Collaborative Research: OAC Core: Learning AI Surrogate of Large-Scale Spatiotemporal Simulations for Coastal Circulation

近9亿人生活在低洼沿海地区的气候危机前线，死于洪水和风暴的可能性是其他地区的15倍。科学家们需要模拟沿着的洋流循环，以建立预警系统，从而拯救无数生命，防止最易受气候变化影响的发展中国家每年遭受重大损失。传统上，这种模拟是通过在高性能计算（HPC）平台上运行数值模型来进行的，这既昂贵又耗时。在过去的几年里，在深度学习和新兴的图形处理单元（GPU）计算架构的推动下，该领域发生了迅速的转变。其主要思想是训练数值模型的神经网络代理，一旦经过预训练，网络就可以以更快的速度和更小的能量足迹生成模拟。该项目将开发一种新的人工智能替代网络基础设施，用于沿海环流的大规模时空模拟。教育活动将包括课程开发，在夏令营的人工智能研讨会上指导广大的高中生，以及为选定数量的高中生举办为期一年的区域科学博览会竞赛项目。该项目将在人工智能和网络基础设施研究方面提供几项创新。首先，它将研究一种新的人工智能代理模型架构，以捕获沿海环流模拟的独特时空数据特征。其次，它将探索几种策略来优化模型的时间和GPU内存效率。最后，在多GPU集群上设计并实现了一个可扩展的模型训练和推理流水线。该项目由美国国家科学基金会的国家发现云气候倡议资助。该奖项反映了NSF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。

Nearly 900 million people live at the front line of the climate crisis in low-lying coastal zones and are 15 times more likely to die from flooding and storms. Scientists need to simulate ocean current circulation along the coasts to develop early warning systems that could save countless lives and prevent significant annual losses in developing countries that are most vulnerable to the impacts of climate change. Traditionally, such simulations are conducted by running numerical models on a high-performance computing (HPC) platform, which is both expensive and time-consuming. The last few years have witnessed a rapid transformation of the field driven by advances in deep learning and the emerging Graphics Processing Unit (GPU) computational architecture. The main idea is to train neural network surrogates of numerical models, and once pre-trained, the networks can generate simulations with much faster speed and a smaller energy footprint. The project will develop a novel AI surrogate cyberinfrastructure for large-scale spatiotemporal simulations in coastal circulation. Educational activities will include curriculum development, mentoring a broad group of high school students in AI seminars at Summer Camps, as well as year-long projects for a selected number of high school students for the regional Science Fair competition. The project will provide several innovations in AI and cyberinfrastructure research. First, it will investigate a novel AI surrogate model architecture to capture the unique spatiotemporal data characteristics of coastal circulation simulations. Second, it will explore several strategies to optimize the model for time and GPU memory efficiency. Finally, it will design and implement a scalable model training and inference pipeline on a multi-GPU cluster. This project is funded by the National Science Foundation's National Discovery Cloud for Climate initiative.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.