Scholarlink

<p>This study investigates the efficacy of Bayesian Structural Time Series (BSTS) models for forecasting seasonal climatic variables such as rainfall, maximum temperature, and minimum temperature. The data were sourced from the World Bank Climate Change Knowledge Portal covering the period from January 1901 to December 2022, representing monthly climate records for India. Using this century-long dataset, the study compares Bayesian methods (BSTS and BSTSX) with traditional statistical methods (SARIMA, SARIMAX) and advanced machine learning approaches (XGBoost, LSTM). Lower values of root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and mean absolute scaled error (MASE) were used to evaluate each model. The study reveals that Bayesian models outperform competing methods in terms of accuracy and interpretability for all three seasonal time series. The Bayesian methods improved forecast accuracy by approximately 59–66% for the rainfall series, 90–93% for maximum temperature, and 40–45% for minimum temperature, relative to competing models, based on MAPE computed from the testing set. It is acknowledged that SARIMA achieved marginally lower testing-set RMSE (0.516) and MAPE (2.230) than BSTS (RMSE: 0.614; MAPE: 2.935) for the minimum temperature series. The results show that the Bayesian structural framework is a reliable tool for climate-sensitive applications, allowing for precise forecasts that support strategies for climate adaption, water resource management, and agricultural planning.</p> Graphical Abstract <p></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

Uncertainty Quantification in Climate Forecasting: How Bayesian Structural Time Series Models Excel Over Traditional and Data-Driven Techniques

  • 期刊论文

  • 发表日期：

    2026年6月20日

查看全文
  • Achal Lama,
  • Soumik Ray,
  • Pradip Kumar Sahu,
  • K. N. Singh,
  • Bishal Gurung,
  • Bikramjeet Ghose,
  • Pradeep Mishra,
  • Tufleuddin Biswas

摘要

This study investigates the efficacy of Bayesian Structural Time Series (BSTS) models for forecasting seasonal climatic variables such as rainfall, maximum temperature, and minimum temperature. The data were sourced from the World Bank Climate Change Knowledge Portal covering the period from January 1901 to December 2022, representing monthly climate records for India. Using this century-long dataset, the study compares Bayesian methods (BSTS and BSTSX) with traditional statistical methods (SARIMA, SARIMAX) and advanced machine learning approaches (XGBoost, LSTM). Lower values of root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and mean absolute scaled error (MASE) were used to evaluate each model. The study reveals that Bayesian models outperform competing methods in terms of accuracy and interpretability for all three seasonal time series. The Bayesian methods improved forecast accuracy by approximately 59–66% for the rainfall series, 90–93% for maximum temperature, and 40–45% for minimum temperature, relative to competing models, based on MAPE computed from the testing set. It is acknowledged that SARIMA achieved marginally lower testing-set RMSE (0.516) and MAPE (2.230) than BSTS (RMSE: 0.614; MAPE: 2.935) for the minimum temperature series. The results show that the Bayesian structural framework is a reliable tool for climate-sensitive applications, allowing for precise forecasts that support strategies for climate adaption, water resource management, and agricultural planning.

Graphical Abstract