Redes neurais LSTM na modelagem de poluentes atmosféricos em região urbana agrícola

Abstract

Atmospheric pollution represents a global challenge with severe impacts on public health and the environment, especially in growing urban areas like Francisco Beltrão in Paraná. This study evaluated the effectiveness of Artificial Neural Networks, specifically the Long Short-Term Memory (LSTM) model, in predicting local concentrations of atmospheric pollutants such as formaldehyde (HCHO), nitrogen dioxide (NO2), and sulfur dioxide (SO2), by integrating NASA satellite data with local meteorological variables. The LSTM model demonstrated the ability to capture the complex temporal and nonlinear relationships within the data, resulting in promising predictions for the concentrations of these pollutants, especially for NO concentration, with a mean squared error (MSE) de 3,7885 × 10−10. Analysis of the results revealed the significant influence of meteorological conditions on the dynamics of atmospheric pollution, highlighting the importance of considering them in predictive models particularly the impact of temperature on the concentration of HCHO. Despite limitations such as the lack of monitoring data in the city center, this study provides an embryonic basis for the development of effective tools for monitoring and managing air quality in Francisco Beltrão, with potential applications in public health, urban planning, and environmental policies. These promising results pave the way for the development of air quality alert and forecasting systems, aiding decision-making and the implementation of effective public policies for managing atmospheric pollution in this municipality.