Collaborative Research: High-Resolution Aerial Forest Mapping Infrastructure and Database to Support Forest and Disturbance Ecology Research

森林清查是了解生物形态和进程的重要资源，但传统上需要耗时的地面调查。小型无人驾驶飞机系统（sUAS，或“无人机”）和人工智能的最新进展正在开启森林研究的新时代，在这个时代，无需大量的地面调查，就可以在广阔的地区对单个树木进行测绘、测量和识别属或种。尽管现在已经有了低成本的基于无人机的森林测绘技术，但使科学家能够制作和获取广泛的森林地图的基础设施是有限的。该项目建立并促进了一个由100多个森林盘查地块组成的网络的未来扩展，每个地块约25公顷。精细尺度、大范围的森林清查数据使我们对形成森林群落和生态系统的复杂过程有了新的认识。随着干旱和高度严重的野火等压力因素导致美国西部和全球森林发生巨大变化，包括向非森林植被的转变，了解这些动态变得越来越紧迫。生态学家和森林管理者需要森林对这些新情况的反应数据来制定管理战略，但是最近变化的速度和幅度挑战了传统的基于实地的数据收集方法。该项目通过森林生态无人机飞行员学徒和强调代表性不足社区的外展活动，向下一代科学家介绍基于无人机的森林测绘工具。它利用国家科学基金会在公共网络基础设施方面的现有投资，并培训科学家将其用于云原生研究。它通过绘制森林地图以支持多方利益攸关方的森林恢复伙伴关系，展示了森林测绘基础设施与森林管理规划的相关性。在招募工作人员和学生参与者时，该项目吸引了支持代表性不足的学生和学者的团体，选择过程采用整体审查和远程旅行标准。该项目涉及开发三个互补的网络基础设施创新，以支持和扩展森林生态学和干扰生态学研究的能力：(1)可扩展、可复制、支持人工智能的软件工作流，用于将低成本无人机的图像处理为森林库存数据（例如，按大小和属或种划分的单株树木地图）；(2)一个可搜索的、可公开访问的、可扩展的树木地图数据库，其中包括100张25公顷的地图，这些地图与森林盘查样地网络（包括NSF国家生态观测网络，NEON）沿着重要的非生物和干扰历史梯度对齐；(3)记录和培训，包括虚拟和面对面的研讨会，使研究人员能够生产和贡献他们自己的数据和分析工具。该软件工作流结合了用于三维林分结构建模的摄影测量和用于分类分类和拒绝假阳性树木检测的多视图计算机视觉（通过人工神经网络），将科学家和资源管理者可实现的森林调查范围扩大了100倍。该项目利用了CyVerse，这是NSF在研究网络基础设施方面最大的投资之一，用于数据处理和数据托管。由此产生的公共森林清查数据库支持云原生研究，以改进目前受有限数据约束的森林格局和过程模型。该奖项反映了美国国家科学基金会的法定使命，并通过基金会的知识价值和更广泛的影响审查标准进行了评估，认为值得支持。

Forest inventories are critical resources for understanding biological patterns and processes, but they have traditionally required time-consuming ground-based surveys. Recent advances in small uncrewed aerial systems (sUAS, or “drones”) and artificial intelligence are enabling a new era of forest research in which individual trees can be mapped, measured, and identified to genus or species across broad areas without extensive ground surveys. Although the technology for low-cost drone-based forest mapping now exists, infrastructure to enable scientists to produce and access extensive forest maps is limiting. This project establishes and facilitates future expansion of a network of over 100 forest inventory plots of approximately 25 ha each. Fine-scale, broad-extent forest inventory data allows for new insight into the complex processes shaping forest communities and ecosystems. Understanding these dynamics is increasingly urgent as stressors such as droughts and high-severity wildfires drive dramatic shifts in forests–including conversion to non-forest vegetation–in the western U.S. and globally. Ecologists and forest managers require data on forest response to these novel conditions to develop management strategies, but the rate and magnitude of recent changes challenge traditional field-based data collection approaches. This project introduces drone-based forest mapping tools to the next generation of scientists via a Forest Ecology Drone Pilot Apprenticeship and via outreach events emphasizing underrepresented communities. It leverages existing investments in public cyberinfrastructure by NSF and trains scientists in its use for cloud-native research. It is demonstrating the relevance of the forest mapping infrastructure to forest management planning by mapping forests to support a multi-stakeholder forest restoration partnership. In recruiting staff and student participants, the project engages groups supporting underrepresented students and scholars, and the selection processes use holistic review and distance-traveled criteria.This project involves development of three complementary cyberinfrastructure innovations to support and extend the capacity of forest ecology and disturbance ecology research: (1) a scalable, reproducible, AI-enabled software workflow for processing imagery from low-cost drones into forest inventory data (e.g., maps of individual trees by size and genus or species); (2) a searchable, publicly accessible, extensible database of tree maps, initiated with 100, 25-ha maps aligned with forest inventory plot networks (including the NSF National Ecological Observatory Network, NEON) along important abiotic and disturbance history gradients; and (3) documentation and training, including virtual and in-person workshops, to enable researchers to produce and contribute their own data and analytical tools. The software workflow, which incorporates photogrammetry for 3D stand structure modeling and multi-view computer vision (via artificial neural networks) for taxonomic classification and rejection of false-positive tree detections, expands the forest survey extents achievable by scientists and resource managers by 100-fold. The project leverages CyVerse, one of NSF’s largest investments in research cyberinfrastructure, for data processing and data hosting. The resulting public forest inventory database supports cloud native research to improve models of forest pattern and process currently constrained by limited data. Open-source software development and project results are available at openforestobservatory.org.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.