Walkability at Street Level: An Indicator-Based Assessment Model

Abstract

Walking is recognised as a healthy and sustainable mode of transport. Providing adequate infrastructure is pivotal for the promotion of walking and, subsequently, for achieving the benefits derived from its numerous positive effects. However, efficiently measuring the walkability at the street level remains challenging. In this paper, we present an indicator-based assessment model that can be used with open spatial data to evaluate segment-based walkability. The model incorporates eleven indicators describing the street segments and their close surroundings that are relevant for pedestrians, such as the presence and type of pedestrian infrastructure, road category, noise levels, and exposure to green and blue space. A weighted average calculation results in walkability index values for each street segment within a road network graph. The model’s generic approach and the ability to be used with open data ensure its reproducibility, adaptability, and scalability. The feasibility of the walkability model was shown using a case study for Salzburg, Austria. The model’s validity was evaluated through a large-scale study involving 660 full responses to an online survey. Participants provided ratings on the walkability of randomly selected street segments in Salzburg, which were compared with the calculated index, revealing a strong correlation (Spearman’s rank correlation = 0.82).

1. Introduction

Increasing the mode share of active transport is an urgent matter. As a result of the rapid growth of private motorization since the 1950s, we now face high levels of air pollutants, greenhouse gas emissions, noise, and traffic congestion [1,2], all of which negatively affect our environment. Moreover, and linked, we have seen a rise in sedentary lifestyles in Western countries. According to the WHO [3], between 2020 and 2030 almost 500 million people will develop heart disease, obesity, diabetes, or other noncommunicable diseases (NCDs) directly attributable to physical inactivity. If no actions are taken to encourage more physical activity among the population, this study calculated that this epidemic will cost the world economy USD 27 billion annually [4]. In response to this multi-faceted situation, the promotion of active mobility has gained momentum and is now included in the Sustainable Development Goals (SDGs), both directly (in SDG 11) and indirectly (in SDG 1 on poverty, as well as in SDG 5, 10, and 15) [5]. However, changing mobility behaviour is highly complex, as it depends on various interrelated factors [6]. Encouraging people to walk or cycle relies greatly on adequate infrastructure and a physical environment that is safe and pleasant [7]. Since this study focuses on pedestrians, we primarily examine the influence of the physical environment on walkability, which has been extensively researched [7,8,9,10,11,12,13,14]. To identify road spaces that are not yet sufficiently walkable, we need reliable methods to measure walkability at street level in order to support decision-making. With the help of such methods, we can contribute to the necessary redesigning of road spaces in favour for active modes.

Alongside audits and qualitative surveys that assess walkability, geographic information system (GIS)-based walkability indices (WI) are widely employed. The “5 Ds”—diversity, density, design, destination accessibility, and distance to transit—are widely recognised criteria for evaluating urban walkability [7]. Traditional indices often aggregated walkability indicators, such as land use mix, street connectivity, net residential density, and proximity to destinations, in specific areal units, such as raster cells, tessellation grids, or administratively defined regions [15,16,17]. One of the most widely known examples is the US-based Walk Score^®, which quantifies functional walking potential based on proximity and accessibility to destinations [18]. However, due to the aggregation level, the results from these WI do not provide a detailed insight into the walkability of the road space itself. In reaction to that and to find ways on how to provide a more complete portrait of the complex determinant of walking behaviour [19], several studies focused on developing methods to assess walkability at the street scale, using street-level variables for their WI calculation, also known as microscale variables [20,21,22,23,24,25]. Such variables are, for instance, traffic speed, lighting, crossing possibilities, street width, and motorised traffic volume [20]. Green spaces, noise levels, overall aesthetics, and well-maintained pedestrian infrastructure, which also count as microscale variables, not only contribute to mental and physical health but also help to reduce stress and improve happiness [11,26,27].

A major challenge in developing such indices lies in the collection and quality assurance of microscale variables. In situ inspections of street indicators, virtual audits, or using Google Street View are common ways to accurately identify the indicators [22,25,28]. However, these approaches are labour-intensive and costly, which limits scalability and transferability. To address this, some studies propose automated audits based on computer vision techniques as a scalable alternative to observational approaches [21,29]. These approaches, however, are not always easy or quick to implement.

Thus, a recurring challenge in walkability assessments is how to obtain meaningful micro- or macroscale indicators while ensuring the easy reproducibility, transferability, and scalability of the assessment model.

A comparatively easy and cost-efficient approach that also offers a considerable degree of transferability is to use spatial data to obtain indicators. A recent study by Guzman et al. [23], for instance, demonstrates how spatial data can be used for walkability index calculation. The study also focuses on the integration of non-observable (subjective) variables, such as perceptions of safety, accounting for different pedestrian groups. This reflects another growing research trend in walkability studies, which seeks to incorporate subjective perceptions into walkability indices [30,31], recognising that pedestrians have diverse needs, abilities, and purposes [24].

Additionally, a recent study [32] explores temporal and seasonal variations in walkability. To accommodate these emerging perspectives on walkability, we present a walkability model that is adjustable and scalable to different use cases.

Based on these findings, we conclude that current developments in the field of walkability assessment research lack models that are easily reproducible, scalable, and capable of working exclusively with open spatial data, making them cost-efficient while remaining highly adaptable to emerging research interests, such as pedestrian types and different use-case scenarios for walkability assessments. In addition, several recently published walkability studies have included some kind of validation [23,25,28,30], but there is a lack of large evaluation studies for the proposed WI. We therefore address the following research questions in this study:

• How should we design a walkability assessment model that is cost-efficient, adaptable, and scalable and can potentially account for various user groups, purposes, and locations?
• How does the initial selection and weighting of indicators perform in a real-world example and how are those results evaluated through a large-scale study?

We first proposed a method for a GIS-based walkability assessment at the segment level. The model was then applied to a real-world environment. The results were validated in a large evaluation study. Outcomes and learnings are reflected in the Discussion Section before we provide a conclusion.

2. Materials and Methods

This section is divided into two parts. The first part introduces the walkability model and the selected indicators, and the second part explains the evaluation study.

2.1. Walkability Assessment Model

Our proposed model is based on the concept of the bikeability assessment model for street networks, originally introduced by Loidl and Zagel [33] and later refined by Werner et al. [34]. The walkability model presented in this paper is integrated into the open-source application NetAScore [35], publicly available on GitHub (version 0.9.0).

While the basic approach of the model remains consistent with the bikeability model, this study places emphasis on the selected indicators specific to walkability, which have not yet been published. The walkability model provides a default parameter set for a generic walkability assessment, focusing on utilitarian pedestrian trips. Although we propose a general parametrization for selecting and weighting indicators, the method itself is scalable and adaptable. Additional indicators can be incorporated, and indicator weightings can be adjusted to suit different use cases.

The model workflow includes preparing a road network graph, computing and weighting indicator values, and calculating the final WI. The output is an enriched road network graph, where each road segment is assigned a WI value ranging from 0 (exceptionally good walkability) to 1 (poor walkability).

The steps involved and the description of the selected indicators are presented in the following sections. An overview on the generic workflow presented in this paper is presented in Figure 1.

2.1.1. Network Graph Preparation

The walkability model needs a road network graph as a basis input, with common attributes for street segments (edges), such as road category, information on access to different transport modes, the availability and type of pedestrian infrastructure, and the speed limit. These attributes are needed for the computation of indicators, which is explained Section 2.1.2. The road network graph used in our model is computed from OpenStreetMap (OSM) data. In cases where a road network graph lacks certain attributes necessary for the WI calculation, data may be procured from additional data sources. However, it is crucial to note that the presence and quality of attributes exhibit substantial variations across different sources. The quality of the output depends upon the quality of the input data, with high quality input data producing more reliable results.

2.1.2. Defining and Computing Indicator Values

The walkability model is based on indicators that quantify measurable factors of the physical environment of street segments, which are known to influence walkability. Since the relationship between the physical environment and walkability is well researched [16,17,18,36,37,38,39,40,41,42], we first identified relevant indicators based on the existing literature. Since our model is designed to work with open data sources, the final selection represents a trade-off between theoretical relevance of indicators and the availability of open spatial data of adequate quality. When open data for a crucial factor, such as traffic volume, is unavailable or impractical to obtain, we use proxies. Since road category is found to correlate with the annual daily traffic volume [43] and more traffic lanes can accommodate more traffic volume, we substitute traffic volume with the indicator number of lanes and road category, which can be derived as an attribute from most road network graphs.

For our generic parametrization, we propose eleven indicators, grouped into four aspects of walkability inspired by Alfonzo’s hierarchy of walking needs [8]: aesthetics, comfort, safety, and functionality. Among these, we consider safety the most critical aspect and allocate five indicators to it: pedestrian infrastructure, road category, maximum speed, number of lanes, and crossing availability. The comfort aspect includes noise levels, gradient, and building ratio, while aesthetics is represented by greenness ratio and presence of water bodies. The functionality aspect is captured by the facility ratio.

While some indicators can be derived directly from the road network graph, others require additional datasets and GIS-based methods to link the relevant information to the road segments. Most data can be derived from OSM, while other datasets are commonly available as Open Governmental Data (OGD). The processing of indicators involves classification and transferring attribute values to numerical values, indicating whether they positively (0) or negatively (1) influence walkability. For instance, pedestrian infrastructure separated from car lanes by physical barriers (e.g., trees or green strips) is more walkable than a standard sidewalk. Consequently, the classified indicator value “Pedestrians Separated” is mapped to 0 (indicating a positive effect on walkability), while “Sidewalk” is mapped to 0.5. The absence of infrastructure receives the numerical value of 1 (indicating a negative effect on walkability).

More details on each indicator follow in the next sections. An overview of the indicators, their classified values, and the assigned numerical and weighting values can be found in

Table 1. Indicators with brief descriptions, main aspect of walkability, weighting, and indicator values and their numerical mapping (default walkability profile).

Indicator and Short Description	Main Aspect of Walkability	Default Indicator Weight	Indicator Value with Default Numerical Mapping
Pedestrian infrastructure (availability and type of pedestrian infrastructure)	safety	0.4	No infrastructure: 1
			Sidewalk: 0.5
			Stairs: 0.3
			Pedestrians separated but mixed with cyclists: 0.15
			Pedestrian street: 0
			Pedestrians separated: 0
Road category (proxy for traffic volume and potentially the share of heavy goods vehicles)	safety	0.3	Primary: 1
			Secondary: 0.8
			Residential: 0.2
			Service: 0.15
			Calmed: 0.1
			Path: 0
			No motorised traffic: 0
Maximum speed (km/h) (speed limit of motorised traffic)	safety	0.3	≥100: 1
			≥80: 0.8
			≥70: 0.7
			≥60: 0.6
			≥50: 0.4
			≥40: 0.15
			≥30: 0.1
			<30: 0
Gradient (steepness of the road or path: physical effort needed)	comfort	0.3	≥12 %: 0.75
			<12%: 0.5
			<6%: 0.3
			<3%, >−3%: 0
			>−6 %: 0.3
			>−12 %: 0.5
			≤−12 %: 0.75
Number of lanes (proxy for traffic volume)	safety	0.1	>4: 1
			4: 0.9
			3: 0.8
			2: 0.5
			0–1: 0
Availability of facilities (points of interest in proximity)	functionality	0.3	No: 1
			Yes: 0
Greenness ratio (greenness of surrounding)	aesthetics	0.3	≥75%: 0
			≥50%: 0.1
			≥5: 0.2
			>0: 0.3
			0: 1
Presence of water bodies (proximity to water bodies)	aesthetics	0.4	No: 1
			Yes: 0
Building ratio (density of built environment)	comfort	0.1	≥80%: 1
			≥60%: 0.8
			≥40%: 0.6
			≥20%: 0.4
			>0%: 0.2
			0%: 0
Crossing availability	safety	0.2	No, and road category is primary or secondary: 1
			No, and road category is residential: 0.5
			Yes, and road category is smaller than residential: 0
Noise level (db) (ambient noise level)	comfort	0.3	≥70: 1
			≥55: 0.4
			≥10: 0.2
			≥0: 0
		∑3

.

• Pedestrian infrastructure

Appropriate, well-designed walking possibilities are mandatory for a safe and comfortable walking experience [42]. This indicator is a common attribute of the road network graph. Categorization and numerical mapping are based on the premise that greater allocated space for pedestrians, coupled with reduced interaction with other modes, contributes to enhanced safety, resulting in a more favourable rating [37,40].

• Road category

Higher road categories mean higher traffic volumes and therefore reduce the appeal of walking along these roads. This indicator is a common attribute available in road network graphs and is used as a proxy for traffic volume, a common attribute used in WI [20].

• Maximum speed

The speed of motor vehicle traffic is confirmed to significantly affect pedestrians’ sense of safety. As speed increases, pedestrian discomfort increases as they feel less safe [40]. Maximum speed is also a common attribute available in road network graphs.

• Gradient

The gradient of the segment reflects pedestrian comfort, with steep slopes generally being less attractive for walking compared to gentler inclines [36]. To calculate the steepness of a segment, a digital elevation model is needed. The gradient is calculated by subtracting the height of the first vertex of a segment from the height of the last vertex and then normalising by segment length. The steeper a segment is, the less comfortable for walking it is. Segments with no or little slope are preferred.

• Number of lanes

The more lanes a street has, the less attractive walking next to it is. This indicator, together with the road category, serves as proxy for traffic volume and is a common attribute in road network graphs.

• Availability of facilities

This indicator represents a frequently employed factor in mesoscale GIS-based WI, particularly the aspects of functionality and proximity to facilities [16,17,18]. In contrast to preceding indicators, its computation requires an analysis of the street segments surroundings. It is determined by calculating the number of distinct facilities within a 30 m radius around a segment. To normalise the resulting value, it is divided by the length of the segment. The facilities consist of categories such as entertainment, retail, and institutional establishments, including universities, schools, pharmacies, retailers, theatres, and museums. The data can be obtained from OSM. We suggest classifying this indicator into a Boolean value, indicating either a presence or absence of facilities along segments.

• Greenness ratio

Green space not only has an aesthetic appeal to pedestrians but also helps to lower temperature, provide shade, and enhance air quality. Therefore, green space is a frequently used indicator in frameworks assessing walkability [38]. This indicator requires additional data (e.g., from OSM) that include green spaces, such as parks, forests, recreational grounds, and meadows. Within a 30 m buffer around each segment, the cumulative area of green space is computed. The computed green area is normalised by the length of the segment. The higher the ratio of green space along the segment, the more positive the influence on walkability.

• Presence of water bodies

Similarly to the greenness ratio, water bodies, such as lakes and rivers, have a positive association for physical activities [39]. To quantify the existence of blue spaces along a segment, we choose different approaches for larger water bodies, such as lakes, and for flowing water, like river and streams (data obtainable from OSM). For segments intersecting with lakes or ponds within a 30 m buffer, the value is numerically mapped to most favourable. For rivers or streams, their mere intersection with the buffered segment is not meaningful, as we consider it necessary for a segment to follow a stream or river for a while in order to benefit from its aesthetic value. We therefore specify that only segments with an overlap ratio of 80% between a stream/river and the buffered segment can be considered to be following rivers or streams and are thus mapped with the best numeric value.

• Crossing availability

The availability of crossings is particularly important on streets with higher motorised traffic [42]. To accommodate this, the importance of crossings is coupled to road category. For residential street segments with low-speed limits that do not offer a crossing, the indicator value receives a medium rating for walkability (numeric value of 0.5). If primary and secondary road segments do not offer a designated crossing possibility, the indicator value for this segment is mapped with the highest numeric value (1), indicating a negative influence on walkability. Segments with crossing possibilities on road categories below residential (e.g., service roads, paths, etc.) receive the most favourable numeric value (0).

Crossing data can also be downloaded from open data providers, such as OSM. To spatially match the crossing to the segment, a small 10 m buffer around the segment is calculated and intersected with crossing possibility.

• Building ratio

We suggest that the indicator building ratio is an indicator which reflects block length, density, and proximity of buildings along a street segment. As Singh [41] states, vehicle-dominated urban areas usually have larger blocks and therefore less street connectivity. As walking distances are longer in such areas, it is likely that fewer people will walk there. To obtain this indicator, building polygons are required (derivable from OSM). The total area of buildings within a 20 m buffer around the segment is computed and normalised by the segment length. A higher percentage signifies denser and closer building placement to the street segment, consequently indicating a less pedestrian-friendly environment.

• Noise level

High noise levels are known to be discomforting. To match noise levels to road network graph segments, a noise polygon dataset is needed. Such datasets are frequently available via OGD portals. The noise polygons are intersected with the road network segments. The noise level is then assigned proportionally to the segments. The higher the noise level, the less favourable a segment is for walking.

2.1.3. Weighting of Indicators

In our walkability model, we propose a generic set of weights for the indicators, developed through an iterative refinement process in collaboration with local experts in active mobility, as suggested by Loidl and Zagel [33]. Over several weeks, this process involved multiple evaluation rounds, during which different weighting sets were tested by checking several street segments, cross-referencing with Google Street View, and looking for the overall plausibility of results by comparing the results across various locations. The final set of weights allocated 43% of the weights to indicators representing the aspect of safety. A total of 23% was allocated to aspect aesthetics, 23% was allocated to comfort, and 10% was allocated to the aspect of functionality. However, we emphasise that these weights can be adjusted for specific use cases, such as assessing walkability for pedestrian groups.

For details on implementation, see Supplementary Materials.

2.1.4. Walkability Index Calculation

A weighted average is computed across all indicators to determine the joint WI for each network segment. Consequently, the resulting metric adheres to the unit interval [0, 1], where 0 indicates excellent and 1 is poor walkability. We define segment-based walkability $w\left(e\right)$ for a graph consisting of vertices and edges $G(V,E)$ as

$$w\left(e\right)={\displaystyle \frac{\sum _{k=1}^n{w}_k{i}_k\left(e\right)}{\sum _{k=1}^n{w}_k}}$$

(1)

for $e \in E$ and $n$ indicators, where $w$ represents indicator weights and $i$ refers to numerical indicator value.

Since a weighted average may not always adequately capture walkability for certain indicator combinations, we incorporate the possibility for rule-based overrides to enhance flexibility in modelling individual perspectives on walkability. Specifically, we implement an override option for segments categorised as primary and secondary roads that provide only minimal pedestrian infrastructure, such as a basic sidewalk. In this case, the WI value is corrected to 0.8. This adjustment is incorporated because these segments may still receive high walkability scores due to positive values for indicators representing aesthetics, comfort, and functionality.

By integrating this rule-based override option, we ensure that the walkability assessment is able to consider specific safety needs of pedestrians and local particularities, respectively.

2.2. Evaluation

We evaluated the applicability of the model and the soundness of its outputs through an online survey. To set up this evaluation study, we first ran the walkability model on a road network graph for the city of Salzburg, Austria. Salzburg, with its population of 155,000, is a compact city where 40% of all trips are shorter than 2.5 km and 70% are shorter than 5 km [44]. This makes it an ideal setting for active mobility.

2.2.1. Data Sources and Index Calculation

We acquired OSM data for the city of Salzburg from planet.osm.org [45]. A digital elevation model (DEM) with a 10 m spatial resolution for Austria was obtained from data.gv.at [46] and clipped to the size of the city of Salzburg. Noise level data were also obtained from data.gv.at [47]. We loaded the data sources in a PostgreSQL database with PostGIS spatial extension.

In the first step, our SQL-script generated a network graph consisting of edges and nodes, with the edges carrying relevant attributes derived from the OSM data. In the second step, the walkability indicators were calculated. Depending on the specific indicator, values were either derived through combinations of network graph attributes or OSM tags or computed via spatial analyses using the DEM, the noise level dataset, or other OSM data.

In the final step, the walkability index was calculated as a weighted average. To achieve this, indicator values were mapped to numerical scores and multiplied by their corresponding weights. The sum of these weighted values was then divided by the total sum of weights, as shown in Equation (1). The result was a road network graph containing both the indicators and the walkability index for each network segment.

For further technical details regarding the implementation, the complete source code is available in the Supplementary Materials and is accessible via the provided repository.

2.2.2. Online Survey

The online survey aimed to provide study participants with the opportunity to rate randomly selected locations within the city of Salzburg based on their perceived walkability. These ratings were then compared to the calculated WI values.

To select the random locations, the calculated WI values for the road network graph in Salzburg were categorised into four groups: 0–0.25, 0.25–0.5, 0.5–0.75, and 0.75–1. Three segments were randomly selected from each group using the ‘Random points along line’ tool in QGIS 3.40.4. To present these locations in the online survey, photos were taken under consistent (sunny) weather conditions on the same day. To provide more context, textual descriptions were added to the images, including details about the noise level, street category, pedestrian infrastructure, and maximum speed (see Figure 2).

In addition to the twelve randomly selected locations, two locations were included, one known to be very walkable and one known to be less walkable (according to local knowledge). These extreme examples were first presented to the participants and served as benchmarks. The ratings for these benchmark locations were excluded from the final study evaluation.

Study participants were asked to rate the selection of images based on the images, the environmental descriptions, and a link to Google Street View, which provided a 360-degree view of the locations, using a slider from 0 (not pedestrian-friendly) to 100 (very pedestrian-friendly) with increment steps of 10. Beyond the image-based walkability ratings, we included additional questions. In the first section of the survey, participants were asked about their familiarity with the city of Salzburg, followed by brief demographic questions (year of birth, age, and gender). Additionally, participants were asked to rate the importance of each indicator. Each indicator was briefly described, and participants rated its importance using the options: ‘not important’, ‘less important’, ‘important’, or ‘very important’.

We implemented the survey using a LimeSurvey (https://www.limesurvey.org/) instance hosted by the University of Salzburg. The call for participation was distributed through our professional networks and university newsletters. No specific selection criteria were applied in recruiting participants.

3. Results

The results are presented in two subsections. The first section presents the results from the walkability assessment in the city of Salzburg. The second part presents the results from the online study. Discussion of the results follows in Section 4.

3.1. Model Results: Walkability

We computed segment-based walkability following the method outlined in Section 2.1 using the set of eleven indicators for the city of Salzburg, Austria.

The computed network for the city of Salzburg consisted of 58,091 segments with a total length of 3330 km and an average segment length of 57 m. A total of 97.4% of the segments were assigned with a WI value ranging from 0 (best) to 1 (worst). The remaining 2.6% were segments without access for pedestrians, for example highways. A total of 4.8% of segments were had an exceptionally good walkability (<0.25), 61.8% of segments had good walkability (0.25–0.5), 26.8% had an WI value between 0.5 and 0.75, and 4.0% had the worst walkability (>0.75). The histogram in Figure 3 displays the WI value frequency in the city of Salzburg. A map visualising the calculated walkability on street-level is shown in Figure 4.

The overall walkability of the city of Salzburg appears to be relatively high, particularly within the city centre. Unsurprisingly, the Salzach river that runs through the city from south to north exhibits high walkability, facilitated by well-established pedestrian and cycling infrastructure along this axis. The recreational areas in the city centre, particularly the two hills, Kapuzinerberg and Mönchsberg, exhibit good walkability. Walkability along busy main roads is tendentially weak, which is also explainable with the use of the overwrite option. Additionally, clusters of residential streets within neighbourhoods throughout the city exhibit weak walkability.

3.2. Online Survey Results

A total of 660 participants completed the online survey, which was distributed through our professional networks and university newsletters. The majority of participants indicated being very familiar (54%) or familiar (28%) with the city of Salzburg, a small percentage mentioned being not that familiar or not familiar at all with Salzburg. A total of 12% fell within the intermediate familiarity range. This insight was considered important, since we aimed for a substantial number of individuals familiar with Salzburg, who would have experience walking in the city and, ideally, be acquainted with some featured locations. Additional descriptive statistics about the participants can be obtained from

Table 2. Descriptive statistics about online survey participants.

Category	Subcategory	Count (%)
Gender	male	224 (33.9%)
	female	431 (65.5%)
	diverse	5 (0.8%)
Age (in years)	<20	11 (1.7%)
	20–34	357 (57.1%)
	35–49	137 (20.8%)
	50–65	113 (17.1%)
	>65	42 (6.4%)
Familiarity with the city of Salzburg	very familiar	355 (53.8%)
	familiar	187 (28.3%)
	somewhat familiar	77 (11.7%)
	not familiar	36 (5.5%)
	not familiar at all	5 (0.8%)

.

3.2.1. Comparison Between Participant’s Walkability Ratings and WI Value

The slider-generated (0–100) walkability ratings were converted to 0–1 range of the WI.

Table 3. Results from the evaluated locations according to their WI value and the participants’ ratings of walkability.

WI Value Category	Location Name	OSM ID	WI Value	Participants’ Ratings (Mean) (n = 660)	Participants’ Ratings (Standard Deviation)
0.0–0.25	Mirabellgarten	54833010	0.159	0.06	0.13
	Ignaz-Rieder-Kai	37890723	0.191	0.23	0.21
	Josef-Mayburger-Kai	106620210	0.216	0.16	0.18
0.25–0.5	Verbindungsweg	390117044	0.379	0.23	0.21
	Schwarzenbergpromenade	33038192	0.417	0.20	0.19
	Treppelweg	138326112	0.362	0.18	0.20
0.5–0.75	Metzgerstraße	67997439	0.552	0.63	0.20
	Galhamerweg	43366570	0.572	0.6	0.23
	Kendlerstraße	30510578	0.545	0.52	0.24
0.75–1	Gaisbergstraße	195910772	0.8	0.92	0.15
	Salzburgerstraße	156209933	0.8	0.29	0.19
	Innsbrucker Bundesstraße	314952977	0.8	0.54	0.23

and Figure 5 illustrate the comparison between WI values and survey means for each street segment. Spearman’s rho correlation coefficient of 0.8234 between the mean survey value of each location and the respective WI value indicates a strong correlation between the WI values and the ratings provided by participants (

Table 4. Results from comparing WI values to the mean participants’ ratings on walkability using Spearman’s rank correlation coefficient, p-value, and mean absolute deviation.

Statistics	(n = 12)
Spearman’s rank correlation coefficient	0.8234 *
p-value	0.002 *
Mean absolute deviation (MAD)	0.149 *

* Calculated between the walkability WI value and the mean value from participants’ ratings.

).

3.2.2. Participants’ Ratings of Indicators

In the online survey, participants were asked to provide a rating regarding the importance of each used indicator. A list of all indicators together with a brief description was presented to the study participants, with the possibility of rating each of them as very important, important, less important, or not important. The outcomes of these responses are presented in Figure 6.

4. Discussion

In this section, we interpret the results presented in Section 3 by addressing the following key topics: indicator selection and weighting and model output.

4.1. Indicator Selection and Weighting

Ten out of eleven indicators were rated by the study participants with the highest share as either important or very important, which we interpret as general agreement with our selection (see Figure 6). The only indicator considered less important by the majority was gradient. Since our model assigns a relatively high weight to this indicator (0.3), this finding will be taken into account in future model improvements.

In addition to this, we observed that our assigned indicator weights corresponded well to the ratings of participants for some indicators, especially for pedestrian infrastructure, road category, maximum speed, availability of facilities, and noise level.

The largest discrepancy, aside from gradient, concerns the crossing availability indicator. This indicator was rated as very important by 66% of participants, yet it received a relatively low weight in our model (0.1). Crossings occur on less than 10% of all segments in the road network of Salzburg. Increasing the weight for this indicator further would lower walkability scores in all segments without crossing opportunities, which must be carefully considered.

Since participants received only limited information about the indicators and how the model functions, some of the discrepancies between participant ratings and expert-driven weights may be due to a lack of context. More nuanced participant feedback on the weights could be obtained through workshops, as was done in the bikeability assessment model [34], or through qualitative interviews and audits.

Furthermore, improvements in the weighting process itself could be achieved by integrating structured decision-making methods, such as the analytic hierarchy process (AHP), to enhance the robustness and transparency of the assigned weights [48]. In addition, sensitivity analysis could be employed to assess the impact of weight variations on the final results [49].

As a final discussion point for this section, we want to emphasise that this model was developed to be adjustable and scalable. Therefore, indicators and weights can be modified, making the model flexible for various use cases. This approach aligns with current research, which indicates that walkability is subject to different perceptions, needs, and types of pedestrians [23,24,30,31]. Beyond accommodating diverse use cases, the model is also suitable for implementation in different geographic and cultural contexts. However, due to such contextual differences, adjustments to indicators and weights may be necessary. For instance, in a city like Salzburg, where steep slopes are present, the gradient indicator is relevant. In contrast, this indicator may be unnecessary in flat urban environments. Therefore, applying the model in different settings requires the careful consideration of its parametrization.

4.2. Model Outputs

The model outputs, computed for the city of Salzburg, demonstrate that segment-based walkability can be effectively estimated using open data. The strong Spearman’s correlation (0.82) between computed WI values and subjective human ratings indicates robust model performance. However, certain locations (e.g., Treppelweg and Schwarzenbergpromenade) exhibited notable discrepancies, which we attribute to data quality issues and the inherent subjectivity of walkability perception. The segment at Schwarzenbergpromenade was classified by our model as having only sidewalks, whereas, in reality, a separate pedestrian and cycling path runs alongside the road. In the case of Treppelweg, the discrepancy is likely due to subjective perceptions of walkability. This area is recognised as pedestrian-friendly by locals, with rich greenery and proximity to the Salzach River. However, the computed indicators for this segment show that there is a lack of dedicated pedestrian infrastructure, that motorised vehicles are allowed along this segment, and that there is an absence of facilities, which justifies the lower walkability value.

4.2.1. Overwrite Option

We tested the model both with and without the overwrite function and compared the results with participant ratings. Interestingly, the correlation was slightly higher when the overwrite function was not applied, increasing from Spearman’s correlation of 0.82 to 0.90. All three randomly selected locations for the 0.75–1.0 value group were affected by the rule-based overwrite. Without the overwrite option, the location on Gaisbergstraße was indexed as 0.688, Salzburgerstraße as 0.521 and Innsbrucker Bundesstraße as 0.666. While the segment at Gaisbergstraße received the lowest overall walkability rating from the study participants and therefore matched well with the overwrite value of 0.8, the segments at Salzburger Straße and Innsbrucker Bundesstraße were rated significantly better by the study participants than was suggested by our WI value. Participants rated Innsbrucker Bundesstraße with a medium walkability, despite its limited pedestrian infrastructure, consisting only of a narrow sidewalk next to a primary road. This suggests that minimal infrastructure may be considered acceptable under certain conditions. In contrast, the discrepancy in ratings for Salzburger Straße can be explained by an attribute error: the network graph used for location selection incorrectly classified the segment as a primary road with only a sidewalk, whereas the image presented to the study participants reveals a separated bike and pedestrian path alongside the road. As the overwrite option caused controversial results for this evaluation study, the option was implemented in NetAScore as an optional feature. However, we want to highlight that we see the usefulness of the overwrite option for specific application purposes. For example, the WI with the overwrite served as impedance factor for pedestrian routing in an agent-based simulation mode and significantly enhanced the model’s accuracy [50]. Hence, in cases where the WI is used for routing purposes, we recommend testing out the overwrite function.

4.2.2. Reflections on Data Quality

An important aspect of our study is the reliance on open data, primarily OSM. While community-generated OSM data are generally of high quality in the European context [51], occasional tagging inaccuracies (e.g., the misclassification of pedestrian infrastructure) can significantly impact the WI, as illustrated in the previous section. Since indicator calculations are based on attributes and tag information derived from OSM, it is important to acknowledge the potential for such inaccuracies. The model’s performance has not yet been tested outside of Europe and remains a subject for future research.

However, given the cost-efficient and scalable approach we present, particularly in comparison to walkability assessment methods that rely on in situ inspections, virtual audits, or Google Street view for indicator identification and quality control [22,25,28], our results are highly promising. The strategic combination of OSM attribute information and tags, along with the use of proxy indicators (e.g., road category and number of lanes as proxies for traffic volume), represents an effective, low-cost approach to assessing walkability at the street level.

5. Conclusions

This paper presents a generic, indicator-based walkability assessment model that uses open data to evaluate street-level walkability. We propose a generic set of eleven indicators while maintaining flexibility in the model’s design to allow for scalability and adaptability—indicators can be added or removed and weights can be adjusted for different use case studies. The model, with the generic parametrization, was successfully applied to the city of Salzburg. Given that walkability assessment models are rarely evaluated with large participant groups, we conducted an evaluation study with 660 participants who completed an online survey. In the evaluation study, the WI values were compared with participants’ ratings of walkability at randomly selected locations in our study area. The comparison between the participants’ subjective walkability ratings and our WI showed a strong correlation, demonstrating the overall effectiveness of the model. By using widely available open data sources such as OSM, the approach remains both cost-effective, transferable, and scalable, which addresses a key research need, particularly in accommodating different pedestrian types and diverse use cases in walkability assessments.

However, the following limitations should be noted. While over 600 responses to the evaluation study represents a solid sample size, we acknowledge that testing the walkability model in different cities would have strengthened the overall validity and generalizability of the findings. Additionally, the online study was not specifically designed to evaluate the indicator weights. Although this was not the primary objective, a different study design could have provided even more insights into this aspect. To achieve this, participants would have required additional context to fully understand how the weights influence walkability assessments. As mentioned above, workshops or qualitative interviews would be more suitable for evaluating indicator weights. Moreover, the online survey may not fully capture the multisensory experience of walking in situ. Still, we have taken a huge step forward by conducting a comprehensive evaluation, which is hardly ever integrated into comparable studies.

While our model is designed to be open, scalable, and adaptable, it necessarily abstracts reality and relies on proxy indicators and simplified rules (e.g., the overwrite option). Aspects such as intersection design, but also more subjective measures, such as the environment perceived by different pedestrian groups, are not yet integrated. Future research should therefore focus on incorporating subjective factors by defining different pedestrian types to gain further insights to walkability of different scenarios.

Moreover, since the model relies on open data, the results are highly dependent on its availability and quality. If such data are unavailable or of poor quality, commercial data sources may be required.

Despite these limitations, the proposed walkability assessment model can be regarded as a valuable tool for analysing street network graphs, supporting decision-making, and enhancing sustainable transport. Beyond urban planners, the proposed walkability assessment model holds value for a wide range of stakeholders. Public health professionals can use it to support interventions promoting active lifestyles [52] while policymakers and local governments could rely on it to guide sustainable transport and infrastructure planning. Researchers can benefit from its flexibility for comparative studies, and real estate developers can use it to assess the attractiveness of locations. This broad applicability underlines the model’s potential as a versatile and impactful tool across multiple domains. Moreover, the model is publicly available, ensuring the overall transparency of indicator parametrization and model functionality.