**2. Methods**

#### *2.1. Study Area*

The City of Eugene (Figure 1; population 156,000; median income \$44,859; area 113 km2) sits within the southern Willamette Valley in western Oregon [45], an area with a Mediterranean climate (Köppen *Csb*) of long cool rainy winters and warm dry summers. The valley surroundings promote winter temperature inversions and summer wildfire smoke collection, causing Eugene to rank among the twenty worst cities in the USA for short-term small particulate (PM2.5) air pollution [46]. The City of Eugene is also currently required by its National Pollution Discharge Elimination System (NPDES) permit to reduce the waterborne discharge of pollutants from the municipal system to the maximum extent possible [47]. The City of Eugene Park system possesses nearly 2000 ha of natural areas and open space, but their aggregation on the outskirts of town [48] limits their contributions to ES in urban neighborhoods.

**Figure 1.** Location of the Friendly Area Neighborhood in Eugene, Oregon.

Within the city, the Friendly Area Neighborhood (FAN; Figure 1; population 7000; area 3.7 km2) is zoned primarily (~75%) for low-density residential development (8–10 dwelling units/ha) and consists largely of single-detached units, with a median tax lot parcel area slightly below the USA median (0.073 ha vs. 0.083 ha) [45]. Nearly all streets in the neighborhood contain vegetated planting strips within city right-of-way easements, while sidewalks are intermittent. The FAN median annual household income (\$46,300) is \$7000 below state and \$11,300 below national medians [49,50], but its access to public green space is above average, with >10% of its land devoted to public parks and schoolyards and ~95% of residents living within a five-min walking distance along roads (i.e., <400 m) of a public park or schoolyard (Figure S1); the neighborhood is therefore comparable to top cities in the United States for such access [51].

#### *2.2. Public Green Space Mapping and Urban ES Quantification*

Although privately owned land is important in providing urban ES [52,53], this study focuses on public green space in which urban ES delivery is managed by the City. To characterize this space, we used multiple mapping strategies to inventory five distinct vegetated land cover types in the neighborhood (Table S1). Each street was traversed on foot in 2017 to identify lawn within the public right-of-way, and lawn without tree canopy cover was geospatially located and measured using a Garmin GPSMAP 62S handheld Global Positioning System (Garmin Ltd., Olathe, KS, USA). In tax lot parcels without adjacent sidewalks, right-of-way boundaries were assumed to extend 3 m on either side of the roadway. The boundaries of woodlands, classified as clustered trees clearly distinguishable from the U.S. Department of Agriculture's 2016 National Agriculture Imagery Program (NAIP) imagery, were assessed visually in ESRI ArcMap 10.7 (ESRI, Redlands, CA, USA) [54] and confirmed in the field. All other vegetation classifications (i.e., trees, tall shrubs, and short shrubs, as well as lawns located in parks) were made using normalized di fference vegetation indices (NDVIs) and height; the NDVI was calculated on a continuum from −1 to +1 using the NAIP four-band imagery with 1 m resolution. The NDVI range for each vegetation class was determined by comparing NDVI and color composite images [55]. The minimum NDVI threshold for all vegetation classes was set at 0.25, with the exception of lawn, which was identified using a minimum NDVI threshold of 0.0 (Table S1). Vegetation height was derived from 2015 light detection and ranging (LiDAR) point-cloud data [56] that were used to generate digital elevation and digital surface models. Digital elevation model values were subtracted from the digital surface model to create a digital height model at 1 m resolution, and vegetation was classified by combining NDVI thresholds with height ranges determined by Derkzen et al. (Table S1) [14].

The accuracy of each NDVI/LiDAR-derived land cover classification was evaluated through a process in which four hundred points, or 100 for each of the four vegetation types classified using NDVIs and LiDAR, were randomly selected and validated visually with NAIP imagery. Air photo interpretation was used to determine land cover type for all points clearly and obviously identifiable from the air photo. Land cover types for all remaining unidentified points were confirmed in the field (Table S2). NDVI/LiDAR-derived public green space land cover quantities were adjusted using validation proportions from Table S2 (see Table S3 footnotes), and five urban ES were quantified from these adjusted spatial data using indicators and supply rates compiled by Derkzen et al. for each of the five green cover types—vegetative ground cover (i.e., lawn), short shrub, tall shrub, tree, and woodland (Table S1) [14].

#### *2.3. Urban ES Supply Rates*

For the existing land cover, supply rates of five urban ES (air purification, carbon storage, runo ff retention, cooling fraction, and outdoor recreation) provided by the five green cover types described above were estimated according to Derkzen et al. [14], in which urban ES supply rates from numerous studies were integrated for the analogous Mediterranean (*Csb*) climate of Rotterdam, NL (Table S1). Although numerous modeling techniques exist for urban ES assessment [16], we chose this straightforward approach, consistent with recent recommendations and used by other case studies [57], as one that would be accessible to a wide range of urban planning practices.

In exploring potential future alternative planting regimes (Section 4.2.), we included stormwater filtration facilities (e.g., stormwater planters and rain gardens) that are not currently present in the neighborhood, estimating their stormwater reduction potential using the Simplified Approach described in Eugene's Stormwater Management Manual [58]. Impervious surface area, a necessary input, was calculated for the neighborhood using image segmentation and supervised learning in ESRI ArcGIS Pro 2.6 (ESRI, Redlands, CA, USA) [59] based on infrared, red, and blue bands from 2016 NAIP four-band imagery. To assess the accuracy of impervious and pervious surface classification, 100 random points were selected for each land cover type, and every point was validated visually with the NAIP imagery. The overall accuracy of the supervised segmentation classification was 94.5% (Table S9).

Urban green space also has the potential to provide ecosystem disservices, including pollen production that exacerbates allergies; a volatile organic compound release that contributes to ground-level ozone formation in the presence of automobile exhaust; and growth of tree limbs that may interfere with electricity lines or fall during storms, causing property damage [60,61]. These may also be estimated quantitatively in some cases (e.g., [62]), but we have not included these considerations here.

#### *2.4. Resident Surveys*

Non-stratified random sample surveys were administered to residents of the FAN to determine their urban ES priorities for public green space and the potential for increased funding for green infrastructure development. A random sample of 500 residential tax lot parcels was selected using county tax lot parcel data for the FAN as a sampling frame. Each selected lot was visited once on a weekday between 5 and 7 PM, and 19.4% of these visits yielded a completed survey (*n* = 97). The majority of the recorded non-responses resulted from resident absences, suggesting that repeated visits could have increased the response rate, and homes with posted "Do Not Disturb" or "No Soliciting" signs were also recorded as non-responses. Among residents who answered their doors, over half agreed to participate. Surveys were conducted orally in a format approved by the University of Oregon's Institutional Review Board. To minimize the survey's perceived invasiveness, sociodemographic information was not collected, although it could have been informative.

Residents were asked to rate 17 randomly ordered urban ES according to their importance for public green space in their neighborhood using a five-point Likert scale from 1 ("very unimportant") to 5 ("very important") (detailed in Supplementary Materials Section S2). They were then asked whether they supported the managemen<sup>t</sup> of public green space to increase urban ES delivery and whether they would be willing to support such e fforts financially, through personal donations or taxes, and through direct contribution of volunteer time.

Resident priorities for public green space urban ES were evaluated using Pearson's chi-square tests for both pairwise and aggregate comparisons, and chi-square tests were further used to compare priorities among urban ES classification types (i.e., provisioning, regulating, cultural, and supporting). Results for each urban ES classification type were tested for internal consistency using Cronbach's alpha (α), and values above 0.7 were regarded as acceptable [63]. "Priority" urban ES were defined as those with Likert responses of 4 ("moderately important") or 5 ("very important"), and Likert responses were reclassified as either priority (values 4 and 5) or non-priority (values 1–3) for data analysis. Descriptive statistics were used to compare residents' willingness to support green infrastructure development. All statistical analyses were conducted in R [64].

#### *2.5. Delphi Method*

We used an iterative survey process, known as a Delphi analysis, to consult with a group of individuals with specific knowledge of the planning and managemen<sup>t</sup> of public green space in Eugene [65,66]. Of the 34 people invited to participate on the basis of their expertise in public policy and green space management, 15 agreed, including nine members of the Eugene Public Works Department (including Parks and Open Space, Stormwater Management, and Urban Forestry), two City Planning and Development members, two local environmental non-profit representatives, one City Council member, and one University of Oregon Landscape Architecture faculty member.

In the first survey, participants ranked the 17 urban ES used in the resident survey in order of importance for public green space managemen<sup>t</sup> in Eugene; those urban ES with mean and median rankings below the top 10 were eliminated from the second survey. In addition, seven open-ended questions asked participants to describe and explain their perspectives on urban ES opportunities and barriers further. In the second round, participants were asked to review the collective results and

representative responses from the first round before again selecting the urban ES they considered to be priorities and expressing their levels of agreemen<sup>t</sup> with responses to the open-ended questions of the first round. To reflect di fferences in management, safety, and ecological benefit potential, these questions distinguished between parks and right-of-way planting strips (Supplementary Materials Section S3).

No particular proportion of agreemen<sup>t</sup> defines "consensus" in the Delphi method, and documented thresholds have varied from a simple majority to 95% [67–70]. Here, we chose a consensus threshold of two-thirds (67%), consistent with practices in many city governments [71,72].
