Scholarlink

<p>This study conducts an observation-constrained intercomparison of 23 CMIP6 global climate models (GCMs) for precipitation over Japan. We quantify relative agreement between models and a regional reference analysis from the JMA mesoscale system (GPV-MSM) for 2015–2023. Eight precipitation indices (mean, 90th percentile, max5d, CWD, etc.) and four statistical metrics (correlation, Pbias, NRMSE, Taylor skill score) are synthesized via entropy-weighted TOPSIS to obtain a proximity-to-reference coefficient and a similarity-based ordering. To limit interpolation artifacts, all fields are remapped to a 1.0° × 1.0° common grid. Because the CMIP6 Historical experiment ends in 2014, ScenarioMIP SSP585 simulations are used for 2015–2023. Axis-specific results show a stable latitudinal top-tier cohort comprising ACCESS-CM2, KACE-1-0-G, and UKESM1-0-LL, each with Top-3 inclusion probabilities ≥ 89%. In the longitudinal direction, a distinct three-model cohort—KACE-1-0-G, CNRM-ESM2-1, and CESM2-WACCM—emerges with Top-3 probabilities near 60%. Bootstrap resampling　indicates consistent rank structures (median Kendall’s Tau = 0.87 for latitude and 0.72 for longitude), with wider intervals in the longitudinal rankings indicating larger sampling variability. Across indices, the intercomparison indicates systematic model-reference differences: many models show lower amplitudes in extreme-intensity indices (90P, max5d) together with positive differences in wet-spell persistence (CWD, CWD10). The synthesis therefore identifies cohorts with comparatively higher similarity across multiple indicators (e.g., ACCESS-CM2, KACE-1-0-G, UKESM1-0-LL) and others with lower similarity (e.g., INM-CM4-8, IPSL-CM6A-LR). All findings are interpreted as relative agreement with an observation-constrained regional analysis over 2015–2023 and are intended to support bias-aware, purpose-specific use of CMIP6 precipitation in regional applications.</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

Observation-Constrained Intercomparison of CMIP6 GCMs Precipitation Projections around Japan

  • 期刊论文

  • 发表日期：

    2026年1月5日

查看全文
  • Do-yup Kim,
  • Naoki Shirakawa

摘要

This study conducts an observation-constrained intercomparison of 23 CMIP6 global climate models (GCMs) for precipitation over Japan. We quantify relative agreement between models and a regional reference analysis from the JMA mesoscale system (GPV-MSM) for 2015–2023. Eight precipitation indices (mean, 90th percentile, max5d, CWD, etc.) and four statistical metrics (correlation, Pbias, NRMSE, Taylor skill score) are synthesized via entropy-weighted TOPSIS to obtain a proximity-to-reference coefficient and a similarity-based ordering. To limit interpolation artifacts, all fields are remapped to a 1.0° × 1.0° common grid. Because the CMIP6 Historical experiment ends in 2014, ScenarioMIP SSP585 simulations are used for 2015–2023. Axis-specific results show a stable latitudinal top-tier cohort comprising ACCESS-CM2, KACE-1-0-G, and UKESM1-0-LL, each with Top-3 inclusion probabilities ≥ 89%. In the longitudinal direction, a distinct three-model cohort—KACE-1-0-G, CNRM-ESM2-1, and CESM2-WACCM—emerges with Top-3 probabilities near 60%. Bootstrap resampling　indicates consistent rank structures (median Kendall’s Tau = 0.87 for latitude and 0.72 for longitude), with wider intervals in the longitudinal rankings indicating larger sampling variability. Across indices, the intercomparison indicates systematic model-reference differences: many models show lower amplitudes in extreme-intensity indices (90P, max5d) together with positive differences in wet-spell persistence (CWD, CWD10). The synthesis therefore identifies cohorts with comparatively higher similarity across multiple indicators (e.g., ACCESS-CM2, KACE-1-0-G, UKESM1-0-LL) and others with lower similarity (e.g., INM-CM4-8, IPSL-CM6A-LR). All findings are interpreted as relative agreement with an observation-constrained regional analysis over 2015–2023 and are intended to support bias-aware, purpose-specific use of CMIP6 precipitation in regional applications.