Scholarlink

<p>Reliable rainfall projections are essential for climate adaptation in tropical megacities, where rapid urbanization amplifies exposure to hydroclimatic extremes. This study conducts a comprehensive comparative evaluation of 13 established bias-correction methods, including traditional statistical techniques and machine-learning models, applied to daily rainfall from eighteen CMIP6 GCMs at 1&#xa0;km grid spacing using a high-resolution observational reference dataset over Greater Kuala Lumpur (GKL), Malaysia. Bias-correction methods were assessed using the non-parametric Kling–Gupta Efficiency (KGE) for spatial and seasonal performance and the Perkins Skill Score (PSS) for distributional accuracy. By comparing methods across spatial, seasonal, and distributional performance dimensions, the study provides evidence-based guidance for bias-correction method selection in tropical urban settings. The results show that Detrended Quantile Mapping (DQM) achieved the best overall performance (average rank = 2.00), followed by Delta correction (average rank = 2.33) and Quantile Delta Mapping (average rank = 3.67). DQM reduced inter-model rainfall spread by 92%, improved RMSE by 53%, and maintained spatial correlation (<i>r</i> ≈ 0.66) with observations during the 1975–2014 baseline. The bias-corrected ensemble revealed robust future rainfall intensification under all SSPs. Annual precipitation (PRCPTOT) is projected to increase by 7–16%, while extreme one-day rainfall (Rx1day) rises by 9–42%, with the greatest changes under SSP3-7.0 and SSP5-8.5 (far-future). Spatially, PRCPTOT increases dominate western GKL due to enhanced moisture convergence from the Strait of Malacca, while Rx1day intensifies in the eastern highlands driven by orographic uplift. The findings highlight DQM’s superior trend-preserving ability, providing a robust foundation for locally adjusted climate risk assessments at 1&#xa0;km grid spacing and flood adaptation strategies in rapidly urbanizing Southeast Asian megacities.</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

Evaluating statistical and machine learning bias-correction methods for CMIP6 rainfall projections in a tropical megacity

  • 期刊论文

  • 发表日期：

    2026年4月18日

查看全文
  • Nirwani Devi Miniandi,
  • Mohamad Hidayat Jamal,
  • Mohd Khairul Idlan Muhammad,
  • Shamsuddin Shahid

摘要

Reliable rainfall projections are essential for climate adaptation in tropical megacities, where rapid urbanization amplifies exposure to hydroclimatic extremes. This study conducts a comprehensive comparative evaluation of 13 established bias-correction methods, including traditional statistical techniques and machine-learning models, applied to daily rainfall from eighteen CMIP6 GCMs at 1 km grid spacing using a high-resolution observational reference dataset over Greater Kuala Lumpur (GKL), Malaysia. Bias-correction methods were assessed using the non-parametric Kling–Gupta Efficiency (KGE) for spatial and seasonal performance and the Perkins Skill Score (PSS) for distributional accuracy. By comparing methods across spatial, seasonal, and distributional performance dimensions, the study provides evidence-based guidance for bias-correction method selection in tropical urban settings. The results show that Detrended Quantile Mapping (DQM) achieved the best overall performance (average rank = 2.00), followed by Delta correction (average rank = 2.33) and Quantile Delta Mapping (average rank = 3.67). DQM reduced inter-model rainfall spread by 92%, improved RMSE by 53%, and maintained spatial correlation (r ≈ 0.66) with observations during the 1975–2014 baseline. The bias-corrected ensemble revealed robust future rainfall intensification under all SSPs. Annual precipitation (PRCPTOT) is projected to increase by 7–16%, while extreme one-day rainfall (Rx1day) rises by 9–42%, with the greatest changes under SSP3-7.0 and SSP5-8.5 (far-future). Spatially, PRCPTOT increases dominate western GKL due to enhanced moisture convergence from the Strait of Malacca, while Rx1day intensifies in the eastern highlands driven by orographic uplift. The findings highlight DQM’s superior trend-preserving ability, providing a robust foundation for locally adjusted climate risk assessments at 1 km grid spacing and flood adaptation strategies in rapidly urbanizing Southeast Asian megacities.