LTREB: Integrating real-time open data pipelines and forecasting to quantify ecosystem predictability at day to decadal scales

由于人类活动，许多生态系统正在表现出变异性的增加。这种环境的可变性给管理者和决策者带来了巨大的挑战，他们不能再使用历史基线来指导对未来生态系统状况的预测。因此，提高对影响水质的一系列物理、化学和生物生态系统变量的未来预测能力对于改善资源管理至关重要。为了满足这一需求，环境生物学长期研究（LTREB）项目将支持两个饮用水供应水库的现场监测和数据共享计划。将收集的生态系统数据（包括水温、清晰度、化学成分和浮游植物等变量）将用于生成和评估每日至年度的实时预测（对未来生态系统状况的预测，以及与之相关的不确定性）。预测是量化生态可预测性的有力方法，因为它需要使用代表我们对生态系统功能的最佳假设的模型来预测未来的生态条件。随着新数据的收集，对这些预测进行迭代评估，将揭示哪些模型在不同的环境条件下表现最好，并确定可以提前一天到十年准确预测未来不同变量的距离。这个项目将能够检验关于生态系统可预测性的基本假设；制定新的工作流程，将环境观测纳入实时预报和数据发布；扩大来自代表性不足群体的学生在环境数据科学领域的参与。此外，所有预报都将实时分发给供水公司，使它们能够立即用作水管理的决策工具。这个LTREB项目将是对生态系统动态可预测性的首次系统分析之一，从而提供有关不同生态系统之间和生态系统变量之间可预测性的梯度和控制的宝贵信息。重要的是，研究人员将能够比较不同预测模型的性能与生态系统动力学的竞争表示（例如，不同的驱动变量，模型结构），以测试关于可预测性的生态假设，并检查对生态系统功能的控制。例如，将两个储层的预测精度进行比较，这两个储层的所有特征都相似，除了氧气可用性，以确定全球水体中不断增加的缺氧如何改变可预测性。此外，该LTREB项目将与环境数据倡议（EDI）合作，开发新颖的FAIR（可查找、可访问、可互操作、可重用）数据发布工作流程，以提高生态学的可重复性，支持环境数据科学教育，并使生态预测能够扩展到其他站点。总而言之，该项目将产生水库物理、化学和生物生态系统变量的数据产品，可用于实时自动预测；一套不同的预测模型和评估预测；预测和数据发布工作流程和软件；最重要的是，获得了关于储层动态可预测性的大量生态系统知识。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。

Many ecosystems are exhibiting increased variability as a result of human activities. This environmental variability poses substantial challenges for managers and decision-makers, who can no longer use historical baselines to guide predictions of future ecosystem conditions. Consequently, advancing the capacity to predict the future for a range of physical, chemical, and biological ecosystem variables that influence water quality is paramount for improving resource management. In response to this need, this Long Term Research in Environmental Biology (LTREB) project will support a field monitoring and data-sharing program at two drinking water supply reservoirs. The ecosystem data that will be collected (which will include water temperature, clarity, chemistry, and phytoplankton, among other variables) will be used to generate and evaluate real-time forecasts (predictions of future ecosystem conditions, and the uncertainty associated with them) at daily to annual scales. Forecasting is a powerful approach for quantifying ecological predictability, as it requires using models that represent our best hypotheses about how ecosystems function to predict ecological conditions into the future. Iteratively evaluating these forecasts as new data are collected will reveal which models perform best in different environmental conditions and identify how far into the future different variables can be accurately predicted, from one day to one decade in advance. This project will enable the testing of fundamental hypotheses about the predictability of ecosystems; develop novel workflows for integrating environmental observations into real-time forecasting and data publishing; and broaden the participation of students from underrepresented groups in environmental data science. Moreover, all forecasts will be disseminated to water utilities in real time, enabling their immediate use as decision-making tools for water management. This LTREB project will represent one of the first systematic analyses of the predictability of ecosystem dynamics, thereby providing valuable information on the gradients and controls of predictability between contrasting ecosystems and among ecosystem variables. Importantly, researchers will be able to compare the performance of different forecast models with competing representations of ecosystem dynamics (e.g., varying driver variables, model structures) to test ecological hypotheses about predictability and examine the controls on ecosystem function. For example, forecast accuracy will be compared between two reservoirs that are similar in all characteristics except for oxygen availability to determine how anoxia, which is increasing in waterbodies globally, alters predictability. In addition, this LTREB project will develop novel FAIR (Findable, Accessible, Interoperable, Reusable) data-publishing workflows in collaboration with the Environmental Data Initiative (EDI) that advance reproducibility in ecology, support environmental data science education, and enable the scaling of ecological forecasting to other sites. Altogether, this project will result in data products for water reservoir physical, chemical, and biological ecosystem variables available in real-time for automated forecasting; a suite of different forecasting models and evaluated forecasts; forecasting and data-publishing workflows and software; and most critically, substantial ecosystem knowledge gained about the predictability of reservoir dynamics.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.