Updating land cover automatically based on change detection using satellite images: case study of national forests in Southern California

Observing dynamic change patterns and higher-order complexities from remotely sensed images is warranted, but the main challenges include image inconsistency, plant phenological differences, weather variations, and difficulties of incorporating natural conditions into automatic image processing. In this study, we proposed a new algorithm and demonstrated it by producing 2002–2008 and 2010 land-cover maps in heterogeneous Southern California based on an existing 2009 land-cover map. The new algorithm improves the baseline land-cover map quality by discarding potential bad land-cover pixels and dividing each land-cover type into several subclasses. Time series Landsat images were used to detect changed and unchanged areas between baseline year and target year t. Subsequently, for each individual year t, each pixel that was identified as unchanged inherited the baseline classification. Otherwise, each pixel in the changed areas was classified by a similar surrogate majority classifier. The demonstration results in Southern California showed that the land-cover temporal pattern captured the observed successional stages of the ecosystem very well. The accuracy assessment had an overall classification accuracies ranging from 81% to 86% and overall kappa coefficients ranging from 0.79 to 0.83.