RII Track-4: NSF:Assessing Dynamic Connectivity of Streams and Wetlands across Spatial and Human Gradients with Deep Learning

湿地景观是河流和湿地相互连接的挂毯，对维持水质标准至关重要，但近几个世纪以来，由于发展和农业扩张，其覆盖范围急剧减少。这些系统去除污染物的程度，如硝酸盐，可以极大地影响下游水生系统。在了解通过湿地拦截、保留和传输硝酸盐的动态连接以及自然和人类特征在多大程度上控制这些连接的强度方面存在重大差距。这些空白限制了预测这些景观功能和有效管理它们的能力。这项研究将为一名终身助理教授和一名研究生提供机会，通过与俄亥俄州辛辛那提市的美国环保署环境测量和建模中心（CEMM）的合作，推动湿地景观连通性和水质建模领域的研究。PI和一名学生将与CEMM科学家合作，他们已经建立了最先进的基于过程的模型，通过深度学习建模框架推进建模前沿，该框架可以克服现有方法的局限性。PI的家乡堪萨斯州已经失去了48%的湿地覆盖率。因此，通过对学生的教育和开发与流域科学家、管理者和利益相关者共享的开放获取工具，这项工作将改善建模、管理和维护这些景观特征的基础设施。这项研究基础设施改善轨道4 EPSCoR研究人员（RII轨道4:NSF）项目将为堪萨斯大学的助理教授提供奖学金，并为研究生提供培训。这项工作将与美国环保署环境测量和建模中心（CEMM）的研究人员合作进行。从历史上看，湿地连通性和污染物去除效率一直是通过静态镜头来观察的，由于数据和计算的限制，经常提出的问题是“在季节性或年度期间，硝酸盐的净去除量是多少？”然而，溪流和湿地是高度动态的，有时会发生不成比例的清除。此外，人们通常认为湿地效能是由水文控制的，水文决定了硝酸盐的停留时间，但最近的研究表明，在单个湿地尺度上的效能与总体去除减少的流域尺度相比存在脱节，在单个湿地尺度上，去除似乎很高。目前存在两种知识差距，这两种差距是拟议的研究旨在弥补的。第一个是超越对连通性的静态评估，第二个是对自然和人类对湿地景观去除效率的控制的评估。该研究小组将与美国环保署环境测量与建模中心（CEMM）的研究人员合作，该中心开发了最先进的基于过程的湿地模型。虽然基于过程的模型可能是有效的工具，但它们通常高度参数化到单个位置，难以跨尺度转移。因此，本项目的研究目标是：(1)通过深度学习模型和基准来改善动态湿地连通性的表征，并将模型性能与现有的基于过程的模型相比较；(2)通过跨区域和尺度的深度学习建模来量化水文、人为和地貌对硝酸盐去除效果的控制。这个RII Track-4:NSF奖学金将为PI提供一个机会，在识别、量化和预测湿地景观行为方面取得根本性进展，并开发利益相关者可以用来改善土地管理成果的工具。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。

Wetlandscapes, the interconnected tapestries of streams and wetlands, are crucial for maintaining water quality standards, yet their coverage has dramatically decreased in recent centuries due to development and agricultural expansion. The extent to which these systems remove pollutants, such as nitrate, can greatly impact downstream aquatic systems. There exist major gaps in understanding the dynamic connections that act to intercept, retain, and transmit nitrate through wetlandscapes and the extent to which natural and human features control the strength of these connections. These gaps have limited ability to predict the function of these landscapes and manage them effectively. This research will provide an opportunity for a tenure-track assistant professor and a graduate student to propel research in the area of modeling wetlandscape connectivity and water quality through collaborations with the USEPA Center for Environmental Measurement and Modeling (CEMM) in Cincinnati, Ohio. The PI and a student will work with CEMM scientists, who have established state-of-the-science process-based models, to advance the modeling frontier with a deep learning modeling framework that can overcome the limitations of existing approaches. The PI’s home jurisdiction, Kansas, has lost 48% of its wetland coverage. Thus, this effort will improve infrastructure for modeling, managing, and maintaining these landscape features through the education of students and the development of open access tools to be shared with watershed scientists, managers, and stakeholders. This Research Infrastructure Improvement Track-4 EPSCoR Research Fellows (RII Track-4:NSF) project would provide a fellowship to an Assistant Professor and training for a graduate student at the University of Kansas. This work would be conducted in collaboration with researchers at the USEPA Center for Environmental Measurement and Modeling (CEMM). Historically, wetlandscape connectivity and contaminant removal efficiency has been viewed through a static lens whereby, due to data and computational limitations, the question often posed is “What is the net removal of nitrate over seasonal or annual periods?”. However, streams and wetlands are highly dynamic and there exist moments where disproportionate removal occurs. Further, it is often assumed that wetlandscape efficacy is controlled by hydrology, which determines the residence time of nitrate, but recent work shows a disconnect between efficacy at the individual wetland-scale, where removal appears high, compared to the catchment scale where overall removal decreases. Two knowledge gaps exist that the proposed research aims to close. The first is a move beyond a static assessment of connectivity and the second is an evaluation of the natural and human controls of wetlandscape removal efficiency. The research team will collaborate with researchers at the USEPA Center for Environmental Measurement and Modeling (CEMM) who have developed state-of-the-science, process-based wetland models. While process-based models can be effective tools, they are often highly parameterized to a single location and difficult to transfer across scales. Thus, the research objectives of this project are to (1) improve the representation of dynamic wetlandscape connectivity with a deep learning model and benchmark that model performance against existing process-based models and (2) quantify the hydrologic, anthropogenic, and geomorphic controls on nitrate removal efficacy using deep learning modeling across regions and scales. This RII Track-4:NSF fellowship will provide an opportunity for the PI to generate fundamental advances to identify, quantify, and predict wetlandscape behavior and to develop tools that stakeholders can use for improved land management outcomes.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.