Scholarlink

<p>This study applied a hybrid entropy-TOPSIS-LSTM framework approach for bias correction and CMIP6 model selection to improve the projection of extreme climate events in the Tendaho catchment, Ethiopia. Climate model outputs are essential in assessing regional extreme climate events; yet their systematic biases require correction for credible projections. The observed data were collected from the Ethiopian National Meteorological Agency. 33 CMIP6 models for temperature and 36 models for rainfall were statistically downscaled. The preprocessed climate models with CDO and extracted with the CMhyd were bias-corrected using multi-bias corrections. This study conducted a multi-model ensemble before and after the selection of bias corrections. The performance bias correction techniques, climate models, and multimodel ensemble were evaluated using the RMSE, r, MAE, and NSE. The 90th and 10th percentiles were utilized further to assess the effectiveness of the bias correction and the models in reproducing extreme conditions. The three steps performed in the selection of bias correction and climate models via a hybrid entropy-TOPSIS were: (1) accuracy score, (2) extreme score, and (3) combined scores. The distributed mapping showed superior performance consistently across all models. The top-ranked models for both maximum and minimum temperatures were FGOALS-G3, EC-Earth3-Veg-LR, and EC-Earth3-Veg. MICOR6 and EC-Earth3 models for maximum temperatures, while the MPI-ESM1-2-HR and ACCEES-ESM1-5 for minimum temperatures. The top-ranked models for rainfall were ACCEES-SEM1-5, BCC-CSM2-MR, EC-Earth3-Veg-LR, INM-CM5-0, and MICOR6. LSTM performed better than Random Forest and the mean ensemble of the multimodel outputs. Projections of extreme temperature and rainfall indices exhibit notable variation under different climate scenarios. This hybrid enhances projection reliability and supports climate-sensitive planning and policy development in the Tendaho catchment.</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

Multi-criteria hybrid-based bias correction selection, and machine learning ensemble modeling to improve extreme climate event projections in the Tendaho Catchment, Ethiopia

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Alemayehu Kabeta Guyasa,
  • Zhang Danrong,
  • Yiqing Guan,
  • Alain Niyongabo,
  • Wang Ziyuan,
  • Muhammad Imran,
  • Oli Ifeanyi Chidozie,
  • Tariku Tadesse Gameda,
  • Haihua Ying

摘要

This study applied a hybrid entropy-TOPSIS-LSTM framework approach for bias correction and CMIP6 model selection to improve the projection of extreme climate events in the Tendaho catchment, Ethiopia. Climate model outputs are essential in assessing regional extreme climate events; yet their systematic biases require correction for credible projections. The observed data were collected from the Ethiopian National Meteorological Agency. 33 CMIP6 models for temperature and 36 models for rainfall were statistically downscaled. The preprocessed climate models with CDO and extracted with the CMhyd were bias-corrected using multi-bias corrections. This study conducted a multi-model ensemble before and after the selection of bias corrections. The performance bias correction techniques, climate models, and multimodel ensemble were evaluated using the RMSE, r, MAE, and NSE. The 90th and 10th percentiles were utilized further to assess the effectiveness of the bias correction and the models in reproducing extreme conditions. The three steps performed in the selection of bias correction and climate models via a hybrid entropy-TOPSIS were: (1) accuracy score, (2) extreme score, and (3) combined scores. The distributed mapping showed superior performance consistently across all models. The top-ranked models for both maximum and minimum temperatures were FGOALS-G3, EC-Earth3-Veg-LR, and EC-Earth3-Veg. MICOR6 and EC-Earth3 models for maximum temperatures, while the MPI-ESM1-2-HR and ACCEES-ESM1-5 for minimum temperatures. The top-ranked models for rainfall were ACCEES-SEM1-5, BCC-CSM2-MR, EC-Earth3-Veg-LR, INM-CM5-0, and MICOR6. LSTM performed better than Random Forest and the mean ensemble of the multimodel outputs. Projections of extreme temperature and rainfall indices exhibit notable variation under different climate scenarios. This hybrid enhances projection reliability and supports climate-sensitive planning and policy development in the Tendaho catchment.