Measuring wall albedo: in situ and using street view images

With climate change heat waves are becoming more frequent and therefore heat stress in especially urban areas increases. To locate heat spots in cities high-resolution urban atmospheric modelling is key, but the variability of one important factor influencing the temperature is often ignored. The albedo of walls is by default assumed to be constant in most urban schemes used in simulation models such as the Weather Research and Forecasting (WRF) model, while the amount of reflection from different surfaces can have a significant effect on the local temperature. Estimating or measuring the albedo of every street facet in an entire city can be very time-consuming. Although there have been studies where albedo was derived from airborne measurements, these have failed to capture the albedo of walls due to a limited view from above. In this study the use of crowdsourced street view images (SVI) from Mapillary is explored to determine if these images could be used to obtain estimates of wall albedo.

This study builds upon previous work of the Urban-M4 project, which is a project focused on developing a python package for segmenting and analysing street view images for urban modelling applications. This code was used in this study to segment the street view images using a combination of the GroundingDINO model for object detection and the Segment Anything Model (SAM) for instance segmentation to select the building façades. Afterwards, the relative brightness of the selected pixels was calculated in three different ways. The first method used the Lightness component from the HSL model, the second the Value from the HSV model and the third used the luminance of the building in comparison to the brightest pixel in the image. To determine the accuracy of each calculation-method, field measurements were done in Amsterdam and Wageningen and first results show that the albedo calculated with the HSV model is closest to the albedo from field measurements.

More information on the Urban-M4 project can be found here: https://research-software-directory.org/projects/urban-m4