Predicción de patrones de comportamiento de la temperatura mediante herramientas estadísticas usadas en los modelos de circulación atmosférica 2020-2050

Abstract

The graduation work had the objective of predicting the behavior patterns of the temperature, where all the provinces of Ecuador were taken as the study population, there are few analysis dedicated to predicting behavior patterns, which is why the information obtained from the future atmospheric circulation models of the CHIRPS data bank was studied, this information had to be transformed into a two-dimensional matrix order to work with it, since it was in the form of a multidimensional matrix. When developing an exploratory analysis, an increasing trend was found that was not very pronounced yet clear, where it is evident that with the passage of time the temperature will have an increase, in addition its data come from a normal distribution, after applying several dimensionality reduction algorithms. As a result, it was found that the UMAP algorithm is better to find a two dimensional representation of this matrix with several columns in the present research, helping to preserve its multidimensional topological structure, likewise the Chebyshev distance was used to obtain better results, where the algorithm applied to make the k-means groupings was the best to group this information, with a Silhouette index of 0.72 where a clear separation of 10 or 15 years in the groups was evidenced, when comparing this range of years of four categories with the four sets that groups k-means, a 90% accuracy was found, so it was concluded with the prediction that these groups belong to high, very high, low and normal temperatures, as analyzed in the heat maps, line graphs and frequency histograms. It is recommended to validate each of the techniques to have effective models that allow decision making in the area of climatology.