EAGER: CAS-Climate: AI-driven Probabilistic Technique, Quantile Regression based Artificial Neural Network Model, for Bias Correction and Downscaling of CMIP6 Projections

全球气候模式（GCMs）通常用于发展气候预估，以预测极端事件（如干旱、洪水）。由于计算能力的提高，GCM预估的空间分辨率有所提高，但对于需要极端事件预测以实现规划的流域尺度应用仍然不足。该项目的研究将开发一种基于人工智能的技术，以改善流域尺度上的水文气候预测。人工智能技术在模拟全球气候数据方面非常强大，可以开发更精细的时空未来气候预测。潜在的影响是在流域尺度上改进对极端事件的规划和抵御能力。本研究将开发一种基于人工智能的概率方法，该方法使用基于分位数回归的人工神经网络（ANN） （QR-AI）模型进行偏差校正和统计缩小（BCSD）耦合模型比对项目（CMIP6）预测。具体而言，研究将开发3个CMIP6在美国大陆（CONUS）预估的BCSD数据产品：1)1950-2014年GCMs降水和温度的历史模拟；2)以CO2排放和减缓情景为代表的4种不同共享社会经济路径的近期（30年）降水和温度预估。发展预报和历史预估的BCSD将提供一个机会，通过比较估计的气候变量的不确定性与观测到的CONUS上的降水和温度的边际密度，来验证QR-AI方法。BCSD CMIP6关于降水和温度的产品将采用人工智能方法为整个CONUS开发，并通过项目网站发布。BCSD的数据也将存档在figshare和github中，以供分发。此外，研究人员将与水库管理和社交媒体等重点用户群体合作，积极传播开发的BCSD产品。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Global Climate Models (GCMs) are typically used to develop climate projections to predict extreme events (e.g., droughts, floods). Spatial resolution of GCM projections has improved due to increasing computational power, but is still inadequate for watershed-scale applications where extreme event prediction is needed to enable planning. The research undertaken in this project will develop an AI-based technique to improve hydroclimatic projections at the watershed scale. AI techniques are quite powerful in modeling global climate data and could develop finer spatial and temporal future climatic projections. The potential impact is improved planning for, and resilience to, extreme events at the watershed scale.This research will develop an AI-based probabilistic approach that uses a Quantile Regression based Artificial Neural Network (ANN) (QR-AI) model for bias-corrected and statistically downscaled (BCSD) Coupled Model Intercomparison Projects (CMIP6) projections. Specifically, the research will develop three BCSD data products of CMIP6 projections over the continental U.S. (CONUS): 1) Historical simulations (1950-2014) of precipitation and temperature of GCMs; 2) Near-term (30 year) hindcasts of precipitation and temperature from relevant GCMs and 3) Near-term (30 year) projections of precipitation and temperature for four different Shared Socioeconomic Pathways, which are represented by CO2 emission and mitigation scenarios. Developing BCSD of both hindcasts and historical projections will provide an opportunity to validate the QR-AI methodology by comparing the uncertainty in the estimated climate variables with the observed marginal density of precipitation and temperature over the CONUS. The BCSD CMIP6 products on precipitation and temperature will be developed using the AI method for the entire CONUS and disseminated through the project website. BCSD data will also be archived in figshare and github for dissemination. Additionally, the investigators will work with focused user groups, such as reservoir management and social media, for active dissemination of the developed BCSD products.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.