Scholarlink

A vital phase in reducing simulation uncertainty is to identify the most skillful GCMs based on their statistical ranking and create multi-model ensembles (MMEs). This research focuses on the statistical evaluation of 20 general circulation models (GCMs) under IPCC’s sixth assessment report. As observed data, daily temperature data from 1965 to 2014 were acquired from the India Meteorological Department (IMD). A gridded resolution of 2° × 2° was used to evaluate the maximum and minimum temperatures for North-East India. Absolute normalized mean bias deviation (ANMBD), normalized root mean square deviation (NRMSD), correlation coefficient (CC), and skill score (SS) were the performance metrics employed. To get a relative ranking, weights were assigned to each performance metric using the dynamic weights approach. Compromise Programming was employed to rank each grid individually, and a group decision-based method was adopted to rank all grids collectively. MMEs of GCMs are established using machine learning-based Support Vector Regression (SVR) and Random Forest Regression (RFR) techniques. For maximum temperature, ACCESS-ESM1-5, NESM3, and INM-CM5-0, and for minimum temperature, TaiESM1, IITM-ESM, and INM-CM5-0, were among the top 3 ranked GCMs. For the study area, RFR is preferable over SVR-based ML techniques. For MaxT and MinT, the optimal GCM number for generating MMEs is around 35% and 25% of the top-ranked GCMs, respectively.

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-Model Ensemble Prediction of Maximum and Minimum Temperature Using Machine Learning Approach for North-East India

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Abdul Rahman,
  • Sreeja Pekkat

摘要

A vital phase in reducing simulation uncertainty is to identify the most skillful GCMs based on their statistical ranking and create multi-model ensembles (MMEs). This research focuses on the statistical evaluation of 20 general circulation models (GCMs) under IPCC’s sixth assessment report. As observed data, daily temperature data from 1965 to 2014 were acquired from the India Meteorological Department (IMD). A gridded resolution of 2° × 2° was used to evaluate the maximum and minimum temperatures for North-East India. Absolute normalized mean bias deviation (ANMBD), normalized root mean square deviation (NRMSD), correlation coefficient (CC), and skill score (SS) were the performance metrics employed. To get a relative ranking, weights were assigned to each performance metric using the dynamic weights approach. Compromise Programming was employed to rank each grid individually, and a group decision-based method was adopted to rank all grids collectively. MMEs of GCMs are established using machine learning-based Support Vector Regression (SVR) and Random Forest Regression (RFR) techniques. For maximum temperature, ACCESS-ESM1-5, NESM3, and INM-CM5-0, and for minimum temperature, TaiESM1, IITM-ESM, and INM-CM5-0, were among the top 3 ranked GCMs. For the study area, RFR is preferable over SVR-based ML techniques. For MaxT and MinT, the optimal GCM number for generating MMEs is around 35% and 25% of the top-ranked GCMs, respectively.