Using Advanced Remote Sensing Data Fusion Techniques for Studying Earth Surface Processes and Hazards: A Landslide Detection Case Study

Monday, 15 December 2014
David Hulslander, Exelis VIS, Boulder, CO, United States
A major problem in earth surface process and hazards research is we have little to no knowledge of precisely where and when the next significant event may occur. This makes it nearly impossible to set up adequate instrumentation and observation ahead of time. Furthermore, it is not practical to overcome this challenge by instrumenting and observing everywhere all the time. We can’t be everywhere and see everything. Remote sensing helps us to fill that gap with missions such as Landsat and WorldView2 offering regular global coverage. However, remote sensing systems for global monitoring have several inherent compromises. Tradeoffs must be made between data storage, processing capacity, spatial resolution, spectral resolution, and temporal resolution. Additionally, instruments and systems must be designed in advance and from a generalized standpoint to serve as many purposes as possible, often at the expense of high performance in specific tasks.

Because of these practical constraints, when using remote sensing data to study earth surface processes it is critical to maximize signal content or information obtained from all available data. Several approaches, including multi-temporal data fusion, multi-sensor data fusion, and fusion with derivative products such as band ratios or vegetation indices can help expand how much information can be extracted from remote sensing acquisitions. Fused dataset results contain more coherent information than the sum of their individual constituents. Examples using Landsat and WorldView2 data in this study show this added information makes it possible to map earth surface processes and events, such as the 2011 Cinque Terre landslides, in a more automated and repeatable fashion over larger areas than is possible with manual imagery analysis techniques and with greater chance of successful detection.