Scholarlink

<p>Global climate models (GCMs) are complex mathematical models which simulate various physical processes across different time periods and locations. Multimodel ensembles often provide more accurate predictions than individual models. However, existing ensemble methods often fail to optimally balance individual model performance and interdependence between them. This study introduces a novel distance-based weighting ensemble (DBWE) that integrates two critical factors: Firstly, the historical performance of models based on their deviations from observed data and secondly the structural independence among models. Thus the proposed method provides a robust framework for ensemble construction by addressing internal variability and model uncertainty. The evaluation of proposed ensemble is rigorously compared with existing ensemble approaches namely: Simple model average ensemble (SMAE), Conditional multimodel ensemble (CMME) and Bayesian model average (BMA). For this purpose, precipitation simulations of 18 GCMs within the Coupled Model Intercomparison Project Phase 6 have been selected for Tibetan Plateau region. The proposed ensemble scheme provides an optimal combination of weights that consistently outperforms SMAE, CMME and BMA across key evaluation metrics. On the average it achieves the lowest NRMSE (0.4175) compared to SMAE (0.4989), CMME (0.4930) and BMA (0.5365), the lowest NRAE (2.4933) and combined accuracy (CA = 9.1348) in estimating precipitation. However the CMME approach shows better correlation (0.7518) with observed precipitation as compared to other ensembles. We also investigate the skill of our approach by comparing its performance to the existing methods using training and validation data. During the training period, DBWE shows superior performance with lowest average NRMSE (0.4359) and NRAE (2.5001) as compared to SMAE (0.5181, 2.9722), CMME (0.5196, 2.970) and BMA (0.5373, 3.0228). In testing phase, the proposed ensemble again shows better ability than same and CMME in reducing uncertainty by capturing precipitation variability. It performed similar to BMA at various locations. The CMME on the other hand provided stronger correlations with observed data to better represent temporal patterns.</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

A novel statistical framework for the reduction of uncertainties in multimodel ensemble of global climate models of precipitation simulations

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Rashida Khalil,
  • Zulfiqar Ali

摘要

Global climate models (GCMs) are complex mathematical models which simulate various physical processes across different time periods and locations. Multimodel ensembles often provide more accurate predictions than individual models. However, existing ensemble methods often fail to optimally balance individual model performance and interdependence between them. This study introduces a novel distance-based weighting ensemble (DBWE) that integrates two critical factors: Firstly, the historical performance of models based on their deviations from observed data and secondly the structural independence among models. Thus the proposed method provides a robust framework for ensemble construction by addressing internal variability and model uncertainty. The evaluation of proposed ensemble is rigorously compared with existing ensemble approaches namely: Simple model average ensemble (SMAE), Conditional multimodel ensemble (CMME) and Bayesian model average (BMA). For this purpose, precipitation simulations of 18 GCMs within the Coupled Model Intercomparison Project Phase 6 have been selected for Tibetan Plateau region. The proposed ensemble scheme provides an optimal combination of weights that consistently outperforms SMAE, CMME and BMA across key evaluation metrics. On the average it achieves the lowest NRMSE (0.4175) compared to SMAE (0.4989), CMME (0.4930) and BMA (0.5365), the lowest NRAE (2.4933) and combined accuracy (CA = 9.1348) in estimating precipitation. However the CMME approach shows better correlation (0.7518) with observed precipitation as compared to other ensembles. We also investigate the skill of our approach by comparing its performance to the existing methods using training and validation data. During the training period, DBWE shows superior performance with lowest average NRMSE (0.4359) and NRAE (2.5001) as compared to SMAE (0.5181, 2.9722), CMME (0.5196, 2.970) and BMA (0.5373, 3.0228). In testing phase, the proposed ensemble again shows better ability than same and CMME in reducing uncertainty by capturing precipitation variability. It performed similar to BMA at various locations. The CMME on the other hand provided stronger correlations with observed data to better represent temporal patterns.