RII Track-2 FEC: IGM--A Framework for Harnessing Big Hydrological Datasets for Integrated Groundwater Management

地下水枯竭是全球关注的主要水管理问题。在当地，由于水资源管理不善，美国东南部经历了越来越大的水资源压力，特别是在干旱时期。农业的快速扩张和无计划的城市化进一步加剧了这一问题。鉴于与水相关的产业贡献了超过1500亿美元的年收入，淡水资源的长期可持续性对该地区至关重要。虽然绘制表土、水库和河流中的水的有效性的地图继续受到关注，但目前还缺乏在大范围内以良好的时空分辨率绘制地下水储量变化的地图。这一点很重要，因为在毗邻的美国，地下水贡献了大约40%的淡水使用量，而在东南部的一些州，例如密西西比州，地下水的贡献超过了三分之二。地下水还间接支持地表水资源，因此其对淡水使用的实际贡献甚至比报告的更大。该项目的目标是利用大数据实施地下水综合管理框架，该框架将提供新的科学见解，并以前所未有的精细时空分辨率作出有用的地下水预测。IGM框架将水文、地质和卫星数据集与机器学习工具和高分辨率模拟模型相结合。所产生的信息将通过一个网络平台向广泛的利益攸关方提供，以帮助制定工程和政策解决方案。研究任务和劳动力发展工作将由来自阿拉巴马大学、路易斯安那州立大学、密西西比大学、塔斯基吉大学和南方大学这五所大学的跨学科研究人员团队共同完成。由于缺乏关于补给通量的信息，因此在精细时空尺度上预测地下水储量变化是具有挑战性的，因为补给通量受到自然地表过程(例如降水和蒸散)以及人为干预(如灌溉和抽水)变化的影响。无法绘制地下非均质性的地图是另一个主要限制。在这项研究中，我们将利用基于科学的过程模型和机器学习工具的大型水文数据集来开发精细时空尺度的地下水位和补给图。这项工作的新贡献将包括开发新的机器学习算法(如受守恒原则约束的卷积和长期-短期记忆网络)、从测井数据得出新的水文地质数据库、从测井数据计算地质横断面的新机器学习工具、使用新方法估计气候压力下植物蒸腾的实际过程模型，以及分享地下水位和补给数据集的新网络平台。地下水综合管理框架将有助于回答几个重要的科学问题：1)我们能够以精细的时间分辨率预测地下水水位和补给；2)数据驱动模型与基于过程的模型相比，获取地下水补给的效率有何不同？混合方法的优势是什么？3)地下水干旱恢复过程的实际控制是什么？这项裁决反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Groundwater depletion is a major water management problem that is of global concern. Locally, the Southeastern US has experienced increased water stress due to the mismanagement of its water resources, especially during drought periods. Rapid agricultural expansion and unplanned urbanization have further aggravated this problem. Given that water-related industries contribute to over 150 billion of US dollars in annual revenues, the long-term sustainability of freshwater resources is of paramount importance to this region. While mapping the availability of water in topsoil, reservoirs, and rivers continues to receive much attention, mapping of groundwater storage changes at a fine spatiotemporal resolution over large areas is currently lacking. This is important because groundwater contributes around 40 percent of freshwater usage in the conterminous US, and its contribution in some Southeastern states, e.g., Mississippi, is over two-thirds. Groundwater also indirectly sustains surface water resources, and hence its actual contribution to freshwater usage is even larger than reported. The goal of this project is to harness the big data to implement an integrated groundwater management (IGM) framework that will provide new scientific insights and make useful groundwater predictions at an unprecedented fine spatiotemporal resolution. The IGM framework integrates hydrological, geological, and satellite datasets with machine learning tools and high-resolution simulation models. The information generated will be made available to a wide group of stakeholders through a web-based platform to help develop engineering and policy solutions. The research tasks and workforce development efforts will be jointly accomplished by a team of interdisciplinary researchers at five universities: The University of Alabama, Louisiana State University, University of Mississippi, Tuskegee University, and Southern University.Prediction of groundwater storage changes at fine spatiotemporal scales is challenging due to lack of information about recharge fluxes, which are influenced by variations in natural land surface processes (e.g., precipitation and evapotranspiration) and anthropogenic interventions such as irrigation and pumping. The inability to map subsurface heterogeneities is another major limitation. In this study, we will harness big hydrologic datasets using science-based process models and machine learning tools to develop groundwater level and recharge maps at fine spatiotemporal scales. Novel contributions from this effort will include the development of new machine learning algorithms (such as convolutional and long-short term memory networks constrained by conservation principles), a new hydrogeological database derived from well log data, new machine learning tools for developing geological cross-sections from well log data, physically-realistic process models that use novel methods for estimating plant transpiration under climatic stress, and a new web platform for sharing groundwater level and recharge datasets. The integrated groundwater management framework will help answer several important science questions: 1) How well can we predict the groundwater levels and recharge at fine temporal resolution? 2) How different is the efficiency of data driven models compared to process-based models for obtaining groundwater recharge, and what are the advantages of a hybrid approach? and 3) What are the physical controls on groundwater drought-recovery processes?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.