publications image

Low-Impact Development - San Francisco’s Green Approach to Stormwater Management Promises to Reduce Energy Use, Increase Natural Habitat, and Enhance the Quality of Urban Life

Low-Impact Development - San Francisco’s Green Approach to Stormwater Management Promises to Reduce Energy Use, Increase Natural Habitat, and Enhance the Quality of Urban Life

Lori Kennedy, Lydia Holmes, Steve McDonald, Rosey Jencks, and Greg Braswell
Published In: 
Water Environment & Technology Magazine,  
April 2008

San Francisco is one of only two cities in California with a combined sewer system. The system is the oldest on the West Coast, with most sewers more than 70 years old. While it meets current regulatory requirements, the aging infrastructure is occasionally overwhelmed by wet weather flows, resulting in street flooding and combined sewer overflows (CSOs). The San Francisco Public Utilities Commission (PUC) is developing a Sewer System Master Plan, a 30-year roadmap for sewer system improvements. The master plan aims to address many challenges, such as aging infrastructure, odors, and CSOs, while improving the sewer system and quality of life for city residents.

As part of the master plan, PUC requested an analysis of low-impact development (LID) to reduce wet weather flows into the combined sewer system. LID is an innovative approach to stormwater management that relies on decentralized, small-scale stormwater facilities and site-design techniques to reduce stormwater volume and peak flow rates, as well as to remove stormwater pollutants. Unlike traditional stormwater management, LID relies on natural hydrologic systems to slow down, capture, infiltrate, and treat rainwater where it falls. Typical LID practices for urban areas include ecoroofs, vegetated swales, rain gardens, permeable pavement, urban forestry, roof-drain disconnection, rainwater harvesting, and stream daylighting.

In addition to drainage improvements, LID can reduce energy usage, provide natural habitat, and enhance the aesthetics and quality of life in urban neighborhoods. LID may also benefit San Francisco by reducing capital costs and operations and maintenance costs for managing wet weather flows and pollutant loadings and by augmenting the city’s water supply with harvested rainwater. To better understand the potential benefits of LID in San Francisco, the project team analyzed several LID strategies for urban areas, then conducted geographic information system (GIS) spatial analysis and LID modeling with an eye toward policy implications for the city.

LID Toolbox

The project team selected only those LID practices that would be widely applicable across San Francisco and for which modeling could be supported by adequate data and resources. Based on findings from a literature review, the LID practices selected were ecoroofs, bioretention, street trees, permeable pavement, and roofdrain disconnection. Each of these techniques is discussed briefly below.


Ecoroofs, also known as green roofs, living roofs, or vegetated roof covers, are categorized as intensive or extensive, depending on the substrate depth and vegetation type. This analysis focused on extensive ecoroofs, which offer greater benefits for stormwater management. An extensive ecoroof covers a rooftop with shallow layers (<150 mm, or <6 in.) of lightweight growing medium, low-growing vegetation, subsurface drainage, and a waterproof membrane. Ecoroofs can be installed on most types of commercial, multifamily, and industrial structures, as well as single-family homes, garages, and sheds. Roofs with slopes up to 40 degrees are appropriate, though ecoroofs are most suitable on roof slopes between 5 and 20 degrees. Ecoroofs can be used for new construction or reroofing an existing building. Candidate roofs must have sufficient structural support to hold the additional weight of the ecoroof, generally 49 to 122 kg/m2 (10 to 25 lb/ft2), saturated.

Ecoroofs absorb rainfall and release it slowly, thereby reducing the runoff volume, reducing peak flow, and delaying the time to peak. On average, ecoroofs retain 50% to 65% of rainfall annually and reduce peak flows for large rain events (>40 mm, or >1.5 in.) by approximately 50%. Ecoroofs also filter out contaminants as runoff flows across the roof and degrades those contaminants by binding them to the growing medium or via direct plant uptake. Studies have shown reduced concentrations of suspended

solids, copper, zinc, and polycyclic aromatic hydrocarbons in ecoroof runoff, with increased concentrations of phosphorus and nitrogen.


The term bioretention refers to dispersed, small-scale landscape features designed to attenuate and treat stormwater runoff. Examples of bioretention include rain gardens, curb bulb-outs, vegetated swales, and tree-box filters. Though bioretention designs vary widely, they all have the same basic features: a soil mixture, a drainage mechanism, and vegetation. Bioretention systems can be installed in parking lots, roadway median strips and rights-of-way, parks, residential yards, and other landscaped areas. In areas where infiltration is appropriate, bioretention can be designed to store runoff temporarily while it slowly infiltrates into underlying soils. Bioretention facilities also can be designed to discharge runoff to a conveyance, dispersal, or storage facility. The main considerations in siting bioretention facilities are space availability, slope, depth to bedrock and water table, and suitability of the soils for infiltration.

Bioretention systems improve stormwater runoff quality, reduce runoff volumes to varying degrees (depending on sizing and infiltration or flow-through design), and delay and reduce runoff peaks. Bioretention facilities designed as small-scale infiltration systems are most effective for reducing runoff volumes. Those designed as flow-though systems (discharging to the municipal sewer) show lower retention rates but are effective at delaying and reducing peak runoff. Bioretention systems also have demonstrated removal of nutrients, metals, and oil and grease.

Street Trees and Urban Forests

Urban forests, made up of publicly and privately maintained street and park trees, offer myriad benefits to the urban environment, including stormwater mitigation. Trees intercept rainfall before it reaches the ground, retaining a portion in the tree crown and thereby reducing runoff volume and peak flow.

In 2003, the City of San Francisco Street Tree Resource Analysis reported that approximately 56% of all street-tree planting sites in the city were unplanted. The analysis found that San Francisco’s street trees reduce stormwater runoff by an estimated 375,000 m3 (99 million gal) annually, for a total value of $467,000 per year. On average, 3800 L (1006 gal) are intercepted annually per tree in San Francisco. The report identified the number of unplanted street tree sites, by tree size, for each supervisor district. These data were used for modeling street trees in this LID analysis.

Permeable Paving

Permeable paving surfaces are made of a porous, load-bearing surface and an underlying aggregate layer that allows for temporary storage prior to infiltration or drainage to a controlled outlet. Several types of paving surface are available to match site conditions, intended uses, and aesthetic preferences. Permeable pavement systems are most appropriate in areas with low-speed travel and light- to medium-duty loads, such as parking lots, low-traffic streets, streetside parking areas, driveways, bike paths, patios, and sidewalks. Site conditions (including soil type, depth to bedrock and water table, slope, and adjacent land uses) should be assessed to determine whether infiltration is appropriate and to ensure that excessive sediments and pollutants are not directed onto permeable surfaces. Permeable paving surfaces must be cleaned periodically, usually by street sweeping or pressure washing, to prevent clogging.

Permeable pavements are effective at reducing runoff, delaying the onset of runoff, and improving water quality. Infiltration rates of permeable surfaces decline over time to varying degrees depending on design and installation, sediment loads, and maintenance.

Roof Rainwater Harvesting

Downspout disconnection, also called roofdrain redirection, involves diverting rooftop

drainage away from the sewer and into infiltration, detention, or storage facilities. In areas where site conditions allow infiltration, roof drainage can be conveyed to bioretention cells or dry wells, or simply dispersed onto a lawn or landscaped area.

The term rainwater harvesting refers to collecting rainwater from rooftops or other surfaces and storing it for later use. After pretreatment (such as leaf screens and first-flush diverters), collected rainwater is stored in rain barrels or cisterns. Captured rainwater can be used for onsite landscape irrigation, nonpotable household uses, and industrial or commercial uses. Depending on the desired use, a rainwater treatment system may be required.

Roof rainwater harvesting can retain up to 100% of roof runoff onsite, with only that water in excess of storage capacity contributing to stormwater flows. Rainwater can be viewed as a resource, rather than a waste product.

Methodology for Analysis

Two main questions guided the LID analysis:

  1. To what extent could the selected LID techniques be implemented in San Francisco?
  2. What drainage improvements could be achieved by installing LID facilities throughout the city?

This section provides an overview of the methodology used to answer those questions.

Infiltration Analysis

The potential for infiltration practices, such as unlined bioretention and permeable pavement systems, was evaluated by creating an infiltration zone map of San Francisco. The map, developed using GIS software, represented areas that generally would be appropriate or inappropriate for infiltration. Infiltration zones were delineated using the following five variables:

  • slope (0%–5%, 5%–10%, > 10%);
  • depth to bedrock (0–3 m [0–10 ft], > 3 m [> 10 ft]);
  • soil contamination;
  • soil group classification (U.S. Soil Conservation Service groups A, B, C, and D); and
  • liquefaction risk (very low, low, medium,high, and very high).

The first infiltration zone map showed that only 15% of the city would be amenable to infiltration. Subsequent iterations of the infiltration zone map further refined the infiltration zone categories, with a much smaller area found to be unsuitable for infiltration. Still, only a small portion of the city would be considered ideal for infiltration. Based on those findings, the project team decided that infiltration practices should not be included in the LID modeling. These practices, however, could be considered on a site-specific basis.

GIS Spatial Analysis

The purpose of the spatial analysis was to determine the extent to which LID practices could be implemented throughout San Francisco, given a set of physical constraints. The project team utilized aerial photos and GIS software to evaluate potential implementation areas. Due to the limited infiltration areas, the project team chose to evaluate lined bioretention and permeable pavement systems that discharge to the sewer, rather than infiltrate.

Parking lots and rooftop coverage.
GIS data were not available for parking lots or buildings. Consequently, the project team devised an alternative method consisting of three main steps: delineate general zoning classifications for San Francisco, analyze satellite imagery to estimate the percentage of land covered by building roofs and parking lots for each zoning class, and apply those percentages within GIS to estimate the parking lot and roof areas in each subcatchment.

Zoning districts, as defined by the San Francisco Planning Department, were simplified into 10 generalized classifications, it was assumed that they reflected complete yet distinct land uses of the city, and that roof coverage and parking lot coverage would be similar throughout each district.

The percent of rooftop coverage was estimated by analyzing aerial satellite photos of two typical blocks within each zoning class. Blocks that were selected:

  • fit the description of the zoning district code,
  • did not contain a mix of zoning districts or include special land uses,
  • were representative of typical building types (i.e., roof cover and roof slope) for the zoning district, and
  • did not otherwise differ from the norm.

One block with slightly higher density and one with slightly lower density were selected, and the average was taken. Two categories of roof slope then were identified: over and under 20-degree slope. Roof slope was estimated by viewing oblique-angled “bird’s-eye view” aerial imagery.

Large parking lots (more than 0.2 ha [0.5 ac]) were identified by viewing satellite images for each of the eight generalized zoning districts.  Unlike the roof analysis, the parking lot analysis did not rely on representative blocks; rather, parking lot areas were identified for the entire zoning district. Since only parking lots larger than 0.2 ha (0.5 ac) were identified, the parking lot coverage estimates are likely conservative.

GIS mapping. GIS software was used to compile or generate relevant GIS layers and to analyze land uses within each subcatchment. The entire San Francisco watershed consists of 931 subcatchments. Existing GIS data were available from the City of San Francisco for sidewalks, zoning classifications, supervisor districts, subcatchments, depth to bedrock, and slope. A pavement layer, representing paved streets, was generated from the sidewalk and zoning layers.

The newly derived parking lot and rooftop data were combined with existing GIS layers (zoning, sidewalks, and pavement), and these layers were overlaid with the existing subcatchment delineations to break down the land-cover information for each subcatchment. Application of LID siting constraints. For each LID practice, a set of constraints was applied within GIS to limit implementation to the most appropriate areas. These siting constraints were used to determine the eligible area for each LID practice in each subcatchment.

Modeled LID Scenarios

Several scenarios for LID implementation were developed. These implementation scenarios encompassed the baseline (no improvements), individual LID practices, and combinations of LID practices. A 30-year program target was developed to represent a reasonable level of LID implementation that could be achieved.

Simulated Rainfall Events

LID scenarios were modeled with a range of discrete-event design storms and with continuous simulation of a typical year’s rainfall. The design storms selected were those with 3-month, 1-year, and 5-year recurrence intervals, each with a duration of 24 hours, to demonstrate LID effectiveness under a range of rainfall intensities.  The 5-year event is the drainage system design standard used by the San Francisco Department of Public Works.

Model Representations of LID Practices

An InfoworksTM collection system model, developed by Metcalf & Eddy (Wakefield, Mass.), was used as the foundation for LID scenarios. For each LID practice, the drainage parameters and network design were modified from the baseline model to best represent the expected drainage performance. Adjusted subcatchment parameters included depression storage (DS; the volume of water retained on the land surface) and surface roughness (Manning’s n). Two performance levels were modeled for each LID practice, ranging from low to high expectations.

Ecoroofs were represented in the model as a new surface type with DS and Manning’s n values higher than those for impervious surfaces. The area to be converted to ecoroofs was removed from the baseline impervious surface area and placed in the new ecoroofs surface type. DS and Manning’s n were adjusted to match ecoroofs’ expected drainage performance. A DS of 0.4 to 0.8 for ecoroofs corresponded with published research data on their storage capacity. Manning’s n values of 0.15 and 0.6 were selected to reflect the range of vegetation and drainage layer characteristics.

Street trees were represented in the model as a new surface type, with a DS value based on the maximum interception volume for a single storm event (crown storage capacity). Three species of broadleaf evergreens were selected to represent typical small, medium, and large street trees in San Francisco. The trees were modeled at two performance levels: The low-performance level assumed smaller crown sizes and interception rates for each tree than the high-performance level.

Roof disconnection was represented in the model as area removed completely from the subcatchment impervious area. Thus, this area would not produce any runoff. To represent low performance for roof disconnection, only 50% of the roof disconnection area was removed from the subcatchment area. This was done to represent undersized rainwater harvesting systems capturing only half the rainfall volume on average.  The high-performance level was represented by removing 100% of the roof disconnection area, with the assumption that all runoff would be kept from entering the sewer.

Lined bioretention and permeable pavement systems were represented as small-scale detention facilities by creating storage nodes within the sewer network. Bioretention storage volume was calculated based on a unit basin storage of the first 13 mm (0.5 in.) of runoff. Permeable pavement storage volume was calculated based on a storage depth of 200 mm (8 in.). The storage nodes were designed to overflow directly to the sewer network and to drain completely within 72 hours.

Through this phase of modeling, the project team demonstrated the citywide volume and peak reductions that could be achieved based on the data limitations cited above. The following sections present the spatial analysis and modeling results for ecoroofs, roof disconnection, street trees, lined bioretention, and lined permeable pavement.

LID Implementation Potential in San Francisco

The total land area of San Francisco, as calculated in GIS, is 11,626 ha (28,729 ac; excluding public waterbodies and subcatchment areas lying outside city limits). Impervious land surfaces making up the majority of that area are as follows:

  • roofs, 3368 ha (8323 ac; 29% of land area);
  • roofs with less than 20-degree slope, 2585 ha (6387 ac; 22% of land area);
  • sidewalks, 996 ha (2460 ac; 9% of land area);
  • streets, 1922 ha (4749 ac; 17% of land area); and
  • parking lots greater than 0.2 ha (0.5 ac) in size, 234 ha (577 ac; 2% of land area).

Modeling Results

The model results demonstrate that the chosen LID practices can reduce runoff volume and peak flow rate. However, the effect of LID on any particular subcatchment depends on the extent to which LID can be implemented there (for example, the number of available street tree sites), as well as hydrologic characteristics of the subcatchment. The following sections describe citywide results for single design storms and for the typical rainfall year.

Runoff Volume

Modeling results show that implementing ecoroofs, street trees, and roof disconnection will reduce runoff entering the combined sewer. Overall, implementing the 30-year LID target will reduce annual runoff volume by 1.5 million to 2.6 million m3 (400 million to 700 million gal, or 4% to 7% annually). Roof disconnection comprises the majority of runoff volume reduction, followed by street trees and then ecoroofs.

Model results show that ecoroofs reduced runoff volume by approximately 203 to 244 L/m2 (5 to 6 gal/ft2) converted annually, and retained on average 30% to 40% of the annual rainfall. Compared to monitoring data from ecoroofs in other U.S. cities, the annual retention rate of 30% to 40% is conservative.

Modeled street trees intercepted approximately 4164 to 5678 L (1100 to 1500 gal) annually per tree when the trees are 10 to 20 years old. This volume reduction is slightly higher than the annual average reported in the San Francisco Street Tree Resource Analysis. However, the Street Tree Resource Analysis results represented all tree types in San Francisco, while the LID model represented broadleaf evergreens only. Broadleaf evergreens intercept more rainfall than deciduous species where winter rainfall patterns prevail; therefore, higher interception rates are expected.

Disconnecting rooftops from the combined sewer reduced runoff volume by approximately 285 to 570 L/m2 (7 to 14 gal/ft2) disconnected. This large range reflects the assumption that between 50% and 100% of rain falling on the disconnected roofs would be captured and reused onsite.

Runoff Peak Flow Rate

Reductions in runoff peak flow rate are important because they indicate how LID might improve the level of service of an existing sewer, such as reducing street flooding, and reduce combined sewer discharges. The LID model demonstrated peak flow reductions for all practices modeled. Peak flow reductions were determined for design storms only, not the typical rainfall year. The 30-year target reduced citywide peak flows for the 5-year design storm by 10% to 14%, with street trees and roof disconnection contributing the highest reductions.

Policy Implications

The current regulatory framework for San Francisco’s stormwater and combined sewer systems is favorable to LID. PUC currently holds a National Pollutant Discharge Elimination System stormwater permit that regulates flows in separate sewer areas. The San Francisco stormwater permit division has included a provision within permits that LID best practices must be incorporated into stormwater management to protect San Francisco Bay water quality.

LID is a key element of PUC’s urban watershed program, which is developing basin plans for nine major drainage areas. The plans will identify basin-scale opportunities for integrating LID into built-out neighborhoods and harnessing existing green spaces for stormwater management. The plans also will address implementation challenges for LID, such as the predominance of internal roof drains, space limitations for cisterns, public unwillingness to participate, and costs. PUC policies will encourage new and redevelopment projects to incorporate stormwater control into their designs, and will promote LID through a combination of policies, technical assistance, education, incentives, and demonstration projects. Findings from the LID analysis are being used by PUC to shape its LID programs and demonstration projects — particularly to target its investments on the most beneficial LID practices and locations.

Given the potential benefits of LID for San Francisco, PUC is pursuing further studies of LID. The second phase of analysis, which is ongoing, refines the previous analysis with more precise land-cover data, collection system data, and modeling techniques. The focus is on particular neighborhoods that experience periodic street flooding and on subcatchments that are tributaries to CSOs. In addition, infiltration facilities are being modeled in areas with amenable conditions, and stream daylighting and large-scale cisterns are being investigated.

About the authors:
Lori Kennedy, Lydia Holmes, and Steve McDonald are project engineer, associate, and partner, respectively, at Carollo Engineers in Walnut Creek, Calif. Rosey Jencks is a stormwater planner at the San Francisco Public Utilities Commission. Greg Braswell is Sewer Information Systems director at the San Francisco Department of Public Works. The authors wish to acknowledge Reese Madrid (San Francisco Department of Public Works), Sharon Tsay (Metcalf & Eddy [Wakefield, Mass.]), Beth Goldstein (Hydroconsult Engineers [San Francisco]), and Joong Lee (Carollo Engineers) for their valuable contributions to this study.