Métodos de inteligencia artificial para la imputación de precipitación pluvial de la cuenca río Ravelo

Abstract

In the field of hydrology, AI has had a significant intervention. The objec
t is: To evaluate the capacity of different artificial
intelligence algorithms for the imputation of precipitation data in the Ravelo River basin. More than 300,000 rainfall data
measured every 15 minutes from 2019 to 2023 were obtained from the Ravelo River basin at the Ravelo, Tumpeka and Cajamarca
stations. To determine the variables used, the Pearson correlation was applied, obtaining the environmental variables to predict rainfall: Barometric Pressure, Relative Humidity, Wind Speed and Wind Direction. The data set for training was defined, from April 2021 to March 2022. There are 140245 readings, Ravelo: 47331 readings, Tumpeka: 46455 readings and Cajamarca: 46459 eadings.
The test data set was defined, from May 2022 (dry season) and December 2022 (rainy season). There are 17,286 readings, Ravelo: 5,762 readings, Tumpeka: 5,762 readings and Cajamarca: 5,762 readings. The methods used are: Linear Regression, Decision Trees
(regressor, classifier) and Logistic Regression. It was evident that all the methods used are capable of predicting rainfall and herefore can be used in its imputation. It is observed that the most convenient methods for imputation are linear regression and the decision tree for classification, given that, according to the metrics obtained, it has the best performance in prediction.