Although cannabis is the most frequent illicit drug consumed in Sweden, little is known about the situations in which cannabis trade, possession and use occur. Following a recent strand of international research on the efect of recreational drugs on crime, this study uses a unique specially tailored database, Geographical Information Systems (GIS) and regression models, to investigate the situational conditions of cannabis ofenses as they are detected in Stockholm, Sweden. Cannabis coincides with the location of drug markets initially delimited by the police but also extends over to locations far from the radar of the police, such as private residences (comfort places). Modeling results indicate that several land uses (convergent public places) have signifcant predictive value of the geography of cannabis ofenses after controlling for other neighborhood characteristics. The article fnishes by stating new research questions and making recommendations for practice.

Drawing from environmental criminology principles, this article explores the existing literature to assess the utility of remote sensing data in detecting and analysing features in the urban environment that are associated with crime occurrence. A systematic review of the literature in the English language from 2003 until the first half of 2023 from two major databases, Scopus and Science Direct, is carried out. As many as 910 publications were selected, from which 36 publications satisfied the selection criteria. Findings show that neighborhood's design has a quantifiable imprint that is possible to be observed with very high spatial-resolution imagery. Given its high spatial and temporal resolution, remote sensing data can to different degrees support the identification of criminogenic features in urban environments (streets and roads, property boundaries, housing density, characteristics and density of vegetation as well as luminosity levels), but when it is used for the detection of potentially illegal activities, infringement of people's privacy and methods lacking validation still present serious concerns. The article concludes with a discussion of the opportunities and challenges of using remote sensing data in crime analysis.

By deploying remotely sensed data together with spatial statistical modeling, we use regression modeling to investigate the relationship between the density of the built environment and two types of crime. We show how the Global Human Settlement Layer (GHSL) data set, which is a measure of building density generated from Sentinel 2A satellite imagery, can be used to create different indexes to describe the built environment for the purpose of analyzing crime patterns for indoor crimes (residential burglary) and open space crimes (street theft). Analysis is at neighborhood level for Stockholm, Sweden. Modeling is then extended to incorporate six planning areas which represent different neighborhood types within the city. Modeling is further extended by adding selected social, economic, demographic and land use variables that have been found to be significant in explaining spatial variation in the two crime categories in Stockholm. Significant associations between the GHSL-based indexes and the two crime rates are observed but results indicate that allowance for differences in neighborhood type should be recognized. Average income and transport hubs were also significant variables in the investigated crime categories. The article provides a practical demonstration and assessment of the use of high-resolution satellite data to examine the association between urban density and two common types of crime and offers reflections about the use of satellite image data in crime analysis.

This study investigates the efficacy of integrating Remote Sensing (RS) data and Machine Learning (ML) techniques to detect criminogenic environments. The methodology employs ISO clustering to classify land cover from high-resolution aerial images, with subsequent linkage of clustered values to crime and non-crime events. Additionally, the study incorporates various socio-economic, demographic, and land-use variables that influence street crime patterns. By utilising Random Forest classification, the research aims to predict the likelihood of street theft crimes based on environmental characteristics, achieving a high accuracy of approximately 90% as measured by the F1 score. To ensure the robustness of the model, K-fold cross-validation is employed to assess its stability. SHAP is also deployed to understand the importance of variables in prediction modelling. PAI and PEI metrics are used to evaluate the accuracy and efficiency of the risk map. The study's findings culminated in successfully generating a risk prediction map for the city of Stockholm, effectively showcasing the potential of aerial imagery in crime analysis and spatial prediction. Integrating RS data and ML techniques offers a promising avenue for accurately identifying criminogenic environments, with implications extending to urban planning and public safety initiatives. Ultimately, this research underscores the value of interdisciplinary approaches in leveraging advanced technologies to address complex societal challenges such as crime prevention and urban management.

Background

This study explores the potential of predicting crime patterns in Stockholm by integrating environmental features from remote sensing data, weather conditions, socioeconomic factors, and urban structure characteristics. The aim is to develop a predictive model for three types of crime—outdoor rape, residential burglary, and robbery—using a spatio-temporal framework that captures variations in time of day, day of the week, and seasonal changes.

Methods

Historical crime data from 2019 to 2023 are combined with environmental data, including high-resolution aerial imagery and night-time light radiance values, along with socioeconomic and land use factors. Weather conditions are also included as key variables. A Random Forest classifier is applied to different combinations of these datasets, and model performance is evaluated based on temporal factors. SHAP analysis is used to identify the most influential predictors of crime occurrence.

Results

The results show that models accounting for seasonal variation, particularly when incorporating weather variables, perform better than other temporal divisions. Environmental factors, such as night-time lighting and urban structure, emerge as critical for predicting night-time crimes.

Conclusions

The findings suggest that combining diverse data sources can improve the accuracy of crime prediction models, supporting more effective crime prevention strategies. However, challenges such as the resolution limitations of night-time light data and the need for more contextual factors persist. Future research should focus on enhancing model accuracy by incorporating additional real-time data and improving temporal resolution to create safer urban environments.