Evaluación de metodologías para el pronóstico de series de precipitación en la cuenca del rio Sogamoso en el departamento de Santander Colombia

Abstract

The evaluation of different precipitation forecasting models is presented in the Sogamoso River in Santander, Colombia. From the study of each of the selected methodologies, a compilation of the data of the meteorological stations of the study area was carried out, for the identification of the variables that were used in each of the selected models. They were developed according to the data found by each of the models, the consistency analysis of the information in which the respective studies of the behavior of each model were carried out according to their initial and ideal conditions.

For the development of the project, combinations of meteorological stations and data corresponding to the variables (precipitation, relative humidity, evapotranspiration, mean temperature and solar brightness) were established, which were analyzed according to the needs of each of the models, with these combinations and the results obtained according to each model, was answered which is the most suitable for the study area. The models that were analyzed in this project were simple linear regression, multiple linear regression, generalized linear models, principal component analysis and Kalman filter, each one required different data, since the characteristics of each are different.

With the combinations of each of the models, the process of analyzing compliance with the criteria to be evaluated was performed, and for this, 70% of the total data from each of the stations was used for the simulation and with the remaining 30% the forecast of each one. The five proposed models were developed from algorithms and series of variables depending on the criteria of each, based on the results of the forecasts developed in the models, the evaluation of the methodologies was carried out applying the performance metrics.

Based on the performance of each of the models, their results, performance metrics and other characteristics, it was determined which model fits the data of the stations and the requirements of the model in which it was executed, which models were identified. more errors and they are disposable for the area of the basin. As a result, it was found that, according to the models made and analyzed, together with the Gamma family, the model whose best results were obtained are the generalized linear models (MLG).