Stormwater capture systems have the potential to address many urban stormwater management challenges, particularly in water-scarce regions like Southern California. Here, we investigate the potential best-case limits of water supply and stormwater retention benefits delivered by a 10,000 m 3 stormwater capture system equipped with real-time control (RTC) on a university campus in Southern California. Using a copula-based conditional probability analysis, two performance metrics (percent of water demand satisfied and the percent of stormwater runoff captured) are benchmarked relative to (1) precipitation seasonality (historical rainfall and a counterfactual in which the same average annual rainfall is distributed evenly over the year); (2) annual precipitation (dry, median, and wet years); and (3) three RTC algorithms (no knowledge of future rainfall or perfect knowledge of future rainfall 1 or 2 days in advance). RTC improves stormwater retention, particularly for the highly seasonal rainfall patterns in Southern California, but not water supply. Improvements to the latter will likely require implementing stormwater capture RTC in conjunction with other stormwater infrastructure innovations, such as spreading basins for groundwater recharge and widespread adoption of green stormwater infrastructure.

Sensitivity analysis (SA) is commonly used to ascertain the relative importance of input variables, x = {x 1 , …, x d }, in determining the simulated output, y = {y 1 , …, y n }, of some vector-valued function or model, f(x) (Saltelli et al., 2010). A closely related practice is uncertainty analysis, which is concerned with quantifying the confidence (prediction) limits of the simulated output, y t ; t = (1, …, n). This usually involves the use of training data record, 𝐴𝐴 ỹ𝐲 = { ỹ𝑦1, . . . , ỹ𝑦𝑛𝑛} , and requires prior assumptions about the nature and distribution of the residuals, 𝐴𝐴 𝐴𝐴𝑡𝑡 = ỹ𝑦𝑡𝑡 − 𝑦𝑦𝑡𝑡 , and the sources of modeling errors (Beven & Binley, 1992;Liu & Gupta, 2007;Vrugt et al., 2005). Thus, uncertainty analysis quantifies the probability that a certain event (or range of events) will take place, whereas SA does not tell us anything about how likely a certain outcome of the model is. This essay is concerned with a general description of sensitivity across the confidence intervals of the simulated output. This so-called probabilistic SA (Oakley & O'Hagan, 2004) quantifies the sensitivity of simulated and/or forecasted event probabilities for a given multivariate probability distribution of the input variables. Our methodology relies on the analysis of covariance (ANCOVA) to preserve the multivariate character of the 𝐴𝐴 (𝐱𝐱, 𝐟𝐟(𝐱𝐱)) -relationship.SA is an essential step in model development, model calibration and quality assurance (Saltelli et al., 2020). In the past decades, many different SA methods have been developed and used in the mathematical and applied literature. Of these approaches, variance-based methods are particularly attractive because of their innate ability to characterize sensitivity over the entire prior ranges of the input variables and capacity to differentiate between the marginal, joint and total effects of x 1 , …, x d (Cukier et al., 1973;McKay, 1995;Razavi & Gupta, 2016;Sobol', 1990). The prototype of this approach, Sobol' (1990) method, was originally presented in Russian. In the English reprint, Sobol' (1993) showed that the output, y = f(x), of a scalar-valued square-integrable function, 𝐴𝐴 𝐴𝐴 ∈ 𝐿𝐿 2 ( 𝕂𝕂 𝑑𝑑 ) , on the d-dimensional unit cube, 𝐴𝐴 𝕂𝕂 𝑑𝑑 = {𝐱𝐱|0 ≤ 𝑥𝑥𝑖𝑖 ≤ 1; 𝑖𝑖 = 1, . . . , 𝑑𝑑} , can be uniquely decomposed into summands of elementary functions, f i (x i ), f ij (x i , x j ), …, f 12…d (x 1 , x 2 , …, x d ), under orthogonality constraints, to yield