Aplicação de aprendizagem de máquina para o mapeamento de ilha de calor urbano superficial e predição da temperatura de superfície terrestre a partir de variáveis ambientais e socioeconômicas

Abstract

The landscape feature is responsible for changing urban dynamics, causing the formation of Urban Heat Islands (UHI). The Land surface Temperature (LST) extracted from thermal images is a primary source of information for exploiting UHI, characterizing the Surface Urban Heat Islands (SUHI). In addition to LST, a set of environmental and socioeconomic variables has been adopted to explain the SUHI phenomenon. Although machine learning algorithms have demonstrated great potential in several areas of applications, their use to investigate the relationship between socioeconomic and environmental variables and SUHI is still unknown. This work proposes the characterization of SUHI using socioeconomic and environmental variables from machine learning approach. The LST was extracted from 15 images collected by the TIRS sensor, Landsat 8 system, from 2019 to 2021. Data on socioeconomic variables were obtained from the official demographic census. The environmental variables, described by spectral indices of vegetation (NDVI), built-up area (NDBI) and impervious surfaces (ISA), were extracted from Sentinel-2 and Planet images. Machine learning algorithms were tested to evaluate the ability to estimate LST based on the aforementioned variables. The algorithms used in the study were decision tree (DT), k‑nearest neighbor (KNN), linear regression (LR), multilayer perceptron (MLP), support vector regression (SVR) and random forest (RF). The results showed that DT algorithm obtained the best performance (r= 0.96, MAE= 1.49 °C and RMSE= 1.88 °C), followed by the RF. Furthermore, the inclusion of all seasons and socioeconomic variables proved to be relevant in the results. The main contribution of this work is to verify if the algorithms can optimize the SUHI characterization process, analyzing the influence exerted by the studied variables. In the social sphere, the information produced can help urban planning in the construction of smart cities.