1 Introduction

Crime in urban areas has consistently been a subject of significant concern across various sectors of society. Streets, as a crucial component of urban public spaces, guide people’s behavior patterns and serve as the arteries of a city, connecting its various parts (Su et al., 2022). Their planning and construction directly impact the quality of life for urban residents and the city’s image. At the same time, streets are also a common location for urban crime (Prozumentov & Rachkova, 2013; Sanchez, 2019; Scheingold, 2011; Truntsevsky et al., 2018), making it essential to study the urban street environment to prevent and control urban crime. The distribution of urban crime is a result of the combined influence of multiple factors, including the built environment, population distribution, and socio-economic conditions (Xiao & Zhou, 2020). Some environmental criminology theories also suggest that the built environment of streets plays a significant role in the occurrence of crime, and effective urban street design can prevent crime.

Existing research on the impact of street environment on crime has yielded abundant results, but past studies primarily focused on adult crime or failed to distinguish between adult and juvenile crime (Han et al., 2023; Kuen et al., 2022; Lorenc et al., 2012; Luo et al., 2022). “Juvenile delinquency” has been recognized globally as the third most serious social problem, following drug abuse and environmental pollution, and remains a significant social issue worldwide. It not only affects adolescents themselves and their families but also impacts entire nations and ethnic groups. Due to the lack of consensus on the age definition of juvenile delinquency internationally, this paper defines criminal acts committed by individuals under the age of eighteen as juvenile delinquency, based on the New York Family Court Act, to differentiate juvenile delinquency from other adult crimes.

Juvenile delinquency has always been a research hotspot. Early studies on juvenile delinquency primarily adopted perspectives from criminology (Hirschi, 1969), sociology (Lemert, 1972), psychology (Agnew, 1992), and law. With the development of the Chicago School, research on the relationship between juvenile delinquency and time and space gradually emerged (Bursik & Webb, 1982; Schuerman & Kobrin, 1986). Scholars attempted to correlate urban area characteristics with the occurrence of juvenile delinquency. Among them, Shaw and McKay introduced the ecological approach into juvenile delinquency research, discovering the existence of “juvenile delinquency areas” through their study of the spatial distribution of crime rates in the Chicago metropolitan area (Shaw & McKay, 1942). They also proposed the “Chicago Area Project,” the first major crime prevention plan in the world. The research and conclusions of the Chicago School have had a profound impact on the study of juvenile delinquency, shifting the focus from the perpetrator to the criminal act. Since then, practical research based on the “juvenile delinquency area” theory has emerged continuously, typically analyzing the characteristics, influencing factors, and formation mechanisms of juvenile delinquency areas, many of which have been applied to juvenile crime prevention efforts (Law & Abdullah, 2024; Long & Liu, 2022; Revital & Haviv, 2023). Compared to adults, juveniles have lower self-control, are more constrained by family and external factors, and have smaller and simpler daily activity ranges and trajectories. Additionally, juveniles tend to be naive and impulsive, making their crimes often characterized by suddenness and randomness. They are more susceptible to external environmental factors that trigger criminal behavior (Simões et al., 2024). Therefore, research on juvenile delinquency should focus more on the impact of micro-scale spatial environmental characteristics and utilize improvements in the built environment and strengthening of social organizations for prevention.

Due to the difficulty of collecting micro-level built environment data, past juvenile delinquency research has primarily analyzed data from macro and meso scales, lacking fine-grained research from a micro-street perspective. Traditional environmental criminology research has mainly explored the relationship between crime and urban micro-built environments through manual audits and surveys. For instance, Hedayati Marzbali et al. (2016) conducted detailed on-site observations and face-to-face interviews with 456 households in two different neighborhood communities in Malaysia to collect data and investigate how community physical characteristics influence crime rates. Zeng et al. (2021) collected data on street usage, street layout, and environmental item setting for 135 streets in Shanghai’s Pudong New Area through field surveys to create an environmental element set for crime data analysis. While these methods provide relatively detailed data, they are time-consuming and labor-intensive, making it difficult to expand the study area to the entire city and potentially introducing subjectivity and errors. Although satellite imagery data enables large-scale quantification of urban built environments, satellite imagery is primarily captured from an aerial perspective, making it difficult to accurately depict street environments from a human perspective (Wolfe & Mennis, 2012).

Street view imagery, as one of the most effective data forms for reflecting real-world street visual environments (Biljecki & Ito, 2021), offers wide coverage, easy accessibility, and the ability to quickly extract street environment features using computer vision techniques, providing new insights into the relationship between crime and urban micro-built environments. For example, researchers can utilize street view image data to assess information about the material environment and emotional perception of urban spaces, thereby inferring the impact of urban environments on criminal activity (Dubey et al., 2016; He et al., 2017; Kang & Kang, 2017; Zhang et al., 2021). In this field, an increasing number of researchers are employing integrated analysis of multi-source big data, including street view data, to more precisely quantify street environments, enabling a more comprehensive and accurate study of the impact of urban environments on criminal activity (He et al., 2020, 2022; Hipp et al., 2022; Kim & Lee, 2023; Yao et al., 2023; Yunus & Loo, 2024). Due to the wide applicability of such data, researchers have conducted studies in cities with diverse socioeconomic backgrounds worldwide. Zhou et al. (2021) combined Google Street View data with deep learning methods to extract street features in Cincinnati, exploring micro-built environment characteristics associated with drug locations and comparing them to street robbery incidents. They found that high road accessibility might facilitate drug trafficking. Yue et al. (2022) utilized Baidu Street View data to quantify street population and built environment elements in ZG City, a coastal city in southern China. Based on this data, they examined the impact of these factors on street property and violent crime. The results showed that street population had a significant positive impact on street property crime but no significant impact on street violent crime. In conclusion, emerging big data, such as street view images, plays a crucial role in researching the relationship between crime and the urban built environment. With continuous technological advancements and applications, research in this field will achieve more profound and accurate results, providing valuable insights for urban planning and crime prevention.

Juvenile delinquency is a significant topic in modern criminology, and providing a healthy growth environment for adolescents is a key focus of social work in countries worldwide. Overall, there is a lack of large-scale quantitative research on the impact mechanisms of juvenile delinquency from a street environment perspective. Therefore, it is particularly important to utilize street view data and related technical methods to supplement research in this dimension. This study explores the influence of visual environment on juvenile delinquency from a micro-street perspective, addressing the shortcomings of existing research on the relationship between juvenile delinquency and the built urban environment by focusing on a narrower dimension. It also refines traditional street crime research by targeting the adolescent population, providing new strategies and perspectives for the prevention and control of such crimes and the construction of urban streets. Furthermore, we use multi-source data, including street view images, to break through the limitations of traditional data, providing a reference value and theoretical research assistance for further in-depth studies.

2 Data and methods

2.1 Study area and analysis process

Manhattan, primarily consisting of Manhattan Island, Marble Hill, and several surrounding smaller islands, is a significant administrative district in New York City, USA, and a globally renowned financial, commercial, and cultural hub. Manhattan is divided into various zones, with Manhattan Island typically categorized into three sections: Lower Manhattan, Midtown Manhattan, and Upper Manhattan. Upper Manhattan refers to the northern part of Manhattan Island (north of 59th Street), Lower Manhattan to the southern part (south of 14th Street), and the area between 14th Street and 59th Street is known as Midtown. According to the New York City Police Department, the crime rate in Manhattan has consistently remained high. Juvenile delinquency, in particular, has not only affected the security of Manhattan but also the growth and future of adolescents, prompting significant attention. Based on this, we selected Manhattan Island (excluding Central Park) as the study area (Fig. 1) and conducted the research using road segments within the road network as the basic units.

Fig. 1
figure 1

Study area range

Full size image

This study aims to explore the impact of the built street environment on the occurrence of juvenile delinquency. We first establish the theoretical foundation for our research by integrating CPTED (Crime Prevention Through Environmental Design) and Social Disorganization theory. We then utilize deep learning, mathematical statistics, and spatial analysis techniques to process relevant research data, constructing separate street physical and social environment indicator systems. We aggregate crime counts and various indicators to the road segment level for subsequent analysis. Next, we investigate the spatial distribution characteristics of juvenile delinquency to identify key study areas for juvenile crime. We then proceed to develop a regression model to analyze the relationship between juvenile delinquency and the street environment. Since the explained variable in our study, crime count, is non-negative discrete data representing the number of crime events, we consider using a Poisson regression model instead of traditional least squares regression for analysis. Poisson regression, as a type of global regression model, assumes that the explained variable for all samples within the entire space is influenced by the same explanatory variables, meaning that the model parameters are fixed across the entire space. However, this model has limited ability to model spatial heterogeneity and cannot fully account for local features within the space. Compared to global regression models, the GWR (Geographically Weighted Regression) model can better capture local relationships within the space. Compared to spatial lag and spatial error models with fixed coefficients, GWR allows coefficients to vary spatially, enabling more accurate capture of relational differences across locations. This provides more detailed spatial pattern information and facilitates a deeper understanding of the drivers of spatial heterogeneity. However, the GWR model may face challenges in terms of computational efficiency and parameter estimation accuracy, requiring careful consideration based on specific circumstances. Therefore, we will use the results of the Poisson global regression model to determine whether a GWR model needs to be established to explore the spatial heterogeneity of explanatory variables. Finally, by selecting appropriate model results, we will analyze the effects of each indicator on juvenile crime and provide relevant recommendations for urban planning and street construction (Fig. 2).

Fig. 2
figure 2

Analysis framework

Full size image

2.2 Data processing and sources

2.2.1 Crime data

The crime data was sourced from the NYPD Complaint Data Historic dataset. This dataset includes all valid felony, misdemeanor, and violation crimes reported to the New York City Police Department (NYPD) from 2006 to present, and provides 35 attribute fields, including crime date, crime type, perpetrator age range, and latitude and longitude coordinates.

We filtered the data to include only cases with a crime date between 2015 and 2019, a perpetrator age under 18, and a crime location within Manhattan. This ensured that the perpetrators were juveniles and the crimes occurred within the study area. We then further filtered the data by crime type. Juvenile delinquency in the United States can be broadly categorized into four major types: violent crimes, property crimes, drug offenses, and public order crimes (Lu, 2022). Considering the characteristics and frequency of each crime type, we selected violent crimes and property crimes for our study. We excluded crime types that were not related to street crime or were minimally influenced by the built street environment, such as fraud and gambling.

Finally, we extracted crime points within a 50 m buffer zone of the road centerline based on the latitude and longitude coordinates of each crime record, ensuring that the crimes occurred within the visible range of the street. This resulted in 10,358 crime points that met our criteria. These points were then aggregated to the road segment level for subsequent analysis.

2.2.2 Road network data

This study investigates the relationship between the urban street environment and juvenile delinquency using road segments broken down by road intersections as the basic spatial unit. The original road network vector data was obtained from the 2019 TIGER/Line® dataset provided by the U.S. Census Bureau. Streets, typically defined as roads for pedestrian and vehicle traffic, are essential public spaces in cities, serving multiple functions including transportation, commerce, culture, and social interaction. However, highways, expressways, and elevated roads are primarily designed for high-speed vehicle travel and lack the complex functionality and public space characteristics of urban streets. They also do not align with the environmental features associated with street crime. Therefore, we excluded these types of roads and focused our research on road segments within urban arterial roads, collector roads, and local roads.

2.2.3 Street view data

Street view image data was obtained from Google Maps. Drawing on previous research (He et al., 2022; Luo et al., 2022), a road network sampling grid with 50 m intervals was generated prior to data collection to ensure that the acquired images comprehensively capture the built environment information of the corresponding road segments. A request URL was constructed to download street view images. Table 1 presents the parameters and descriptions required for building the URL. The parameter “key” represents the Google API key, which is obtained through application.

Table 1 Details of google street view API parameters
Full size table

In summary, each sampling point required the collection of street view images in four directions to reconstruct the visual environment experienced by a perpetrator on the street. Images with obstructed views or not located on the street (e.g., returning indoor images) were visually eliminated and re-sampled. Although the crime data spans from 2015 to 2019, Manhattan, as the central urban area of New York City, has a long history of development and its environment has not undergone significant changes. Therefore, we consider the crime data and street view image data to be temporally aligned. Additionally, due to limitations in image acquisition methods, some streets may not have available street view images, resulting in blank images for the corresponding sampling points. We screened and excluded these streets without street view images to ensure that all road segments in this study are usable. Ultimately, 55,372 images were collected within the study area, which were then stitched together to produce 13,843 panoramic street images (Fig. 3).

Fig. 3
figure 3

Street panoramic image stitching example

Full size image

We utilized deep learning techniques, employing a PSPNet model (https://github.com/CSAILVision/semantic-segmentation-pytorch) pre-trained on the ADE20K dataset, to perform semantic segmentation and feature extraction on the street view images (Fig. 4). The extracted annotated classes were then categorized into 11 major categories (Table 2) for constructing and calculating street environment feature indicators. The PSPNet model, proposed in 2017, is a highly effective and accurate deep neural network architecture that achieved new records for mIoU accuracy on the PASCAL VOC 2012 and Cityscapes datasets, reaching 85.4% and 80.2%, respectively. This demonstrates the PSPNet model’s ability to accurately segment objects and scenes within images (Zhao et al., 2017). Additionally, due to the pyramid pooling module’s capability to capture information at multiple scales, the PSPNet model can handle complex scenes with fine-grained details. The use of dilated convolutions allows the PSPNet model to retain spatial information while expanding the receptive field, resulting in more accurate segmentation results.

Fig. 4
figure 4

Output semantic segmentation result example

Full size image
Table 2 Annotated categories based on ADE20K dataset
Full size table

Our theoretical framework is grounded in both CPTED and Social Disorganization theory, enabling us to examine juvenile delinquency in urban streets from the perspectives of environmental criminology and urban planning. CPTED is a comprehensive theoretical and practical approach aimed at reducing crime occurrence and fear of crime through environmental design and planning. Its core principle is crime prevention through the creation of safer and more community-oriented environments. CPTED originated in the United States in the 1960s. Building upon Jane Jacobs’ “eyes on the street” theory (Jacobs, 1961), architect Oscar Newman (Newman, 1975) and criminologist C. Ray Jeffery (Jeffery, 1971) formally established the term CPTED, laying the foundation for the first generation of CPTED, which encompassed six concepts: territoriality, surveillance, access control, maintenance, activity support, and target hardening. The first generation of CPTED primarily focused on physical dimensions. The second generation of CPTED theory expanded upon the first generation by incorporating social concepts, placing greater emphasis on understanding the social motivations behind crime and the factors contributing to its occurrence (Mihinjac & Saville, 2019).

Given our focus on the relationship between street built environment and juvenile delinquency, and the difficulty in directly identifying target hardening from street view images, we selected five concepts from the first generation of CPTED as our theoretical foundation: territoriality, surveillance, access control, maintenance, and activity support. CPTED has been widely applied in previous crime research and urban planning practices, including cases targeting youth populations. For example, Bradshaw et al. (2015) developed a School Assessment for Environmental Typology by measuring CPTED indicators such as lighting conditions, graffiti, and surveillance. Lamoreaux and Sulkowski (2020) found that schools and communities applying CPTED principles such as natural surveillance and physical maintenance effectively prevent juvenile violence. Compared to adults, juveniles have lower self-control and are more susceptible to external influences that trigger crime. Therefore, in selecting indicators, we prioritized the impact of order maintenance and social atmosphere creation in the built environment on juvenile psychology, rather than focusing on the suitability of scenarios for premeditated crimes as in adult crime research. Ultimately, we constructed a street physical environment index system based on the characteristics of juveniles and drawing on previous research (Kim & Hipp, 2021; Lin et al., 2021; Su et al., 2023), which includes 10 street view indicators potentially related to crime. The calculation methods for each indicator are described in Table 3.

Table 3 Indicator system for the physical environment of streets
Full size table

2.2.4 Social statistical data

Social statistical data was sourced from the 2019 ACS 5-Year Estimates Detailed Tables dataset provided by the U.S. Census Bureau. This dataset is part of the ACS (American Community Survey) dataset, a significant survey used to determine various characteristics of the American population, such as education level, income level, employment status, poverty rate, housing conditions, and more.

Social disorganization theory, proposed by American criminologists C. R. Shaw and H. D. McKay (Shaw & McKay, 1942), was initially developed to explain the root causes of juvenile delinquency in urban communities. The theory posits that crime in urban communities primarily stems from social disorganization, meaning that the breakdown of social structures and the weakening of social control within communities create environmental incentives for crime and deviant behavior. Manifestations of social disorganization include factors such as low income, low education levels, high unemployment rates, family breakdown, and immigrant migration. These factors can lead to unstable population composition within communities, weak social networks, and insufficient public services. These issues weaken social control mechanisms within the community, leaving juveniles without adequate supervision and guidance, making them more susceptible to engaging in criminal behavior. Therefore, social disorganization theory emphasizes the importance of social structures and social control, suggesting that strengthening community social control and social networks can effectively prevent and reduce crime. This theory has become a crucial foundation for studying urban crime and a valuable reference for crime prevention and community governance.

Guided by social disorganization theory, we extracted seven street social environment indicators potentially related to juvenile delinquency from the aforementioned statistical dataset: population density (people/square meter), median household income (USD thousand), percentage of juvenile population (%), household poverty rate (%), percentage of single-parent households (%), enrollment rate (%), and racial diversity. Racial diversity can be measured using the Shannon Wiener Diversity Index, calculated as follows:

$$\begin{array}{c}H=-\sum\limits_{1}^{S}\left({P}_{i}\mathit{ln}\left({P}_{i}\right)\right)\end{array}$$
(1)

Where H represents the value of the Shannon index, S is the total number of races in the area, and Piis the proportion of the population of the i-th race to the total population. A higher Shannon index indicates greater racial diversity within the area, meaning a rich variety of races with relatively equal proportions. A lower Shannon index indicates lower racial diversity, meaning a limited variety of races or a dominant population of a particular race.

Since our aim is to investigate the impact of the built street environment on juvenile delinquency, these social environment indicators serve as control variables to assist in the analysis of street physical environment indicators, mitigating the influence of social environmental factors on the final results.

3 Results

3.1 Spatial distribution characteristics of juvenile delinquency

This section analyzes the spatial distribution of juvenile delinquency based on the crime categories mentioned earlier. The number of total crimes, violent crimes, and property crimes were aggregated onto the streets, with road segments serving as the basic spatial units. The crime counts were classified using the natural breaks method. Figure 5 shows that the distribution of crime counts on urban streets is uneven, with significant variations in crime counts between different streets. Most streets have low crime counts, typically below 4. However, a small number of streets exhibit higher crime counts and a certain clustering pattern, particularly in the east side of Lower Manhattan, Midtown Manhattan, and Harlem in Upper Manhattan. Comparing violent crimes and property crimes, both types of crimes occur frequently in Harlem, but violent crimes exhibit a more concentrated spatial distribution pattern overall. These findings suggest potential spatial clustering of criminal activity, which requires further investigation using spatial autocorrelation analysis.

Fig. 5
figure 5

Spatial distribution of crime counts for three crime types

Full size image

Spatial autocorrelation analysis is a statistical method used to explore the correlation of spatial data. It can be measured using Moran’s I. By calculating Moran’s I, we investigated whether spatial clustering exists in juvenile crime (Table 4). The table shows a Moran’s I value of 0.083, a Z-score of 18.664, and a P-value of 0.000. Therefore, we can conclude that spatial clustering exists in juvenile crime within the study area.

Table 4 Spatial autocorrelation analysis results
Full size table

Hot spot analysis (Getis-Ord Gi*) is a statistical method used to explore hot and cold spots in spatial data. We used the hot spot analysis tool in ArcGIS Pro to identify cold and hot spot road segments for juvenile crime in the study area and assess their significance (Fig. 6).

Fig. 6
figure 6

Spatial distribution of crime hotspots for three crime types

Full size image

Based on the spatial distribution of juvenile crime cold and hot spot road segments, combined with an analysis of the region’s socioeconomic and cultural environment, we observed the following:

  1. (1)

    Areas with lower household income levels are more prone to juvenile delinquency. For example, Harlem and the Lower East Side, being among the areas with the lowest household incomes in Manhattan, exhibit significantly more crime hot spot road segments compared to the Upper East Side and Upper West Side.

  2. (2)

    Violent crimes are more concentrated in areas with high racial diversity indices. For instance, the Harlem area is predominantly Black but also includes various other races. A high racial diversity index indicates a complex racial composition, which can lead to cultural conflicts and consequently violent crimes.

  3. (3)

    Property crimes tend to occur in economically developed areas with high pedestrian traffic, such as shopping malls, parks, plazas, and transportation hubs. Examples include Madison Square Garden, Times Square, and Rockefeller Center in Midtown.

  4. (4)

    Population density and the percentage of the juvenile population do not have a direct relationship with the clustering of juvenile delinquency. For example, the Upper East Side and Upper West Side, known for their upscale residential areas, do not exhibit a proportional relationship between juvenile crime counts and population density. The western side of Lower Manhattan and the Harlem area in Upper Manhattan both have high percentages of juvenile populations, but they show significant differences in the distribution of crime cold and hot spots. Therefore, compared to population-related factors, other socioeconomic factors such as poverty rate, percentage of single-parent households, and racial diversity have a more significant impact.

In conclusion, the spatial distribution of juvenile delinquency is influenced by multiple factors. Considering the social, economic, and cultural characteristics of different areas in Manhattan, combined with the spatial distribution of crime cold and hot spots, we ultimately identified three key study areas for juvenile crime: Lower Manhattan, Midtown Manhattan, and the Harlem area in Upper Manhattan (Fig. 7).

Fig. 7
figure 7

Key study area

Full size image

3.2 Spatial modeling analysis of juvenile delinquency and street environment indicators

3.2.1 Model selection and performance evaluation

To measure and compare the performance of different models, two diagnostic indicators, PDE (Percent Deviance Explained) and AICc (Akaike Information Criterion corrected), were selected for evaluation. PDE refers to the percentage of variance in the response variable explained by the model, with values ranging from 0 to 1. It is commonly used to assess the interpretability of the model and compare the fitting effects of different models. AICc can be used to examine model performance and compare regression models. Models with lower AICc values fit the observed data better. AICc is not an absolute measure of goodness of fit, but it is very useful for comparing models that are applicable to the same dependent variable but have different explanatory variables. If the AICc values of two models differ by more than 3, the model with the lower AICc value is considered a better model.

Considering that juvenile delinquency is not only influenced by the physical environment of streets but also closely related to the urban social environment, the seven street social environment indicators mentioned earlier that may be related to juvenile delinquency were set as control variables in the study. This was done to more accurately explore the impact of explanatory variables on the final results. Both explanatory variables and control variables were normalized in advance to avoid analysis errors caused by dimensional differences.

The results of Poisson global regression for the two crime types in Upper Manhattan, Midtown Manhattan, and Harlem are shown in Table 5. The PDE values of all six models are below 0.3, indicating that the explanatory power of the Poisson global regression model is not strong. This is likely due to the inability to capture the spatial correlation of variables. Additionally, the table shows that some variables have opposite effects on the two crime types in different areas. For example, AI has a positive effect on both crime types in Lower Manhattan but a negative effect in Midtown Manhattan. This further indicates that the impact of explanatory variables on juvenile crime is spatially non-stationary and requires the establishment of a GWR model to further explore and study the influence mechanism.

Table 5 Poisson regression model analysis results
Full size table

In the GWR model, the kernel function determines the weight distribution of sample points within the neighborhood, while the bandwidth controls the size of the neighborhood. Therefore, choosing the correct kernel function type and bandwidth selection method can enable the GWR model to better capture the local variability and heterogeneity of spatial data, improving the model’s predictive accuracy and explanatory power. This study uses the adaptive Bi-square function to simulate spatial weights. This kernel function has the advantage of better handling uneven data distribution and has better fitting ability compared to traditional fixed bandwidth kernel functions. The optimal bandwidth is determined by the Golden Section search method, which finds the bandwidth size with the smallest AICc value. Its advantages include finding the optimal solution in a shorter time and being more robust and reliable compared to other bandwidth selection methods. The GWR model type is still chosen as Poisson. The final analysis results of the GWR model for the two crime types in the three areas are shown in Tables 6, 7 and 8.

Table 6 GWR model analysis results of Lower Manhattan
Full size table
Table 7 GWR model analysis results of Midtown Manhattan
Full size table
Table 8 GWR model analysis results of Harlem region
Full size table

Comparing the Poisson global regression model and the GWR model, we observe a significant decrease in AICc values and a noticeable increase in PDE values. This indicates that the GWR model has a stronger fitting ability and model performance compared to Poisson global regression, and it also has a better explanatory power for variables. Therefore, the analysis results of the GWR model should be used to analyze and study the relationship between juvenile crime and street environment indicators.

3.2.2 GWR model analysis results

Impact of territoriality indices on juvenile crime

Territoriality indices include SBI and EPI. The spatial distribution of the regression coefficients obtained from the GWR model for these two indices is shown in Figs. 8 and 9.

Fig. 8
figure 8

Spatial distribution of SBI regression coefficients

Full size image
Fig. 9
figure 9

Spatial distribution of EPI regression coefficients

Full size image

Figure 8 shows the spatial distribution of the coefficients for the impact of SBI on the two crime types in the three areas. In most areas of Lower Manhattan, SBI has a negative impact on both violent crimes and property crimes. As SBI increases, the number of both crime types decreases significantly. This means that increasing the number of street enclosure elements can enhance the sense of boundary, effectively inhibiting juvenile crime. Near the Holland Tunnel in northern Tribeca, SBI has the highest positive impact on both crime types. This may be because the tunnel area has a strong sense of boundary but a low population density, with sparse pedestrian activity and mainly vehicle traffic, which does not provide suitable opportunities for criminals. In Midtown Manhattan, the negative impact of SBI on both crime types is more significant. For both crime types, SBI only shows a positive impact in the central area of Midtown around the Empire State Building and near the Midtown Tunnel along the East River in Queens. In Harlem, the distribution impact of SBI on violent crimes and property crimes is roughly the same. It has a positive impact in the western part of Manhattanville, the central part of Harlem, and East Harlem, and a negative impact in other areas.

Figure 9 shows the spatial distribution of the coefficients for the impact of EPI on the two crime types in the three areas. As shown in the figure, in most areas, EPI has a negative impact on both violent crimes and property crimes. As EPI increases, the number of both crime types decreases significantly. This means that increasing the ratio of buildings, trees, and other enclosure elements to roads enhances people’s perception of enclosure on the street, strengthens the sense of domain, and effectively inhibits juvenile crime. Compared to violent crimes, the negative impact of EPI on property crimes is more pronounced. This may be because a strong sense of enclosure not only enhances natural surveillance by pedestrians but also increases the sense of oppression on the street. For juveniles with emotional instability, this can induce violent conflicts while inhibiting crimes such as theft.

In conclusion, for both crime types, the distribution of SBI and EPI is similar. They are significantly negatively correlated in most areas and positively correlated in areas with low population density. This indicates that strengthening spatial boundaries can generally enhance pedestrian safety, deter potential criminals, and reduce crime. It is worth noting that in areas with low population density, increasing spatial boundaries can actually induce criminal behavior. This may be because in areas with low pedestrian traffic, there is a lack of natural surveillance, and the addition of enclosures can provide criminals with hidden crime locations, reducing the cost of crime. Therefore, the strengthening of spatial domains should be determined based on different urban areas and should not be overly emphasized, leading to an increased sense of oppression on the streets.

Impact of surveillance indices on juvenile crime

Surveillance indices include SLI, SVI, and TFI. The spatial distribution of the regression coefficients obtained from the GWR model for these three indices is shown in Figs. 10, 11 and 12.

Fig. 10
figure 10

Spatial distribution of SLI regression coefficients

Full size image
Fig. 11
figure 11

Spatial distribution of SVI regression coefficients

Full size image
Fig. 12
figure 12

Spatial distribution of TFI regression coefficients

Full size image

Figure 10 shows the spatial distribution of the coefficients for the impact of SLI on the two crime types in the three areas. In the southern end of the Financial District in Lower Manhattan, the intersection of Greenwich Village and SoHo, the East Village, and most areas of Midtown and Harlem, SLI is negatively correlated with both crime types. As SLI increases, street lighting space increases, and the number of crimes decreases significantly. This indicates that lighting can provide a sense of security for pedestrians and enhance natural surveillance on the street, increasing the likelihood of potential criminals being detected, achieving a crime prevention effect. Contrary to adult crime [33], for economically developed street areas, the increase in SLI does not inhibit the occurrence of juvenile property crimes, but rather promotes it. For example, near Times Square in Midtown, SLI is positively correlated with property crimes. As the commercial center of Manhattan, this area is economically developed and densely populated. In addition to streetlights, the buildings in Times Square are also brightly lit, easily attracting pedestrian attention. This type of street environment inhibits the occurrence of violent crimes but provides opportunities for juvenile property crimes.

Figure 11 shows the spatial distribution of the coefficients for the impact of SVI on the two crime types in the three areas. In the East Village and Lower East Side in Lower Manhattan, Midtown except for the central area, and Harlem except for the central and western areas, SVI is significantly positively correlated with crime. This means that as SVI increases, spatial visibility decreases, and the number of juvenile crimes increases. This indicates that an increase in visual obstructions in the streets of these areas can induce crime. This may be because an increase in obstructions can increase visual blind spots on the street, providing opportunities for crime. In Greenwich Village in Lower Manhattan, near Times Square in Midtown, and the central and western areas of Harlem, SVI is significantly negatively correlated with both crime types. These areas are economically developed, with well-maintained streets. The increase in obstructions like poles and pillars can enhance the sense of order on the street, guiding clear sightlines, thereby reducing crime opportunities and lowering crime fear.

Figure 12 shows the spatial distribution of the coefficients for the impact of TFI on the two crime types in the three areas. As shown in the figure, in most areas, TFI has a positive impact on both crime types. As TFI increases, the number of both crime types increases significantly. This means that an increase in pedestrians and vehicles can induce juvenile crime. This conclusion differs from the statement in CPTED that pedestrians and passing vehicles can effectively monitor the street, thereby inhibiting crime. The reason may be that in this study, the number of pedestrians and vehicles was extracted from street view images, but there are many vehicles parked illegally and occupying road space, which do not play a role in natural surveillance. Instead, they trigger the “broken window effect”, creating a chaotic environment that induces crime. While crowds can play a role in natural surveillance to some extent, in bustling urban areas, crowded crowds can actually provide hiding spaces for criminals.

In conclusion, for both crime types, the influence mechanism of natural surveillance-related factors varies greatly in different urban areas. In economically developed areas, good lighting can inhibit crime. In economically underdeveloped areas, high-visibility spaces are more likely to play a role in natural surveillance, reducing crime.

Impact of access control indices on juvenile crime

Access control-related indices include PSI and AI. The spatial distribution of the regression coefficients obtained from the GWR model for these two indices is shown in Figs. 13 and 14.

Fig. 13
figure 13

Spatial distribution of PSI regression coefficients

Full size image
Fig. 14
figure 14

Spatial distribution of AI regression coefficients

Full size image

Figure 13 shows the spatial distribution of the coefficients for the impact of PSI on the two crime types in the three areas. As shown in the figure, PSI has a significant positive impact on property crimes in Midtown. As PSI increases, the number of property crimes increases significantly. This means that as pedestrian ground elements increase, the likelihood of criminals escaping increases, thereby inducing juvenile crime.

Figure 14 shows the spatial distribution of the coefficients for the impact of AI on the two crime types in the three areas. In most areas of Lower Manhattan, AI has a positive impact on property crimes. As AI decreases, the number of property crimes decreases significantly. This means that as the ratio of pedestrian and vehicular road elements to fence elements decreases, accessibility decreases. By controlling street traffic, juvenile property crimes are effectively inhibited.

In conclusion, for both crime types, the overall impact of access control-related indices is positive. Increasing fences, landscaping, and other measures to restrict passage or control traffic in the streets, clearly distinguishing between public and private spaces, can effectively prevent crime.

Impact of maintenance indices on juvenile crime

Maintenance indices include SI and GVI. The spatial distribution of the regression coefficients obtained from the GWR model for these two indices is shown in Figs. 15 and 16.

Fig. 15
figure 15

Spatial distribution of SI regression coefficients

Full size image
Fig. 16
figure 16

Spatial distribution of GVI regression coefficients

Full size image

Figure 15 shows the spatial distribution of the coefficients for the impact of SI on the two crime types in the three areas. In the streets near the western side of the West Village in Lower Manhattan, Tribeca, the southern side of Chelsea in Midtown, and the southern part of Harlem, SI is significantly positively correlated with both crime types. This means that as sky visibility increases, the number of juvenile crimes increases. These areas have wide roads, low building heights, and a poor sense of enclosure, with open spaces leading to a decrease in pedestrian safety and making it easier for criminals to search for targets and escape quickly after committing a crime. In other areas, SI has a negative impact on violent crimes and property crimes. This may be because in areas with high building density and dense pedestrian traffic, increasing sky visibility can provide a comfortable environment, enhancing people’s sense of belonging and visibility, thereby reducing crime.

Figure 16 shows the spatial distribution of the coefficients for the impact of GVI on the two crime types in the three areas. In the streets of the Financial District, Tribeca, SoHo in western Lower Manhattan, the central area of Midtown, and the central and western areas of Harlem, GVI has a negative impact on both crime types. As GVI increases, the number of juvenile crimes decreases. This means that increasing vegetation elements can inhibit juvenile property crimes. In the streets of the Holland Tunnel, Manhattan Bridge, and East Harlem in Lower Manhattan, the impact of GVI on both crime types is positive. This indicates that in street areas with low pedestrian activity or poor socioeconomic conditions, the obstruction of street greenery can provide criminals with a hidden crime environment, creating blind spots and increasing crime opportunities. In economically developed and well-constructed street environments, the presence of vegetation can attract pedestrian traffic, increase social supervision, and have a positive impact on people’s emotions and behavior, thereby inhibiting crime.

Impact of activity support indices on juvenile crime

Figure 17 shows the spatial distribution of the coefficients for the impact of SAI on the two crime types in the three areas. In the streets of Greenwich Village, SoHo, East Village, and Lower East Side in Lower Manhattan, the streets along the East River in Midtown, and the streets in the south-central part of Harlem, SAI has a positive impact on violent crimes and property crimes. In other areas, it has a negative impact. This means that as SAI increases, the number of benches, trash cans, and other small features in the street increases, and the number of both crime types decreases. From the regression coefficients, the impact of SAI on both crime types is not significant. In Lower Manhattan and the East River area of Midtown, where building density and height are low, the lack of street furniture makes it difficult for people to gather, thereby reducing opportunities for juvenile crime. In other areas, such as the central area of Midtown, where pedestrian activity density is high, increasing landscape features and providing convenient facilities can improve spatial convenience and street appearance, increasing people’s willingness to be active in the space and their sense of belonging, making it difficult for criminals to commit crimes and achieving crime prevention.

Fig. 17
figure 17

Spatial distribution of SAI regression coefficients

Full size image

3.2.3 Urban design recommendations

This study focuses on the impact of street built environment on juvenile delinquency, therefore, the proposed policy recommendations revolve around the physical environment construction of urban streets.

Optimize fencing and building layout. In areas with high pedestrian traffic, such as urban centers, setting up fences, railings, and other street enclosures can clearly define street space, enhancing the sense of spatial boundaries and making juveniles more aware of the increased risk of being caught for committing crimes. Reasonable control of the height and spacing of street-facing buildings and trees, optimizing street enclosure based on regional characteristics, focusing pedestrian lines of sight, and strengthening territoriality and natural surveillance functions are crucial. Excessive openness in street enclosure can evoke a sense of emptiness and insecurity, while excessive enclosure can create a sense of oppression, which can more easily trigger violent conflicts among juveniles.

Control street space visibility. In street environments with strong enclosure, numerous obstructions, and already robust natural surveillance, excessive street lighting may increase the exposure of potential victims, potentially inducing criminal activity. Therefore, it is advisable to enhance natural surveillance by appropriately increasing street lighting. Additionally, controlling building density on plots, rationally designing street landscapes, and other measures can enhance spatial visibility, increasing the likelihood of potential criminals being detected by the public and consequently raising juveniles’ risk perception of attempting criminal acts.

Maintain street environment order and create a comfortable street environment. A chaotic spatial environment represents a lack of control and can more easily stimulate juvenile criminal desires. Maintaining a good street environment image can enhance juveniles’ sense of belonging. Specific measures include regular cleaning, maintenance, and renewal of facilities and buildings in the environment, improving street greening, strengthening management of roadside parking areas, and other measures to maintain the cleanliness and orderliness of public spaces.

Rational placement of landscape elements to enhance street space vitality. Through rational street design, public spaces can be infused with vitality, providing effective surveillance sources and promoting legitimate activities. Specifically, adding landscape elements to create resting spaces that combine privacy and security, providing convenient facilities (such as trash cans and benches), and other measures can increase the attractiveness and convenience of the space, increasing people’s willingness to engage in activities within the space, building positive neighborly relationships, strengthening community cohesion, and ultimately achieving the goal of preventing juvenile delinquency.

4 Conclusion and discussion

4.1 Conclusion

Based on the above analysis, the following conclusions can be drawn regarding the impact mechanism of street environment on juvenile delinquency:

  1. (1)

    Territoriality elements (e.g., fences and building enclosure) can significantly reduce juvenile delinquency, but in areas with low population density, the enhancement of spatial boundary sense can actually induce crime. Strengthening territorial boundaries can enhance pedestrian safety and deter potential criminals, but in the absence of natural surveillance, it can reduce the cost of crime, and the oppressive feeling of high enclosure is more likely to induce juvenile violent crime.

  2. (2)

    The impact mechanism of natural surveillance elements (e.g., streetlights or obstructions) on juvenile delinquency varies in different urban areas. In economically developed areas with high pedestrian traffic, good lighting is more likely to induce juvenile property crime. In economically underdeveloped areas, high-visibility spaces are more likely to play a role in natural surveillance, reducing juvenile crime.

  3. (3)

    Access control elements have a significant inhibitory effect on juvenile delinquency. By adding fences and other measures to restrict passage or control traffic in the streets, public and private spaces can be clearly distinguished, effectively preventing juvenile delinquency.

  4. (4)

    Maintenance elements can significantly inhibit crime in areas with high pedestrian traffic. A good built environment can attract pedestrian traffic, increase social supervision, and have a positive impact on people’s emotions and behavior, thereby inhibiting crime.

  5. (5)

    In areas with high pedestrian activity density, activity support elements (e.g., street furniture) can to some extent inhibit juvenile delinquency.

4.2 Discussion

This paper uses multi-source data to conduct a quantitative study on the impact mechanism of urban street environment on juvenile delinquency, which is helpful in identifying key street environment factors that affect criminal behavior. The suggestions for urban planning and street design based on this study have certain significance in preventing juvenile delinquency. However, this paper still has the following shortcomings:

  1. (1)

    Data aspects. This paper mainly uses street view data, statistical data, and other geospatial data for analysis, but this is not comprehensive enough for revealing street characteristics. In addition, there are certain errors in the process of extracting indicators from street view images, and the scale of various elements in the street, such as building height and volume, cannot be accurately measured. As a result, the obtained results cannot provide specific street construction standards and can only qualitatively describe the relationship between juvenile delinquency and various street elements. To address these issues, future research should consider introducing aerial data to measure the planform of urban streets and introducing POI data to refine the social attributes of streets, enriching the research dimensions.

  2. (2)

    Technical method aspects. Some street characteristics are currently difficult to quantify on a large scale, such as subjective perception indicators such as the sense of order in street construction. Due to the complex and diverse nature of street characteristics, while this study endeavors to comprehensively quantify street built environment information, it may overlook the potential influence of some unmeasured factors. In terms of modeling analysis, the chosen model possesses the advantage of exploring spatial heterogeneity, but it neglects potential threshold or interaction effects of certain factors. In addition, this paper focuses on the impact of street environment elements on juvenile delinquency, but does not involve crime prediction. In the future, it is possible to consider using artificial evaluation combined with machine learning and deep learning methods to quantify the perceptual characteristics of streets and predict juvenile crime-prone locations, providing more intuitive references for juvenile delinquency prevention.