Perceptible landscape patterns reveal invisible socioeconomic profiles of cities

Background:

This study is led by Dr. Wenning Li and Prof. Ranhao Sun (Research Center for Ecological and Environmental Sciences, Chinese Academy of Sciences). Urbanization is transforming our living environments and economic structures at an unprecedented rate. The assessment of urban socioeconomic development has relied on "invisible" statistical indicators. But the visible and tangible urban landscape, such as architectural style, road layout, and the distribution of parks and green spaces, offers a direct reflection of an urban character and progress. Urban landscape not only affects residents' lives, but also reflects the level of urban development. Traditional methods quantify urban landscape patterns using satellite remote sensing technology, focusing primarily on two-dimensional features while overlooking the complex information of building facades and vegetation, thereby providing a perspective that is substantially removed from that of urban residents. Consequently, these methods may fail to accurately represent residents' perceptions of the urban landscape. With these limitations, this study provides an in-depth interpretation of socioeconomic development through urban street views. Confronted with massive amount street view images data, the technical challenges faced by the study include how to efficiently identify landscape elements in street view images and how to construct scientifically and effective landscape indicators. By solving these issues, this study aims to provide a more robust scientific basis and technical support for urban planning and development decisions, leveraging comprehensive insights.

Results:

In this study, a substantial corpus of urban street view big data was initially acquired. Researchers utilized nighttime light data to delineate the boundaries of 303 cities in China.

Employing the Baidu Street View map open interface, over 4 million street view images were acquired from four direction viewpoints (0°, 90°, 180°, and 270°) at each sampling point. Critical urban landscape elements were extracted using deep learning algorithms on this base. By comparing the accuracy and computational efficiency of multiple deep learning models, DeepLabv3 pre-training model was chosen to complete the segmentation and identification of key landscape elements in street view images, including 19 types of elemental objects such as roads, buildings, vegetation, pedestrians, and cars, which solves the problem of efficiently identifying landscape elements using massive street view data.

To quantitatively capture the characteristics of urban landscapes, the study introduced a hierarchical semantic tree approach, proposing four urban landscapes indicators in three dimensions: greenness, greyness, openness, and crowding. These four indicators represent four aspects of urban areas, namely, the ecological environment, urban development, urban morphology, and anthropogenic activities. Collectively, these indicators comprehensively depict the social and ecological attributes of urban areas, bridging the gap between qualitative interpretation of landscape elements and the quantitative assessment of landscape patterns.

Incorporating factors such as urban geographical location, topographic features, and climatic context, the study established a socioeconomic multiple linear regression model based on urban landscape patterns. For the four predictive models addressing land, population, economy, and society, the inclusion of urban landscape indicators substantially enhanced the explanatory power of the regression models. Compared to traditional models based on remotely sensed indicators, the precision of these new models improved by 19% to 36%. The landscape pattern prediction model proposed by this study allows for more accurate and real-time forecasting of socioeconomic development levels. This tool helps urban planners in gaining a timely understanding of the social and ecological progress of cities, thereby promoting sustainable urban development.

See the article:

Perceptible landscape patterns reveal invisible socioeconomic profiles of cities https://doi.org/10.1016/j.scib.2024.06.022