Data Fusion: Merging Classification Streams

Abstract

AbstractAs the ability to rapidly produce classifications of satellite images grows, it will be increasingly important to have algorithms that can shift through them to separate the signal from inevitable classification noise. The purpose of this chapter is to explore how to update classification time series by blending information from multiple classifications made from a wide variety of data sources. In this lab, we will explore how to update the classification time series of the Roosevelt River found in Fortin et al. (Remote Sens Environ 238, 2020). That time series began with the 1972 launch of Landsat 1, blending evidence from 10 sensors and more than 140 images to show the evolution of the area until 2016. How has it changed since 2016? What new tools and data streams might we tap to understand the land surface through time?