Efficient Updating of Land-Cover Classifications in a Data-Rich Environment

English Abstract:

What was the land cover of Las Vegas, Nevada on July 10, 2014? In the view of Landsat 5, 7, and 8, how much wetland area has been lost or gained annually on the Gulf Coast since Hurricane Katrina? Given observations from both MODIS and Landsat, where has forest in Bolivia been converted to non-treed land cover since 1984?

For many scientists interested in using interpretations of remote-sensing data such as land-cover classifications, questions like these are enticing but remarkably difficult to answer given the current state of the art for interpreting remote sensing image sequences. We developed a new algorithm designed for the continuous updating of land-cover classifications through time in large data sets. The algorithm ingests classified land-cover data from any of the wide variety of earth-resources sensors; it maintains a running estimate of land-cover probabilities and the most probable class at all time points along a sequence of events. We tested this algorithm in central Quebec with Landsat 8 data for summer 2013, when a series of large fires erupted and burned through forests. Across 10 images of widely varying quality, our algorithm detected the fires, ignored contaminated image information, and mapped fire borders consistently throughout the summer. The sequence of land-cover maps through time had producer’s and user’s accuracies that were better than many of the component days, which were often contaminated by fire haze and clouds but nevertheless contained useable information. As we leave remote sensing’s data-poor era and enter a period with multiple looks at Earth’s surface from multiple sensors over a short period of time, this algorithm may help to sift through images of varying quality to extract the most useful information for mapping.