Developing software to estimate thermal properties of urban fabric from open-source street view imagery

Urban weather and climate simulations require increasingly accurate urban morphology information. While spatial maps of commonly used parameters such as building height and density are generally available in The Netherlands, information on radiative properties of urban fabric is typically less abundant. In the Urban-M4 project, we are developing software to automatically retrieve open-source street view images and extract thermal properties like albedo and emissivity of roads and facades. 

We build upon the Global Streetscapes dataset [1], which comprises a wealth of metadata on 10 million crowdsourced street view images spanning 688 cities worldwide. On top of this, we are building a Python package called ‘streetscapes’ that can efficiently query and filter this data to obtain relevant subsets. The corresponding images can be downloaded automatically for further analysis. The streetscapes package wraps state-of-the-art segmentation models to reproduce and extend the original segmentation results included in the Global Streetscapes dataset. This allows the user to flexibly label and extract individual facets in each image and apply further processing steps.  

Our primary interest is to extract albedo and emissivity of roof, road and building facets for use in the Weather Research and Forecasting model. To this end, we are exploring various approaches ranging from direct estimates of reflectivity to indirect methods based on material classification and tabulated radiation data for each material. As these methods are more widely applicable, beyond albedo and emissivity, we are developing the streetscapes package as a generic tool that can be used and extended for analyzing various aspects of street view imagery. The codebase is openly available on https://github.com/Urban-M4/streetscapes.

[1]. doi.org/10.1016/j.isprsjprs.2024.06.023