腾讯会议 (会议ID: 777 666 381)
The fidelity of climate projections is often undermined by biases in climate models due to their simplification or misrepresentation of the climate processes. While various bias correction methods have been developed to post-process model outputs to match observations, existing approaches usually focus on limited, lower-order statistics, or break either spatiotemporal consistency of the target variable, or its dependency upon model resolved dynamics. We develop a Regularized Adversarial Domain Adaptation (RADA) methodology to overcome these deficiencies, and enhance efficient identification and correction of climate model biases. Instead of pre-assuming where the climate model bias lies, we apply discriminative neural networks to learn an arbitrary mismatch between the distribution of historical climate simulations and observations, meanwhile applying the discriminator’s feedbacks to train the conditional domain adaptation neural networks to close the mismatch under cycle-consistent statistical and dynamical constraints. An application to daily precipitation projection over the contiguous United States shows that our methodology can correct all moments of daily precipitation at approximately 1 degree resolution, ensures spatiotemporal consistency and inter-field correlations, and can discriminate between different dynamical conditions. Our methodology offers a powerful tool for disentangling model parameterization biases from their interactions with the chaotic evolution of climate dynamics, opening a novel avenue toward big-data enhanced climate prediction.
潘宝祥，美国劳伦斯利弗摩尔国家实验室博士后，主要研究兴趣包括概率信息理论、结合机器学习与动力模式的天气-气候尺度预报、动力系统可预报性。2012年本科毕业于武汉大学，2015年在清华大学获得工学硕士学位，2019年于加州大学欧文分校（University of California, Irvine）获得工学博士学位，师从Dr. Soroosh Sorooshian, Dr. Kuolin Hsu, Dr. Amir AghaKouchak。