High spatiotemporal granularity prediction of Heat Island within the City Using High-Frequency Urban Sensor Network

Keywords: Urban Sensors Network, Remote Sensing, CyberGIS, high-frequency data
Session Type: Virtual Paper Abstract
Day: Monday
Session Start / End Time: 2/28/2022 02:00 PM (Eastern Time (US & Canada)) - 2/28/2022 03:20 PM (Eastern Time (US & Canada))
Room: Virtual 48

Abstract

As rapid urbanization proceeds, Heat Island within the cities, featuring significantly higher temperatures in metropolitan areas compared to surrounding areas, has caused a negative effect on people living in the cities. Temporal granularity is often ignored in most Urban Heat Island studies due to the use of weekly- or biweekly- based thermal remote sensing imageries. The low-temporal frequency data has also held back the application of complicated models onto the prediction of heat islands within the city. To resolve this limitation, this study developed approaches encompassing traditional polynomial regression model, Artificial Neural Network, Support Vector Machine, and Random Forest Regression for predicting the heat island within the city with high-frequency urban sensing data combined with remote sensing data in Chicago, IL from 2018 to 2020. Enabled by rapid advances in sensors technologies and high-performance computing, we build a comprehensive framework to use deep learning methods to predict the heat island within the city based on environmental data collected with urban sensors network and Landsat remote sensing imageries including PM2.5 concentration, Sulfur dioxide concentration, humidity, light intensity, land property, etc. After comparing the heat island detected with observed data from the urban sensor and the predicted heat island with a deep learning model, it is demonstrated that deep learning models, especially the random forest regression model, can be used effectively in accurately predicting the high-spatiotemporal granularity heat island with high-frequency urban sensing data integrated with remote sensing data.