Scholarlink

<p>The investigation of future marine heatwave (MHW) trends in a changing climate is seriously hindered by the inability of current-generation global climate models (GCMs) to accurately reproduce high-frequency sea surface temperature (SST) variations. This limitation of GCMs is due to their coarse resolution and lack of parameterization for important small-scale processes. In this study, we introduce a novel hybrid model that combines low-frequency GCM simulation data with a machine learning (ML) component to incorporate realistic high-frequency SST variations. The ML model component is trained on 42 years of historical ERA5-reanalysis data. The hybrid model serves as a computationally inexpensive add-on to incorporate realistic high-frequency SST variability into existing GCM ensemble simulations or other low-frequency SST products. We demonstrate that the hybrid model yields MHW statistics that more closely align with past observations compared to data derived solely from MPI-ESM1.2-LR, a commonly used GCM. In particular, the well-known bias of GCMs towards longer and less frequent MHWs vanishes entirely in the hybrid model. We then utilize the hybrid model to examine the temporal evolution of MHW statistics on the North-West European Shelf (NWES) from 1850 to 2100, across multiple state-of-the-art CMIP6 scenarios. MHW statistics are projected to saturate in the future, leading to a counterintuitive decrease in event frequency. A return period analysis reveals that extremely long-lasting MHWs are expected to become 100 times more likely by the end of the century, with cumulative intensities reaching levels that were virtually impossible during pre-industrial times.</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

Improving marine heatwave statistics in global climate models using machine learning: a case study for the north–west European Shelf

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月21日

查看全文
  • Johannes Becherer,
  • Thomas Pohlmann

摘要

The investigation of future marine heatwave (MHW) trends in a changing climate is seriously hindered by the inability of current-generation global climate models (GCMs) to accurately reproduce high-frequency sea surface temperature (SST) variations. This limitation of GCMs is due to their coarse resolution and lack of parameterization for important small-scale processes. In this study, we introduce a novel hybrid model that combines low-frequency GCM simulation data with a machine learning (ML) component to incorporate realistic high-frequency SST variations. The ML model component is trained on 42 years of historical ERA5-reanalysis data. The hybrid model serves as a computationally inexpensive add-on to incorporate realistic high-frequency SST variability into existing GCM ensemble simulations or other low-frequency SST products. We demonstrate that the hybrid model yields MHW statistics that more closely align with past observations compared to data derived solely from MPI-ESM1.2-LR, a commonly used GCM. In particular, the well-known bias of GCMs towards longer and less frequent MHWs vanishes entirely in the hybrid model. We then utilize the hybrid model to examine the temporal evolution of MHW statistics on the North-West European Shelf (NWES) from 1850 to 2100, across multiple state-of-the-art CMIP6 scenarios. MHW statistics are projected to saturate in the future, leading to a counterintuitive decrease in event frequency. A return period analysis reveals that extremely long-lasting MHWs are expected to become 100 times more likely by the end of the century, with cumulative intensities reaching levels that were virtually impossible during pre-industrial times.