Análise de mudanças climáticas no estado do Maranhão: estudo das temperaturas e radiações solares

Abstract

Climate events that represent risks to communities in general are becoming more frequent and arise due to the daily, seasonal and year-to-year variability of the climate as well as from regional climatic differences. The Science of Climate Change studies models in order to ensure prediction of effects and impacts with precision. This work aimed to evaluate the behavior of climatic variables, air temperature and solar irradiation, in five municipalities located in the state of Maranhão and to present a long-term forecast model (one year ahead) of these variables using the Artificial Neural Networks intelligent model. Samples of five automatic meteorological stations were collected from INMET for the historical series from 2008 to 2019 and, to reach a consistent database, the processes of data cleaning and feature selection were performed to assemble the information database for the chosen ANN model. The databases of each municipality were subdivided by seasonal period so that the temporal forecast for the same period could be carried out one year ahead. The metrics for evaluating the forecasts were: Mean Absolute Error (MAE), Absolute Relative Error (RAE) and Root Mean Square Error (RMSE). The climatic variables used for the development of the method were Maximum Temperature (ºC), Relative Air Humidity (%), Minimum Temperature (ºC), Dew Point Temperature (ºC), Wind Speed (m/s), Maximum Gust Wind (m/s), Wind Direction (degrees), Precipitation (mm) and Atmospheric Pressure of the Station (mBar) in addition to the variables Solar Irradiation (kJ/m²) and Air Temperature (ºC) that were being predicted. The collected data were divided into 80% for training and 20% for testing. The results of the Solar Irradiation forecast models were MAE = 0.062 kJ/m², RAE = 0.00005 kJ/m² and RMSE = 0.0002 kJ/m². Values for Air Temperature forecast metrics are MAE = 0.4737 ºC, RAE = 0.0003 ºC and RMSE = 0.0025 ºC. The most consistent bases for the calculation of MAE and RAE were the winter of the municipalities of Imperatriz and Turiaçu, while for the RMSE, the best results were obtained using the spring period database of the municipalities of Caxias and Imperatriz.