A framework to predict hydrologic processes at continental scales

在用水压力日益增大的情况下，河流流量预测对于预测洪水和管理水资源至关重要。为了对所有河流做出可靠的流量预测，包括那些没有流量计的河流，我们需要计算机模型来准确地模拟流域过程，以及它们在美国各地的变化。例如，地表流量、补给、地下水储存和流量模式如何随着流域的不同而变化？最新的水文模型足够灵活，可以模拟空间变化的过程，但我们目前缺乏这些过程如何随流域变化的知识。该项目将通过开发一个新的框架来预测美国各地的流域过程如何变化，从而填补这一知识空白。该方法在利用大陆尺度的机器学习应用中的小规模现场水文知识方面是新颖的。该研究将发现景观特征、水流动力学和流域过程之间的新关系。项目科学家将与NOAA的国家水中心合作，将研究结果应用于下一代国家水模型的设计，该模型将为美国的每条河流提供流量预测。该项目将为代表性不足的少数民族学生提供研究经验，并将开发在线学习材料。该项目的目标是：(1)确定一套景观指标，量化最有可能激活特定径流生成过程的景观特征。(2)通过将河流动态与驱动它们的上游过程联系起来，在美国测量流域的大型数据库中确定主要的水文过程。(3)建立基于景观指标的水文主导过程预测数据驱动模型。(4)通过对一系列地点和案例研究进行测试来评估数据驱动模型。本项目开发的框架将改进以前识别和预测景观和水文指标的方法，通过重新设计指标来针对特定的水文过程。此外，该项目将应用新的机器学习发展来识别和解释景观和过程之间的预测关系。交付成果将包括美国相邻地区水文过程的GIS（地理信息系统）地图，以及从景观特征估计水文过程的开源代码。总体而言，该项目渴望改变大陆尺度水文模型在不同气候和景观下表征水通量的方式。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。

Streamflow predictions are essential for forecasting floods and managing water resources under intensifying pressures on water use. To make reliable streamflow predictions for all rivers, including those with no flow gauges, we need computer models that accurately simulate watershed processes and how they vary across the U.S. landscape. For example, how do surface flows, recharge, groundwater storage and flow patterns change from watershed to watershed? The latest hydrologic models are flexible enough to simulate spatially variable processes, but we currently lack the knowledge of how those processes vary by watershed. This project will fill this knowledge gap by developing a new framework to predict how watershed processes vary across the U.S.. The approach is novel in leveraging small-scale field hydrology knowledge within a continental-scale, machine learning application. The research will discover new relationships between landscape features, streamflow dynamics and watershed processes. Project scientists will work with NOAA’s National Water Center to apply the results in the design of the Next-Generation National Water Model that provides streamflow predictions for every river in the U.S.. The project will provide research experiences for under-represented minority students, and will develop online learning materials. The goals of the project are to (1) Identify a suite of landscape metrics that quantify landscape characteristics most likely to activate specific runoff generation processes. (2) Identify dominant hydrologic processes across a large database of gauged U.S. watersheds, by relating streamflow dynamics to the upstream processes that drive them. (3) Develop a data-driven model that predicts dominant hydrologic processes based on landscape metrics. (4) Evaluate the data-driven model by testing it for a range of locations and case studies. The framework developed in this project will improve on previous methods of identifying and predicting landscape and hydrologic metrics, by redesigning the metrics to target specific hydrologic processes. Further, the project will apply new machine learning developments to identify and interpret predictive relationships between landscapes and processes. Deliverables will include GIS (geographic information system) maps of hydrologic processes across the contiguous U.S., and open-source code to estimate hydrologic processes from landscape characteristics. Overall, the project aspires to transform how continental-scale hydrology models represent water fluxes in diverse climates and landscapes.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.