CRII: III: Disciplinary Knowledge Guided Big Spatial Structured Models for Geoscience Applications

该项目的目标是研究在地球科学应用中以学科知识为指导的数据科学方法的新型计算技术。数据科学领域在过去十年中取得了巨大的成功，不仅在商业领域，而且在科学和工程领域。数据驱动的方法被认为是科学发现的“第四范式”（继实验、理论和计算模拟之后）。然而，在解决跨学科问题时，纯数据驱动的方法往往面临着与该学科现有理论和知识缺乏可解释性和一致性的重大差距，著名的谷歌烟道趋势例子表明了这一点。该项目旨在通过利用学科知识指导数据驱动模型来提高可解释性、一致性和预测准确性，从而填补这一空白。具体来说，该团队将在地球科学应用的空间结构模型的背景下研究这个问题。该团队将研究学科知识在构建新的空间依赖结构中的应用，并探索有效的模型学习和推理算法。提出的方法将在水文学的跨学科应用中得到验证。如果成功，将为21世纪美国的下一代水资源管理做出贡献。所提出的研究不仅可以提高灾害响应机构的态势感知能力，还可以提高国家水模型的洪水预报能力。提出的算法将被应用到开源工具中，这将增强地球科学社区的研究基础设施。教育活动包括课程开发，在阿拉巴马州计算机科学营地指导一大批高中生参加数据科学研讨会，以及为部分高中生进行为期一年的项目，参加区域科学博览会竞赛。该项目预计将导致以下计算机科学创新。首先，研究了一种新的空间结构模型隐马尔可夫地形树（HMTT），它将隐马尔可夫模型从全阶序列推广到偏阶多树。与现有的空间结构模型（如马尔可夫随机场、空间自回归回归）相比，HMTT可以通过来自水文学学科知识（如水流方向）的更复杂的结构约束，潜在地减少样本特征中噪声和大障碍的影响。其次，探索从大量位置构建地形树的高效计算算法。最后，该团队将利用隐藏类层中的多树结构，并探索计算剪枝来减少现有动态规划方法中用于类推理的回溯次数。将学科知识（例如，结构约束）与数据驱动方法相结合的想法可以通过增强模型的可解释性和一致性来潜在地改变数据科学研究。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。

The goal of this project is to investigate novel computational techniques for disciplinary knowledge guided data science methods in geoscience applications. The field of data science has achieved tremendous success over the last decade, not only in business but also in science and engineering. The data-driven approach has been recognized as the "fourth paradigm" of scientific discovery (after experimental, theoretical, and computational simulation). However, when solving interdisciplinary problems, a purely data-driven approach often faces a significant gap in lacking interpretability and consistency with existing theories and knowledge in the discipline, as shown by the famous Google Flue Trend example. The proposed project aims to fill the gap by utilizing disciplinary knowledge to guide data-driven models to enhance interpretability, consistency, as well as prediction accuracy. Specifically, the team will study the problem in the context of spatial structured models for geoscience applications. The team will investigate the utilization of disciplinary knowledge in constructing novel spatial dependency structure and explore efficient algorithms for model learning and inference. Proposed approaches will be validated with interdisciplinary applications in hydrology. The project, if successful, will contribute towards the next generation water resource management for the U.S. in the 21st century. Proposed research can not only improve the situational awareness for disaster response agencies but also enhance the flood forecasting capabilities of the National Water Model. Proposed algorithms will be implemented into open source tools that will enhance the research infrastructure for geoscience communities. Educational activities include curriculum development, mentoring a broad group of high school students in data science seminars at Alabama Computer Science Camps, as well as year-long project for a selected number of high school students for regional Science Fair competition.The project is expected to result in the following computer science innovations. First, a novel spatial structured model called hidden Markov topography tree (HMTT) will be investigated, which generalizes existing hidden Markov models from total order sequences to partial order poly-trees. Compared with existing spatial structured models (e.g., Markov random field, spatial autoregressive regression) that captures dependency based on spatial proximity, HMTT can potentially reduce the impacts of noise and large obstacles in sample features via more complex structural constraints from disciplinary knowledge in hydrology (e.g., flow directions). Second, efficient computational algorithms to construct topography tree from a large number of locations will be explored. Finally, the team will leverage the poly-tree structure in the hidden class layer, and explore computational pruning to reduce the number of backtracking in existing dynamic programming method for class inference. The idea of integrating disciplinary knowledge (e.g., structural constraints) with data-driven methods can potentially transform data science research by enhancing model interpretability and consistency.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.