Scholarlink

<p>Accurate monthly forecasts of the Total Ozone Column (TOC) are essential for UV advisories and environmental planning. Using a 20-year national dataset for Iran, we develop a multi-model benchmark comparing four forecasting families: classical (SARIMA), component-based (Prophet), and deep learning (RNN, LSTM). Prophet achieved the highest nominal accuracy on the test set (<i>R</i><sup>2</sup> = 0.83). Although its performance was not statistically superior to LSTM (<i>R</i><sup>2</sup> = 0.76), it showed significant gains over SARIMA and RNN under a more permissive threshold. Beyond test-set evaluation, two sensitivity analyses were conducted to examine robustness. A fivefold cross-validation confirmed that model behavior remains stable across earlier periods, and a spatial assessment across all 31 provinces demonstrated that Prophet retains its performance advantage in most regions despite climatic variability. From an operational standpoint, Prophet offers a balance of accuracy, computational efficiency, and interpretability through its trend–seasonality decomposition. These results show that reliable nationwide TOC forecasts can be obtained without computationally intensive architectures.</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

Comparative forecasting of Total Ozone Column over Iran: a 20-year multi-model assessment

  • 期刊论文

  • 发表日期：

    2025年12月28日

查看全文
  • Mohammad Hashemzadeh,
  • Faezeh Borhani,
  • Poorya Shahverdi

摘要

Accurate monthly forecasts of the Total Ozone Column (TOC) are essential for UV advisories and environmental planning. Using a 20-year national dataset for Iran, we develop a multi-model benchmark comparing four forecasting families: classical (SARIMA), component-based (Prophet), and deep learning (RNN, LSTM). Prophet achieved the highest nominal accuracy on the test set (R2 = 0.83). Although its performance was not statistically superior to LSTM (R2 = 0.76), it showed significant gains over SARIMA and RNN under a more permissive threshold. Beyond test-set evaluation, two sensitivity analyses were conducted to examine robustness. A fivefold cross-validation confirmed that model behavior remains stable across earlier periods, and a spatial assessment across all 31 provinces demonstrated that Prophet retains its performance advantage in most regions despite climatic variability. From an operational standpoint, Prophet offers a balance of accuracy, computational efficiency, and interpretability through its trend–seasonality decomposition. These results show that reliable nationwide TOC forecasts can be obtained without computationally intensive architectures.