Modelagem geográfica e máxima entropia para contextualização e predição da localização de armazéns em regiões metropolitanas

Abstract

Urban freight transport (UFT) has a direct impact on our daily lives. It is responsible for supplying from the most essential to the most complex activities. However, this activity comes with a series of negative externalities such as the increase of congestion, CO2 emissions, noise and vibration levels in the urban environment. Given the importance of this activity in people's daily lives, we must expect that UFT should be considered in the agenda of public managers. However, this is not the reality. The UFT has little or no mention in the Municipal Master Plans, which demonstrates the unawareness of the phenomenon, and the need to investigate it. Investigating the warehouse location in the Metropolitan Region of Belo Horizonte (BHMR), this dissertation aims to analyze how factors used in urban planning can contribute to the installation of logistics warehouses and thus identify the potential places where warehouses should be located to assist in the introduction of urban cargo transport in urban planning. For this purpose, this work used the Maximum Entropy Modeling (MEM) to obtain the cartographic representation of the potential locations where the warehouses have the potential to be located in the RMBH. It is a probabilistic statistical machine learning model that uses the concepts of entropy to develop its mathematical formulation and build this model based on the points of the geographical location of the phenomenon and factors that explain the phenomenon, represented in a specialized way. Its results, in addition to the cartographic representation with the probability of localization of warehouses in the BHMR territory, provided responsive curves for all factors used in the modeling, with their marginal effect for the factor analyzed in isolation and in the context of the model, which together with the jackknife bar graphs, it was possible to understand how the factors influence the location of the warehouses. Eight different models were developed to better assess the different interactions between the factors. The models aimed to prove the following hypotheses: (i) the logistics warehouses tend to form agglomerations in a metropolitan territory; (ii) logistics warehouses tend to be located close to the consumer market; (iii) logistics warehouses tend to be located close to shippers; (iv) logistics warehouses tend to be located close to the local transport infrastructure; (v) logistics warehouses tend to be located close to the regional transport infrastructure; (vi) logistics warehouses tend to be located in urban areas. The six hypotheses were proven, and the models were validated according to their ROC curve (Receiver Operator Characteristic) and the value of AUC (Area Under the Curve), and all models could be considered valid.