Change and Stability of Canada's Forested and Aquatic Landscapes: Big-data Remote Sensing on Multiple Platforms

卫星遥感的最新发展表明，我们已经进入了一个新时代。在卫星信息一度稀缺和昂贵的地方，大量的档案已经向公众免费开放。在处理能力和磁盘空间一度会阻碍大面积研究计划的地方，具有巨大容量的新工具已经出现，使超级计算机触手可及。这种新模式的遥感进展迅速而令人印象深刻，产生了广泛现象的全球评估：全球潮平图、高分辨率人口估计、森林损失和收益的时间序列图——仅举几例。尽管取得了这一巨大进展，但相当大的障碍限制了我们利用世界地球资源卫星的巨大潜力。将来自多个平台的数据流组合在一起是相当复杂的：不同的传感器提供不同空间分辨率的图像，具有不同的光谱带数量和规格，到达时间不规律，通常部分被云覆盖，并且设计用于广泛的目的。把这些传感器想象成一个单一的地球观测机器是令人兴奋的，它不断地从电磁频谱的不同部分获取视图，但是一次处理超过几个维度的信号是很困难的。这项研究有两个目的。首先，我们将使用我们独特的算法来混合来自各种传感器的信息，使我们能够更快速、更可靠地观察变化和稳定性。其次，我们将整合新的机器学习技术来解释旧的低信息图像，使我们能够看到加拿大在卫星时代开始时的样子。这些目标将更全面地探索从20世纪70年代初到即将到来的夏季陆地和水中大火的地球表面的巨大图像宝库。在接下来的五年里，我的研究团队包括本科生预科生、本科生、硕士生、博士生和博士后，他们将追求四个中期目标：(1)结合多个数据流，每周绘制加拿大各地的收获、火灾边界和其他变化；(2)将火灾增长模型与多个传感器的信息流融合，近实时跟踪野火的增长；(3)对加拿大10万多个湖泊的化学性质作出新的估计；(4)对旧图像实施新的大数据方法，绘制20世纪70年代加拿大土地利用/土地覆盖地图。在我的实验室创建的坚实原型的基础上，在新数据和强大的分析平台的推动下，这项研究可能会为研究人员和非科学家提供具体的好处。我们将更好地向加拿大人通报日益引起公众关注的野火；我们将窥探加拿大丰富的水生资源；我们将开发出我们预计会引起遥感界广泛兴趣的方法。

Recent developments in satellite remote sensing show that we have entered a new era. Where satellite information once quite scarce and expensive, vast archives have been opened for free public access. Where processing power and disk space once could throttle a plan for a large-area study, new tools with enormous capacity have emerged to put supercomputers at anyone's fingertips. Remote sensing's advances with this new paradigm have been rapid and impressive, producing global assessments of a wide range of phenomena: global tidal flat maps, high-resolution population estimates, times series maps of forest loss and gain-- to name only a few. Despite this enormous progress, considerable obstacles constrain our use of the vast potential of the world's earth-resources satellites. Combining data streams together from multiple platforms is quite complex: diverse sensors offer imagery with different spatial resolutions, have different numbers and specifications of spectral bands, arrive at irregular times, are often partially covered by clouds, and were designed for a wide range of purposes. It is exciting to conceive of these sensors as a single Earth-observing machine constantly grabbing views from different parts of the electromagnetic spectrum-but it is difficult to handle more than a few dimensions of this signal at a time.  There are two goals of this research. First, we will use our unique algorithms to blend information from a wide array of sensors, allowing us to view change and stability more rapidly and reliably. Second, we will integrate new machine-learning techniques to interpret older low-information imagery, allowing us to see how Canada looked near the beginning of the satellite era. Together these goals will more fully explore the immense trove of imagery available across Earth's surface, from the early 1970s until the very latest fires of this coming summer, on land and in water. Over the next five years my research team of pre-undergraduates, undergraduates, Master's students, PhD students, and postdocs will pursue four medium-term objectives: (1) Combine multiple data streams for weekly mapping of harvest, fire borders and other change across Canada; (2) track wildfire growth in near-real-time by fusing fire growth models with information streaming from multiple sensors; (3) create new estimations of the chemical properties of more than 100,000 lakes of Canada; and (4) implement new big-data approaches to old imagery to map Canada's land use / land cover in the 1970s. Built on solid prototypes created in my lab and fuelled by new data and powerful analysis platforms, this research is likely to provide concrete benefits to both researchers and non-scientists alike. We will better inform Canadians about the wildfires that increasingly capture the public's attention; we will peer into Canada's great aquatic resource; and we will develop approaches that we foresee will be of broad interest to the remote sensing community.