Supporting the Design of On-Site Infiltration Systems: From a Hydrological Model to a Web App to Meet Pluriannual Stormwater Volume Reduction Targets

Infiltration-based sustainable urban drainage systems (i-SUDS) often turn out to be simple and effective solutions for on-site runoff and pollution control. Their ability to limit the discharge to sewer networks or receiving waters can be broadly assessed in terms of (pluri)annual stormwater volume reduction. Although accepted as a relevant efficiency metric, this long-term volume reduction does not integrate well in design practices that have traditionally relied on event-based approaches. This article introduces a modeling framework, involving a hydrological model and machine-learning emulation, from which a web app was developed to allow practitioners to investigate the relation between i-SUDS design and pluriannual volume reduction efficiencies. The theoretical basis for modeling and a description of the web app are first provided. A diagnosis of the hydrological model is then conducted. The uncertainty caused by model parameters that do not directly relate to i-SUDS design is evaluated through a sensitivity analysis performed over multiple design scenarios. The latter is found to be highly variable and potentially significant, thereby justifying its explicit consideration in the web app. As part of this diagnosis, the impact of a shallow groundwater or a low-permeability layer on simulated volume reduction efficiencies is later evaluated to clarify the validity domain of the model. Practical recommendations on the minimum distance to shallow groundwater or low permeability layer, for the rainfall conditions considered in the web app, are given as a function of project size and the permeability of the soil media. The applicability of the web app is later illustrated from a selection of outputs. Its outcomes are finally compared to those of a simple design rule based on the combination permanent storage (as rainfall depths) and drawdown duration targets. Results confirm the inability of such simple design rules to fully capture pluriannual volume reduction efficiency and point out the risk of oversizing i-SUDS.

Stormwater infiltration in small vegetated systems can effectively reduce runoff and pollutant discharge to surface waters. A well-accepted performance objective for such systems is to achieve a significant reduction of the rainfall volume at the annual scale. However, integrating (pluri)annual volume reduction targets in design practices remains difficult as they do not accommodate well with the back-of-the-napkin, event-based calculations traditionally used by the stormwater profession. This paper introduces a web app that allows practitioners to easily investigate the relation between the design characteristics of infiltration-based solutions and pluriannual volume reduction efficiencies. The approach shows how machine learning can be used to replicate at low computational cost the outputs of specialized hydrological models and be incorporated in larger-audience tools. Through an analysis of the validity domain of the app, the study also points out the potential reduction of efficiency that may result from the presence of a shallow groundwater or a low-permeability layer, an aspect often overlooked in the design of infiltration-based systems. The applicability of the app is illustrated from different usage situations. The relevance of the proposed approach is finally demonstrated through a comparison to a simpler, event-based design method, which proves unable to adequately capture pluriannual volume reduction efficiency.