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

Abstract

The objective of this research was to identify behavior patterns of precipitation for the 2020-2050 period through statistical tools, machine learning techniques and model validation using observed and predicted data. The precipitation database was obtained from the atmospheric models of the National Center for Meteorological Research (CNRM), which is in the project Intercomparisons of Coupled Models Phase 6 (CMIP6), the precipitation data was collected from the period 2020-2050 on a daily scale, for this data analysis the Python programming language was used to elaborate frequency histograms and multivariate techniques for dimensionality reduction; different Machine Learning techniques were applied to identify behavior patterns such as: K-Means, Density-based Spatial Clustering of Applications with Noise (DBSCAN) and Agglomerative Grouping, where the technique (DBSCAN) identified 4 clusters of behavior patterns, in order to validate said model, the Silhouette precision metric and the precipitation information recorded in Ecuador for the year 2021 were used. As a result of this analysis, the best technique for the reduction of dimensionality and identification of patterns of precipitation behavior was the algorithm with Euclidean metrics (DBSCAN), with an effectiveness of 70.92% greater than the other techniques used, the behavior patterns identified with the aforementioned technique were similar to the predicted data. It is recommended to experiment with various machine learning techniques to see which of these techniques is appropriate for the database under study. It is important to go to an expert in the area of climatology for a better interpretation and understanding of weather events.