Inflow Generation for Water System Planning in Multisite Studies Using Target Synthetic Streamflow Generation

Abstract

Robust water-system planning under deep inflow-condition uncertainty requires synthetic streamflow scenarios that capture a wide range of target changes in magnitude, variability, and seasonality; the primary drivers of water-system operations. Targeting these properties ensures that streamflow scenarios align with site-specific operational requirements and effectively capture water-system vulnerabilities. Climate-driven projections provide scenario diversity but do not offer control over streamflow properties. Stochastic generators cannot control streamflow properties, and optimization-based frameworks that do so are computationally expensive. This study introduces a computationally efficient optimization-based framework for generating statistically credible synthetic streamflow scenarios that match target monthly mean and standard deviation at multiple inflow locations. By optimizing the parameters of a widely used streamflow generator, the framework: (a) directly targets streamflow magnitude, variability, and seasonality, (b) enforces physical and mathematical feasibility of targets scenarios, (c) uniformly covers the exposure space, avoiding redundancy and ensuring no plausible condition is overlooked. Monthly sequences are disaggregated to daily using k-nearest neighbors method. The framework is demonstrated in the Winnipeg River Basin, Canada, generating synthetic inflows across 14 locations under historical and future GCM-driven conditions. Results show that optimization improves matching of target statistics and enables uniform exposure space sampling, while preserving or modifying autocorrelation and cross-correlation structures depending on target seasonality. Beyond offering a computationally efficient alternative to prior optimization-based frameworks, the approach complements top-down method: while GCMs provide physically plausible futures, the proposed framework enables transparent, baseline-consistent, and broader exploration of uncertainty informed by top-down insights, supporting stress testing and risk-informed decision-making under deep inflow-condition uncertainty.