Unpacking the perceived cycling safety of road environment using street view imagery and cycle accident data.

Cycling, as a routine mode of travel, offers significant benefits in promoting health, eliminating emissions, and alleviating traffic congestion. Many cities, including London, have introduced various policies and measures to promote 'active travel' in view of its manifold advantages. Nevertheless, the reality is not as desirable as expected. Existing studies suggest that cyclists' perceptions of cycling safety significantly hinder the broader adoption of cycling. Our study investigates the perceived cycling safety and unpacks the association between the cycling safety level and the road environment, taking London as a case study. First, we proposed novel cycling safety level indicators that incorporate both collision and injury risks, based on which a tri-tiered cycling safety level prediction spanning the entirety of London's road network has been generated with good accuracy. Second, we assessed the road environment by harnessing imagery features of street view reflecting the cyclist's perception of space and combined it with road features of cycle accident sites. Finally, associations between road environment features and cycling safety levels have been explained using SHAP values, leading to tailored policy recommendations. Our research has identified several key factors that contribute to a risky environment for cycling. Among these, the "second road effects," which refers to roads intersecting with the road where the accident occurred, is the most critical to cycling safety levels. This would also support and further contribute to the literature on road safety. Other results related to road greenery, speed limits, etc, are also discussed in detail. In summary, our study offers insights into urban design and transport planning, emphasising the perceived cycling safety of road environment.

Associations of street-view greenspace with Parkinson's disease hospitalizations in an open cohort of elderly US Medicare beneficiaries.

INTRODUCTION: Protective associations of greenspace with Parkinson's disease (PD) have been observed in some studies. Visual exposure to greenspace seems to be important for some of the proposed pathways underlying these associations. However, most studies use overhead-view measures (e.g., satellite imagery, land-classification data) that do not capture street-view greenspace and cannot distinguish between specific greenspace types. We aimed to evaluate associations of street-view greenspace measures with hospitalizations with a PD diagnosis code (PD-involved hospitalization). METHODS: We created an open cohort of about 45.6 million Medicare fee-for-service beneficiaries aged 65 + years living in core based statistical areas (i.e. non-rural areas) in the contiguous US (2007-2016). We obtained 350 million Google Street View images across the US and applied deep learning algorithms to identify percentages of specific greenspace features in each image, including trees, grass, and other green features (i.e., plants, flowers, fields). We assessed yearly average street-view greenspace features for each ZIP code. A Cox-equivalent re-parameterized Poisson model adjusted for potential confounders (i.e. age, race/ethnicity, socioeconomic status) was used to evaluate associations with first PD-involved hospitalization. RESULTS: There were 506,899 first PD-involved hospitalizations over 254,917,192 person-years of follow-up. We found a hazard ratio (95% confidence interval) of 0.96 (0.95, 0.96) per interquartile range (IQR) increase for trees and a HR of 0.97 (0.96, 0.97) per IQR increase for other green features. In contrast, we found a HR of 1.06 (1.04, 1.07) per IQR increase for grass. Associations of trees were generally stronger for low-income (i.e. Medicaid eligible) individuals, Black individuals, and in areas with a lower median household income and a higher population density. CONCLUSION: Increasing exposure to trees and other green features may reduce PD-involved hospitalizations, while increasing exposure to grass may increase hospitalizations. The protective associations may be stronger for marginalized individuals and individuals living in densely populated areas.

Using deep transfer learning and satellite imagery to estimate urban air quality in data-poor regions.

Urban air pollution is a critical public health challenge in low-and-middle-income countries (LMICs). At the same time, LMICs tend to be data-poor, lacking adequate infrastructure to monitor air quality (AQ). As LMICs undergo rapid urbanization, the socio-economic burden of poor AQ will be immense. Here we present a globally scalable two-step deep learning (DL) based approach for AQ estimation in LMIC cities that mitigates the need for extensive AQ infrastructure on the ground. We train a DL model that can map satellite imagery to AQ in high-income countries (HICs) with sufficient ground data, and then adapt the model to learn meaningful AQ estimates in LMIC cities using transfer learning. The trained model can explain up to 54% of the variation in the AQ distribution of the target LMIC city without the need for target labels. The approach is demonstrated for Accra in Ghana, Africa, with AQ patterns learned and adapted from two HIC cities, specifically Los Angeles and New York.

Beyond here and now: Evaluating pollution estimation across space and time from street view images with deep learning.

Advances in computer vision, driven by deep learning, allows for the inference of environmental pollution and its potential sources from images. The spatial and temporal generalisability of image-based pollution models is crucial in their real-world application, but is currently understudied, particularly in low-income countries where infrastructure for measuring the complex patterns of pollution is limited and modelling may therefore provide the most utility. We employed convolutional neural networks (CNNs) for two complementary classification models, in both an end-to-end approach and as an interpretable feature extractor (object detection), to estimate spatially and temporally resolved fine particulate matter (PM2.5) and noise levels in Accra, Ghana. Data used for training the models were from a unique dataset of over 1.6 million images collected over 15 months at 145 representative locations across the city, paired with air and noise measurements. Both end-to-end CNN and object-based approaches surpassed null model benchmarks for predicting PM2.5 and noise at single locations, but performance deteriorated when applied to other locations. Model accuracy diminished when tested on images from locations unseen during training, but improved by sampling a greater number of locations during model training, even if the total quantity of data was reduced. The end-to-end models used characteristics of images associated with atmospheric visibility for predicting PM2.5, and specific objects such as vehicles and people for noise. The results demonstrate the potential and challenges of image-based, spatiotemporal air pollution and noise estimation, and that robust, environmental modelling with images requires integration with traditional sensor networks.

Do poverty and wealth look the same the world over? A comparative study of 12 cities from five high-income countries using street images.

Urbanization and inequalities are two of the major policy themes of our time, intersecting in large cities where social and economic inequalities are particularly pronounced. Large scale street-level images are a source of city-wide visual information and allow for comparative analyses of multiple cities. Computer vision methods based on deep learning applied to street images have been shown to successfully measure inequalities in socioeconomic and environmental features, yet existing work has been within specific geographies and have not looked at how visual environments compare across different cities and countries. In this study, we aim to apply existing methods to understand whether, and to what extent, poor and wealthy groups live in visually similar neighborhoods across cities and countries. We present novel insights on similarity of neighborhoods using street-level images and deep learning methods. We analyzed 7.2 million images from 12 cities in five high-income countries, home to more than 85 million people: Auckland (New Zealand), Sydney (Australia), Toronto and Vancouver (Canada), Atlanta, Boston, Chicago, Los Angeles, New York, San Francisco, and Washington D.C. (United States of America), and London (United Kingdom). Visual features associated with neighborhood disadvantage are more distinct and unique to each city than those associated with affluence. For example, from what is visible from street images, high density poor neighborhoods located near the city center (e.g., in London) are visually distinct from poor suburban neighborhoods characterized by lower density and lower accessibility (e.g., in Atlanta). This suggests that differences between two cities is also driven by historical factors, policies, and local geography. Our results also have implications for image-based measures of inequality in cities especially when trained on data from cities that are visually distinct from target cities. We showed that these are more prone to errors for disadvantaged areas especially when transferring across cities, suggesting more attention needs to be paid to improving methods for capturing heterogeneity in poor environment across cities around the world. Supplementary Information: The online version contains supplementary material available at 10.1140/epjds/s13688-023-00394-6.

Changes in life expectancy and house prices in London from 2002 to 2019: hyper-resolution spatiotemporal analysis of death registration and real estate data.

Background: London has outperformed smaller towns and rural areas in terms of life expectancy increase. Our aim was to investigate life expectancy change at very-small-area level, and its relationship with house prices and their change. Methods: We performed a hyper-resolution spatiotemporal analysis from 2002 to 2019 for 4835 London Lower-layer Super Output Areas (LSOAs). We used population and death counts in a Bayesian hierarchical model to estimate age- and sex-specific death rates for each LSOA, converted to life expectancy at birth using life table methods. We used data from the Land Registry via the real estate website Rightmove (www.rightmove.co.uk), with information on property size, type and land tenure in a hierarchical model to estimate house prices at LSOA level. We used linear regressions to summarise how much life expectancy changed in relation to the combination of house prices in 2002 and their change from 2002 to 2019. We calculated the correlation between change in price and change in sociodemographic characteristics of the resident population of LSOAs and population turnover. Findings: In 134 (2.8%) of London's LSOAs for women and 32 (0.7%) for men, life expectancy may have declined from 2002 to 2019, with a posterior probability of a decline >80% in 41 (0.8%, women) and 14 (0.3%, men) LSOAs. The life expectancy increase in other LSOAs ranged from <2 years in 537 (11.1%) LSOAs for women and 214 (4.4%) for men to >10 years in 220 (4.6%) for women and 211 (4.4%) for men. The 2.5th-97.5th-percentile life expectancy difference across LSOAs increased from 11.1 (10.7-11.5) years in 2002 to 19.1 (18.4-19.7) years for women in 2019, and from 11.6 (11.3-12.0) years to 17.2 (16.7-17.8) years for men. In the 20% (men) and 30% (women) of LSOAs where house prices had been lowest in 2002, mainly in east and outer west London, life expectancy increased only in proportion to the rise in house prices. In contrast, in the 30% (men) and 60% (women) most expensive LSOAs in 2002, life expectancy increased solely independently of price change. Except for the 20% of LSOAs that had been most expensive in 2002, LSOAs with larger house price increases experienced larger growth in their population, especially among people of working ages (30-69 years), had a larger share of households who had not lived there in 2002, and improved their rankings in education, poverty and employment. Interpretation: Large gains in area life expectancy in London occurred either where house prices were already high, or in areas where house prices grew the most. In the latter group, the increases in life expectancy may be driven, in part, by changing population demographics. Funding: Wellcome Trust; UKRI (MRC); Imperial College London; National Institutes of Health Research.

Characterisation of urban environment and activity across space and time using street images and deep learning in Accra.

The urban environment influences human health, safety and wellbeing. Cities in Africa are growing faster than other regions but have limited data to guide urban planning and policies. Our aim was to use smart sensing and analytics to characterise the spatial patterns and temporal dynamics of features of the urban environment relevant for health, liveability, safety and sustainability. We collected a novel dataset of 2.1 million time-lapsed day and night images at 145 representative locations throughout the Metropolis of Accra, Ghana. We manually labelled a subset of 1,250 images for 20 contextually relevant objects and used transfer learning with data augmentation to retrain a convolutional neural network to detect them in the remaining images. We identified 23.5 million instances of these objects including 9.66 million instances of persons (41% of all objects), followed by cars (4.19 million, 18%), umbrellas (3.00 million, 13%), and informally operated minibuses known as tro tros (2.94 million, 13%). People, large vehicles and market-related objects were most common in the commercial core and densely populated informal neighbourhoods, while refuse and animals were most observed in the peripheries. The daily variability of objects was smallest in densely populated settlements and largest in the commercial centre. Our novel data and methodology shows that smart sensing and analytics can inform planning and policy decisions for making cities more liveable, equitable, sustainable and healthy.

Street-view greenspace exposure and objective sleep characteristics among children.

Greenspace may benefit sleep by enhancing physical activity, reducing stress or air pollution exposure. Studies on greenspace and children's sleep are limited, and most use satellite-derived measures that do not capture ground-level exposures that may be important for sleep. We examined associations of street view imagery (SVI)-based greenspace with sleep in Project Viva, a Massachusetts pre-birth cohort. We used deep learning algorithms to derive novel metrics of greenspace (e.g., %trees, %grass) from SVI within 250m of participant residential addresses during 2007-2010 (mid-childhood, mean age 7.9 years) and 2012-2016 (early adolescence, 13.2y) (N = 533). In early adolescence, participants completed >5 days of wrist actigraphy. Sleep duration, efficiency, and time awake after sleep onset (WASO) were derived from actigraph data. We used linear regression to examine cross-sectional and prospective associations of mid-childhood and early adolescence greenspace exposure with early adolescence sleep, adjusting for confounders. We compared associations with satellite-based greenspace (Normalized Difference Vegetation Index, NDVI). In unadjusted models, mid-childhood SVI-based total greenspace and %trees (per interquartile range) were associated with longer sleep duration at early adolescence (9.4 min/day; 95%CI:3.2,15.7; 8.1; 95%CI:1.7,14.6 respectively). However, in fully adjusted models, only the association between %grass at mid-childhood and WASO was observed (4.1; 95%CI:0.2,7.9). No associations were observed between greenspace and sleep efficiency, nor in cross-sectional early adolescence models. The association between greenspace and sleep differed by racial and socioeconomic subgroups. For example, among Black participants, higher NDVI was associated with better sleep, in neighborhoods with low socio-economic status (SES), higher %grass was associated with worse sleep, and in neighborhoods with high SES, higher total greenspace and %grass were associated with better sleep time. SVI metrics may have the potential to identify specific features of greenspace that affect sleep.

What you see is what you breathe? Estimating air pollution spatial variation using street level imagery.

High spatial resolution information on urban air pollution levels is unavailable in many areas globally, partially due to high input data needs of existing estimation approaches. Here we introduce a computer vision method to estimate annual means for air pollution levels from street level images. We used annual mean estimates of NO2 and PM2.5 concentrations from locally calibrated models as labels from London, New York, and Vancouver to allow for compilation of a sufficiently large dataset (~250k images for each city). Our experimental setup is designed to quantify intra and intercity transferability of image-based model estimates. Performances were high and comparable to traditional land-use regression (LUR) and dispersion models when training and testing on images from the same city (R2 values between 0.51 and 0.95 when validated on data from ground monitoring stations). Like LUR models, transferability of models between cities in different geographies is more difficult. Specifically, transferability between the three cities i.e., London, New York, and Vancouver, which have similar pollution source profiles were moderately successful (R2 values between zero and 0.67). Comparatively, performances when transferring models trained on these cities with very different source profiles i.e., Accra in Ghana and Hong Kong were lower (R2 between zero and 0.21) suggesting the need for local calibration with local calibration using additional measurement data from cities that share similar source profiles.

A Deep Learning Approach for Meter-Scale Air Quality Estimation in Urban Environments Using Very High-Spatial-Resolution Satellite Imagery.

High-spatial-resolution air quality (AQ) mapping is important for identifying pollution sources to facilitate local action. Some of the most populated cities in the world are not equipped with the infrastructure required to monitor AQ levels on the ground and must rely on other sources, like satellite derived estimates, to monitor AQ. Current satellite-data-based models provide AQ mapping on a kilometer scale at best. In this study we focus on producing hundred-meter-scale AQ maps for urban environments in developed cities. We examined the feasibility of an image-based object-detection analysis approach using very high-spatial-resolution (2.5 m) commercial satellite imagery. We fed the satellite imagery to a deep neural network (DNN) to learn the association between visual urban features and air pollutants. The developed model, which solely uses satellite imagery, was tested and evaluated using both ground monitoring observations and land-use regression modeled PM2.5 and NO2 concentrations over London, Vancouver (BC), Los Angeles, and New York City. The results demonstrate a low error with a total RMSE < 2 µg/m3 and highlight the contribution of specific urban features, such as green areas and roads, to continuous hundred-meter-scale AQ estimation. This approach offers promise for scaling to global applications in developed and developing urban environments. Further analysis on domain transferability will enable application of a parsimonious model based merely on satellite images to create hundred-meter-scale AQ maps in developing cities, where current and historical ground data is limited.

Multimodal deep learning from satellite and street-level imagery for measuring income, overcrowding, and environmental deprivation in urban areas.

Data collected at large scale and low cost (e.g. satellite and street level imagery) have the potential to substantially improve resolution, spatial coverage, and temporal frequency of measurement of urban inequalities. Multiple types of data from different sources are often available for a given geographic area. Yet, most studies utilize a single type of input data when making measurements due to methodological difficulties in their joint use. We propose two deep learning-based methods for jointly utilizing satellite and street level imagery for measuring urban inequalities. We use London as a case study for three selected outputs, each measured in decile classes: income, overcrowding, and environmental deprivation. We compare the performances of our proposed multimodal models to corresponding unimodal ones using mean absolute error (MAE). First, satellite tiles are appended to street level imagery to enhance predictions at locations where street images are available leading to improvements in accuracy by 20, 10, and 9% in units of decile classes for income, overcrowding, and living environment. The second approach, novel to the best of our knowledge, uses a U-Net architecture to make predictions for all grid cells in a city at high spatial resolution (e.g. for 3 m × 3 m pixels in London in our experiments). It can utilize city wide availability of satellite images as well as more sparse information from street-level images where they are available leading to improvements in accuracy by 6, 10, and 11%. We also show examples of prediction maps from both approaches to visually highlight performance differences.

Measuring social, environmental and health inequalities using deep learning and street imagery.

Cities are home to an increasing majority of the world's population. Currently, it is difficult to track social, economic, environmental and health outcomes in cities with high spatial and temporal resolution, needed to evaluate policies regarding urban inequalities. We applied a deep learning approach to street images for measuring spatial distributions of income, education, unemployment, housing, living environment, health and crime. Our model predicts different outcomes directly from raw images without extracting intermediate user-defined features. To evaluate the performance of the approach, we first trained neural networks on a subset of images from London using ground truth data at high spatial resolution from official statistics. We then compared how trained networks separated the best-off from worst-off deciles for different outcomes in images not used in training. The best performance was achieved for quality of the living environment and mean income. Allocation was least successful for crime and self-reported health (but not objectively measured health). We also evaluated how networks trained in London predict outcomes three other major cities in the UK: Birmingham, Manchester, and Leeds. The transferability analysis showed that networks trained in London, fine-tuned with only 1% of images in other cities, achieved performances similar to ones from trained on data from target cities themselves. Our findings demonstrate that street imagery has the potential complement traditional survey-based and administrative data sources for high-resolution urban surveillance to measure inequalities and monitor the impacts of policies that aim to address them.

A hazard-based approach to modelling the effects of online shopping on intershopping duration.

Despite growing prevalence of online shopping, its impacts on mobility are poorly understood. This partially results from the lack of sufficiently detailed data. In this paper we address this gap using consumer panel data, a new dataset for this context. We analyse one year long longitudinal grocery shopping purchase data from London shoppers to investigate the effects of online shopping on overall shopping activity patterns and personal trips. We characterise the temporal structure of shopping demand by means of the duration between shopping episodes using hazard-based duration models. These models have been used to study inter-shopping spells for traditional shopping in the literature, however effects of online shopping were not considered. Here, we differentiate between shopping events and shopping trips. The former refers to all types of shopping activity including both online and in-store, while the latter is restricted to physical shopping trips. Separate models were estimated for each and results suggest potential substitution effects between online and in-store in the context of grocery shopping. We find that having shopped online since the last shopping trip significantly reduces the likelihood of a physical shopping trip. We do not observe the same effect for inter-event durations. Hence, shopping online does not have a significant effect on overall shopping activity frequency, yet affects shopping trip rates. This is a key finding and suggests potential substitution between online shopping and physical trips to the store. Additional insights on which factors, including basket size and demographics, affect inter-shopping durations are also drawn.