Modelos espaciais e espaço-temporais para contagem de chuvas extremas na costa leste do Nordeste do Brasil

Abstract

The study of climatic extremes is essential for understanding the occurrence of natural disasters. It becomes even more important when the tools available for this analysis do not perform their functions accurately. This is the case with the extreme rainfall events that occur on the east coast of north-eastern Brazil, where estimates from the Tropical Rainfall Measuring Mission satellite (TRMM) and the Tropical Rainfall Measuring Mission satellite (TRMM) are not accurate. (TRMM) satellite and the Global Precipitation Measurement (GPM) satellite tend to underestimate extreme precipitation. This study therefore set out to analyse the performance of spatial and spatio-temporal models, available in the literature and implemented in the R language, when applied to extreme rainfall count data from the east coast of the Northeast, as well as comparing the performance of the models with the performance of the satellite for the study context. The aim is to build up the knowledge needed to support the development of mitigation measures for natural disasters. To this end, the study variable was the number of times the daily rainfall accumulated exceeded or was equal to the threshold of 30 millimetres over the course of a year, so that the counts were made per rainfall station. This variable is one of the 27 indices adopted by the World Meteorological Organisation (WMO) to study climate variations. The time series studied has records of daily rainfall accumulations from 36 stations during the period from 1991 to 2022. In addition to precipitation, it contains the latitude, longitude and altitude of the stations. The data was taken from the National Meteorological Institute (INMET) and the National Water and Sanitation Agency (ANA). With regard to the methodology adopted for the main parts of the study, the research began with a literature search for R language packages that worked with count data from a spatial and spatio-temporal perspective. After the search and before using the models, a descriptive analysis of the data was carried out, in which the means, medians, standard deviations, coefficients of variation and amplitudes by year and season were calculated. The investigation continued by analysing the study variable and the covariates latitude, longitude and altitude using scatter plots. Spatial and temporal dependencies were also analysed using variograms. Once the descriptive analysis had been completed, the cross-validation methodology was used to assess the performance of the models. The comparison between the models and the satellite data was made using mean square error, bias and correlation coefficient statistics. The results achieved in this research are encouraging. Extremes occur more frequently in regions close to the coast, with spatial dependence and temporal dependence for only a few stations. Furthermore, in a purely observational analysis, there is an indication of patterns of extremes repeating every 10 years from 1991 onwards. It was also possible to see that satellite data underestimates the extremes in the region studied, especially in the south of the area. Four packages implemented in the R language capable of working with the study data were identified in the literature, and six models from these packages were adjusted. The models obtained excellent results when compared to the satellite estimates. They proved to be better in most of the locations tested. They also show that they have great potential for generating more reliable data for analysing precipitation extremes.