CAREER: Artificial Intelligence for Polarimetric Radar Remote Sensing of Precipitation

偏振多普勒雷达一直是观测云和降水的最重要的遥感仪器，是国家恶劣天气监测和预报基础设施的基石。然而，国家的最先进的方法只能提取部分降水信息的多维极化雷达数据。该研究将通过开发可解释的人工智能（AI）技术，从雷达数据中提取丰富的信息，以提高对复杂降水过程和不同降水环境下大气动力学的理解，为雷达遥感降水开辟新的视野。该项目开发的人工智能方法可能应用于美国和世界各地的全国性天气雷达和许多研究雷达，从而促进天气，水和气候科学和服务。除了本科生和研究生的培训，计划的教育活动分布在北方科罗拉多的多所高中，教学和实地考察旨在向高中生介绍天气观测和人工智能应用，并灌输追求STEM职业的愿望。结合研究和教育活动可以培养下一代科学家，他们将熟悉人工智能和气象领域。为了实现科学目标，本项目重点是：1）研究物理指导的人工智能模型，用于从偏振雷达观测中识别水凝物和反演降水微物理; 2）设计一种具有高泛化能力的密集卷积神经网络框架，用于基于雷达的定量降水估计;（3）建立了一个可解释的人工智能降水临近预报模式，并对暴雨发生、发展和衰减的控制因子进行了研究。这项研究将通过现场数据收集、以模型和数据为中心的深度学习以及对深度学习结果的解释来完成。这项研究将提高估计和预测严重风暴的能力，从而提高对极端天气事件的态势感知，改善决策，该项目由地球科学理事会共同资助，以支持人工智能/ML在地球科学中的进步。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值进行评估而被认为值得支持和更广泛的影响审查标准。

Polarimetric Doppler radars have been the most important remote sensing instrument for observing clouds and precipitation, serving as cornerstones of the national severe weather monitoring and forecast infrastructure. However, state-of-the-art approaches have only been able to extract part of the precipitation information from the multi-dimensional polarimetric radar data. This research will open new horizons for radar remote sensing of precipitation through developing explainable artificial intelligence (AI) techniques which can extract the rich information from radar data to improve the understanding of complex precipitation processes and atmospheric dynamics in different precipitation environments. The AI methods developed from this project can potentially be applied to the nationwide operational weather radars and many research radars in the United States and worldwide, thus can advance weather, water, and climate science and service. In addition to the training of undergraduate and graduate students, the planned educational activities are spread across multiple high schools in Northern Colorado, with instruction and field trips aimed at introducing high school students to weather observations and AI applications and instilling a desire to pursue STEM careers. The integration of research and educational activities can foster the next generation of scientists who will be familiar with both AI and meteorology fields.To achieve the scientific goals, this project focuses on: 1) investigating physics-guided AI models for hydrometeor identification and precipitation microphysics retrievals from polarimetric radar observations; 2) designing a dense convolutional neural network framework that has high generalization capability for radar-based quantitative precipitation estimation; 3) developing an interpretable AI model for precipitation nowcasting and investigating the controlling factors on storm initiation, growth and decay, which are poorly understood. This research will be accomplished through field data collection, model- and data-centric deep learning, and interpretation of the deep learning results. This research will enhance the ability to estimate and predict severe storms, which will lead to improved situational awareness of extreme weather events, improved decision making, and better public safety.This project is co-funded by the Directorate for Geosciences to support AI/ML advancement in the geosciences.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.