Filling and validating rainfall data based on statistical techniques and artificial intelligence

Abstract

The study of the hydric regime of rainfall helps in management analysis and decision-making in hydrographic basins, but a fundamental condition is the need for continuous time series of data. Therefore, this study compared gap filling methods in precipitation data and validated them using robust statistical techniques. Precipitation data from the municipality of Itirapina, which has four monitoring stations, were used. Four gap filling techniques were used, namely: normal ratio method, inverse distance weighting, multiple regression and artificial neural networks, in the period from 1979 to 1989. For validation and performance evaluation, the coefficient of determination (R²), mean absolute error (MAE), mean squared error (RMSE), Nash-Sutcliffe coefficient (Nash), agreement index (D), confidence index were used (C) and through non-parametric techniques with Mann-Witney and Kruskal-Wallis test. Excellent performances of real data were verified in comparison with estimated data, with values above 0.8 of the coefficient of determination (R²) and of Nash. Kruskal-Wallis and Mann-Whitney tests were not significant in Stations C1 and C2, demonstrating that there is a difference between real and estimated data and between the proposed methods. It was concluded that the multiple regression and neural network methods showed the best performance. From this study, efficient tools were found to fill the gap, thus promoting better management and operation of water resources.
 Keywords: artificial neural networks, inverse distance weighting, multiple regression, normal ratio method.

The hunt for data: obstacles faced by researchers in the search for accurate information on precipitation in Brazil

“…They can also enable the incorporation of climate change in water resource planning processes, along with other sectoral policies, taking into account possible impacts that changes in rainy season patterns can cause at local and global levels. 9 Studies by Brubacher et al 10 state that despite faults in extensive series harming studies related to urban and rural planning, to monitor extreme events that may impact society and assist in urban drainage projects aimed at reducing risks inherent to flooding and inundation, and engineering works such as dam sizing, methods for filling gaps such as weighting from Simple or Multiple Linear Regression, mathematical models based on machine learning, such as Artificial Neural Networks, spatial interpolators for gap filling (Inverse Distance, Natural Neighbor, Kriging), can help researchers mitigate these faults and contribute to the correction of missing data.…”