Scholarlink

<p>Machine learning algorithms have gained traction across diverse fields, demonstrating significant success in numerous applications, including precipitation climate prediction. As in all world regions, precipitation is a key meteorological variable in South America, which influences the intensity of rainfall across varying climate seasons. Its importance extends beyond agriculture, civil defense, and tourism in Brazil, as a substantial portion of the country’s electricity generation relies on hydroelectric power plants. This research leverages meteorological variables, including wind fields, temperature, and the previous month’s precipitation, as inputs for monthly precipitation prediction. The Light Gradient Boosting Machine (LightGBM) framework is employed by using the decision tree learning approach. Two gradient boosted decision tree models are designed: one for precipitation estimation and another for uncertainty prediction. Optimal hyperparameters for the machine learning tools are determined through the <i>Optuna</i> optimizer. Data ranging from January 1980 to December 2017 are used for algorithm training, while the years 2018 and 2019 serve as the testing period. The results show a good performance of the proposed methodology for precipitation prediction. Predictability maps for (prediction uncertainty quantification) covering the entire South American territory were carried out as a time series prediction for the computed variance between the difference of observation and prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning algorithm for uncertainty quantification in climate precipitation prediction over South America

  • 期刊论文

  • 发表日期：

    2025年10月15日

查看全文
  • Vinicius Schmidt Monego,
  • Juliana Aparecida Anochi,
  • Haroldo Fraga de Campos Velho

摘要

Machine learning algorithms have gained traction across diverse fields, demonstrating significant success in numerous applications, including precipitation climate prediction. As in all world regions, precipitation is a key meteorological variable in South America, which influences the intensity of rainfall across varying climate seasons. Its importance extends beyond agriculture, civil defense, and tourism in Brazil, as a substantial portion of the country’s electricity generation relies on hydroelectric power plants. This research leverages meteorological variables, including wind fields, temperature, and the previous month’s precipitation, as inputs for monthly precipitation prediction. The Light Gradient Boosting Machine (LightGBM) framework is employed by using the decision tree learning approach. Two gradient boosted decision tree models are designed: one for precipitation estimation and another for uncertainty prediction. Optimal hyperparameters for the machine learning tools are determined through the Optuna optimizer. Data ranging from January 1980 to December 2017 are used for algorithm training, while the years 2018 and 2019 serve as the testing period. The results show a good performance of the proposed methodology for precipitation prediction. Predictability maps for (prediction uncertainty quantification) covering the entire South American territory were carried out as a time series prediction for the computed variance between the difference of observation and prediction.