Modelos estadísticos para la detección de patrones en medio ambiente y economía : aplicación de técnicas bootstrap para establecer umbrales más exigentes al momento de realizar la detección de inhomogeneidades en series meteorológicas.

Abstract

The inhomogeneities in the time series can unleash extreme events which cause a negative impact in the field of study. In meteorology, these disturbances can represent meteorological events such as: heat waves, droughts, floods, strong winds, and even forest fires, and they originate due to errors in the recording of information or the gradual change of the seasons. These inhomogeneities must be detected and corrected from the data set through the homogenization process. With the R Climatol package, this entire process can be carried out using R code, but there is a problem when choosing the inhomogeneity correction threshold. The choice of the threshold is made subjectively, only observing the graphs of the maximum SNHT histograms, but it is not the most appropriate since there is a high probability that certain inhomogeneities in the series will not be detected. For this reason, the project focuses on applying Moving Block Bootstrap (MBB) and Stationary Bootstrap (SB) resampling techniques to obtain more demanding thresholds with 95\% confidence, specifying in a very detailed way the structure and methodology to be applied. used to accomplish this goal. The application was made to a data set with daily measurements of the Relative Humidity variable of 9 meteorological stations that the Alternative Energy and Environment Group (GEAA) monitors in the period 2015 to 2017. The most reliable results that allow minimizing the value of the root mean square error (RMSE) and the value of the standard normal homogeneity test (SNHT) in the homogeneous series were achieved with the threshold obtained with SB resamples and are presented in the Results section.