FILLING OF FAULTS IN CLIMATOLOGICAL AIR TEMPERATURE SERIES IN BRAZILIAN STATE CAPITALS FROM 1980 TO 2017

Resumen

Climatological data and the information obtained from these have great relevance for different activities performed by man, given the results contribute to decision-making in areas such as water management, urban climate and others. Due to their importance, the climatological series can be filled by different techniques according to the interest of the researcher and what is most appropriate for the desired analysis. Thus, this study aimed to evaluate three methodologies in the climatological series of maximum and minimum temperature in 21 meteorological stations located in different state capitals of Brazil for a period of 37 years (1980-2017). For this purpose, multiple linear regression statistical technique, computational technique using artificial intelligence in a artificial neural network of the perceptron multilayer and the self-regressive integrated moving average model were applied. In order to verify the performance of the models used in filling failures, the mean error or bias, absolute mean error and coefficient of determination were used. The results obtained for the maximum and minimum mean temperatures indicated the methods of neural networks and multiple linear regression as appropriate techniques in the process of estimating missing data. Therefore, these results contribute to the development of studies that require climatological time series and even the improvement of the techniques presented here.