A Machine Learning Approach to Monitoring the UHI From GNSS Data

Abstract

The urban heat island (UHI) effect has been widely studied because of its impacts on the environment and human’s wellbeing. The surface UHI (SUHI) is typically monitored with remote sensing products, whereas the air UHI (AUHI) is typically monitored using temperature sensor networks. The first technique requires clear days to obtain good images, and the latter is limited by the number of weather stations in an urban and rural area. An alternative to monitor the AUHI is the usage of global navigation satellite system (GNSS) data which is available in all weather conditions and there are well-developed GNSS receiver networks around the world. In this article, we propose a machine learning (ML) approach to estimating the air temperature from GNSS data to take advantage of the quantity of data. By training a support vector regressor with three datasets, GNSS-derived zenith tropospheric delay (ZTD), temperature, and day-of-the year, we can estimate temperature from GNSS data without the need for other environmental variables. The regressor presented in this article is trained with data from 2016 to 2019 and tested with data from 2020 in five Japanese metropolitan areas. The results show that it is possible to predict temperature from only the ZTD of GNSS signals with high precision, as an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula> between 0.82 and 0.90 is found for the predictions. Predicted temperatures are then used to study the diurnal cycle of the AUHI in Sapporo, Tokyo, Nagoya, Osaka–Kyoto, and Fukuoka with an average precision of 0.8 °C.