Modelagem e análise de cenários através de classificação do índice de variância do campo termal urbano em ilhas de calor urbano em Cuiabá utilizando floresta randômica

Abstract

Urban Heat Islands (UHI) have in their context a very peculiar complexity in their formation that is strongly influenced by human action. Heat retention, unnatural barriers that prevent winds from following the natural path, changes in geomorphology cause huge impacts on local vegetation, society in relation to health and economy. Provide a way of understanding the occurrence of the Urban Heat Island Phenomenon (IHU) and how it impacts microclimatic variables: Soil Temperature, Air Temperature, Direction and Speed of Local Winds, Relative Air Humidity and the Normalized Difference Index of Vegetation (NDVI) is the objective of this study that uses a data analysis approach through Random Forest (Random Forest – RF) with the objective of evaluating and analyzing the influence of the Urban Thermal Field Variance Index (UTFVI) on the variables microclimatic variables and the NDVI spectral index and seek to classify what trends may occur given the circumstances and behavior of these microclimatic variables as well as the NDVI. To achieve the objectives, the model is trained with data from the Landsat 7 and 8 satellite to classify images and obtain the predominance of land cover and ERA5 reanalysis data that provides information on microclimatic variables. Initially, a spatial analysis of the region of interest was carried out using the Google Earth Engine software in order to obtain the spatialization of the image as well as the temporality of the sample data and thus be able to obtain spectral indices related to the phenomenon of island formation. of urban heat, as well as microclimatic data. During the experiment, the model was generated with a data sample reduced to UTFVI indication levels to determine the occurrence of the UHI Phenomenon (strong, very strong and extreme) to balance the rate of false negatives and false positives and thus be able to generate a more accurate predictive model. Obtaining the results demonstrate the importance of machine learning to improve the detection of factors that lead to dangerous conditions in the existence of the UHI Phenomenon and consequent assistance in mitigating the risks inherent to Urban Heat Islands. This work contributes towards identifying which locations in the area under study tend to have a greater intensity in the UHI Phenomenon due to microclimatic variables, or even which locations are likely to have low intensity in the UHI Phenomenon and, consequently, provide public managers with a better understanding of how to solve the problems caused by the formation of urban heat islands.