Quantifying Analyst Bias in Mapping Flooded Areas from Landsat Images
presentationposted on 19.06.2017, 00:00 authored by Caleb McCarragher
Satellite images are classified by image analysts following disasters such as flooding, landslides, and wildfires. Results of the image classification process will be impacted by analyst bias which is unavoidable. If the results of the classification vary too much due to analyst bias, the derived map products might not be useful for post-disaster management. The goal of this study was to quantify the amount of bias introduced by the analysts while classifying the flooded areas in six Landsat images. In this study, six analysts classified a set of four post-flood Landsat images. Each analyst had the flexibility to select the number of clusters along with related parameters while assigning the pixels in each image. Finally there were required to group the pixels in each image and assign them to water (flooded areas) or non-water classes. Variations in the area of flooded areas identified by the six analysts for each image was computed to quantify the extent of analyst bias along with sources of variation in each image.