Monitoring a fuzzy object: The case of Lake Naivasha

This study shows two approaches to including uncertainty of the mapped feature in multi-temporal analysis. This is demonstrated on a series of Landsat ETM+ images of Lake Naivasha, Kenya, with fuzzy boundaries resulting from marshes and floating vegetation. The first approach creates image segments, merges these to image objects through object-based classification and calculates the uncertainty for the lake image object in each image. The second approach uses a soft classifier to calculate memberships for lake and land. The lake area is calculated for 6 different thresholds on membership for each “lake” membership image, reflecting thresholds on the uncertainty in the estimate. The method based on image objects and attached uncertainty provided a quick overview and highlights uncertainty related to image quality and time of observation. The method based on thresholding of membership gave more spatial detail, highlighting the effect of fuzzy boundaries.