Scholarlink

<p>Accurate weather forecasting is crucial for sectors such as public health and agriculture, playing a key role in decision-making during extreme weather events. Traditional numerical weather forecasting (NWF) methods rely on solving complex physical equations, demanding extensive computational resources. Recently, machine learning techniques have emerged as cost-effective alternatives, leveraging technological advancements to improve prediction accuracy. This study compares the performance of five models—SARIMA, RNN, LSTM, XGBoost, and LightGBM—in predicting next day’s maximum and minimum temperatures using 10 years of data from Ulaanbaatar, Mongolia, collected by the National Agency for Meteorology and Environmental Monitoring (NAMEM) from 2013 to 2022. Model performance was evaluated using root mean squared error (RMSE) and R² scores. Results indicate that XGBoost achieved the highest accuracy for minimum temperature prediction with an R² of 0.9713. For maximum temperature, LightGBM and XGBoost performed similarly, with LightGBM attaining the highest R² (0.9184), followed closely by XGBoost (0.9180).</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

Predicting short-term air temperature: a comparison of statistical and machine learning models

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Munkhnasan Choinzon,
  • Narantsetseg Yadmaa,
  • Juhee Choi

摘要

Accurate weather forecasting is crucial for sectors such as public health and agriculture, playing a key role in decision-making during extreme weather events. Traditional numerical weather forecasting (NWF) methods rely on solving complex physical equations, demanding extensive computational resources. Recently, machine learning techniques have emerged as cost-effective alternatives, leveraging technological advancements to improve prediction accuracy. This study compares the performance of five models—SARIMA, RNN, LSTM, XGBoost, and LightGBM—in predicting next day’s maximum and minimum temperatures using 10 years of data from Ulaanbaatar, Mongolia, collected by the National Agency for Meteorology and Environmental Monitoring (NAMEM) from 2013 to 2022. Model performance was evaluated using root mean squared error (RMSE) and R² scores. Results indicate that XGBoost achieved the highest accuracy for minimum temperature prediction with an R² of 0.9713. For maximum temperature, LightGBM and XGBoost performed similarly, with LightGBM attaining the highest R² (0.9184), followed closely by XGBoost (0.9180).