo Partnership (16 partners; canal-only) is dominated by the Compromise design: on average, 37% lower partner gains and slightly more negative average non-partner impacts. Critically, its worst-partner cost of gains exceeds $1000/ML in all 79 scenarios, with two providers frequently experiencing marginal or negative gains and extreme costs; five more exceed $1000/ML in some scenarios. In contrast, no Compromise partner exceeds $407/ML in any scenario, and in 90% of scenarios the worst-off partner pays < $180/ML. - CMIP5 robustness: Results under 20 downscaled CMIP5/VIC sequences are broadly consistent with the synthetic ensemble, with slightly higher gains and somewhat lower risks overall; differences between partnerships dwarf differences between ensembles. - Drivers of dominance: Compromise coupling of canal+bank (synergy), improved partner selection (including non-Friant providers and excluding marginal Friant participants), and better matching of ownership shares to realized gains explain superior performance.

The findings demonstrate that in institutionally complex, interconnected water systems, partnership design itself critically governs performance, with sharp tradeoffs between scale (number of partners), water delivery gains, partner-level financial risk, and external impacts on non-partners. High internal hydroclimatic variability makes realized benefits and financial liabilities highly uncertain over typical 30-year bond horizons. The results directly address the research question by showing that multiobjective intelligent search coupled with a detailed daily, provider-scale water system model can uncover high-performing, more equitable partnership structures that ad hoc planning and expected-value benefit–cost methods miss. Practically, water providers should include financial risk metrics—especially worst-partner exposure—alongside water reliability metrics, carefully assessing the limits of large-scale collaboration. Policymakers should anticipate and manage tradeoffs with non-partners through guidance, mitigation, or compensation mechanisms. The synergy between conveyance and storage argues for integrated planning across multiple investments. The study underscores the value of ensemble-based, high-resolution modeling and multiobjective search for transparent exploration of tradeoffs, uncertainty, and heterogeneity, improving the robustness and perceived fairness of cooperative investments.

The study contributes an integrated framework that combines high-fidelity ensemble water system simulation (CALFEWS) with multiobjective evolutionary search (Borg MOEA) to design cooperative infrastructure partnerships that balance water supply gains, partnership size, partner-level financial risks, and regional externalities. Applied to California’s San Joaquin Valley, the approach discovered 270 Pareto-optimal partnerships, primarily favoring combined canal expansion and groundwater banking. It revealed severe tradeoffs and substantial uncertainty/heterogeneity in partner outcomes. Compared to a real-world status quo partnership, an optimized Compromise Partnership achieved substantially higher water gains with markedly lower worst-partner financial risks, illustrating significant regret from ad hoc planning. Future work should expand to nonstationary climate scenarios and other uncertainties (demand, operations, infrastructure performance, costs), integrate stakeholder processes to capture institutional and legal nuances, incorporate equity and power asymmetries into objectives/constraints, and explore adaptive financing and contractual innovations (e.g., environmental impact bonds, dynamic cost-sharing) to manage financial risk. Building capacity for computation and decision support will be essential for broader adoption.

The analysis is intentionally optimistic: it models internal hydroclimatic variability using stationary synthetic sequences and largely omits nonstationary climate change, though limited CMIP5 tests suggest similar patterns. Financial modeling considers only debt service on capital (excluding O&M, water procurement, and potential legal/transaction costs), likely understating financial risks. Institutional/legal constraints (e.g., third-party injury determinations, permitting, litigation) are simplified; excluding injury-causing partnerships reduces achievable gains and stricter criteria would shrink feasibility further. CALFEWS, while high-fidelity, may not capture all local infrastructural, operational, and legal nuances; results depend on model assumptions and data. Ownership shares are optimized under stylized rules and may not reflect real negotiation dynamics. Computational demands are high, potentially limiting immediate transferability without specialized resources.