Understanding future temperature extremes is pivotal to preparing for and mitigating the impacts of climate change. This study proposed machine learning techniques to develop a multi-model ensemble model for high-resolution projection of global land temperature extremes under different emission scenarios, hence providing enhanced precision over previous climate model projections. By utilizing the NEX-GDDP-CMIP6 dataset with bias adjustment and the Gradient Booster algorithm, we reduced the biases that existed in Global Climate Models. The model significantly reduces the root mean square errors (RMSEs) for both the daily maximum and daily minimum temperature extremes. A future scenario analysis revealed that global temperature extremes would substantially increase under high-emission scenarios, highlighting the urgency for stringent emission reduction commitments. This study also identified regions like Greenland, the Tibetan Plateau, and the regional Arctic Archipelago as potential hotspots of temperature extremes under these scenarios. The multi-model ensemble approach, tuned with machine learning and driven by high-resolution data, contributes to climate science by providing refined insights into future temperature extremes, thereby offering direction to climate change mitigation and adaptation strategies.

中文翻译：

---

未来全球陆地极端温度的变化


了解未来的极端温度对于准备和减轻气候变化的影响至关重要。本研究提出了机器学习技术，以开发一种多模式集成模型，用于在不同排放情景下高分辨率预测全球陆地极端温度，从而提供比以前的气候模式预测更高的精度。通过利用带有偏差调整的 NEX-GDDP-CMIP6 数据集和 Gradient Booster 算法，我们减少了全球气候模型中存在的偏差。该模型显著降低了每日最高温度和每日最低温度极端值的均方根误差 （RMSE）。未来情景分析显示，在高排放情景下，全球极端温度将大幅增加，凸显了严格减排承诺的紧迫性。这项研究还确定了格陵兰岛、青藏高原和北极群岛等地区是这些情景下极端温度的潜在热点地区。多模式集成方法通过机器学习进行调整并由高分辨率数据驱动，通过提供对未来极端温度的精细见解来为气候科学做出贡献，从而为气候变化缓解和适应策略提供方向。