An object-based change detection method accounting for temporal dependences in time series with medium to coarse spatial resolution

• Published in: Remote sensing of environment Vol. 112; no. 6; pp. 3181 - 3191
• Main Authors: Bontemps, Sophie, Bogaert, Patrick, Titeux, Nicolas, Defourny, Pierre
• Format: Journal Article
• Language: English
• Published: New York, NY Elsevier Inc 16.06.2008; Elsevier
• Subjects: Animal, plant and microbial ecology; Applied geophysics; Biological and medical sciences; Change detection; Earth sciences; Earth, ocean, space; Exact sciences and technology; Fundamental and applied biological sciences. Psychology; General aspects. Techniques; Internal geophysics; Mahalanobis distance; Multi-temporal segmentation; Probabilistic change thresholds; Teledetection and vegetation maps; Temporal dependences; South America; Brazil; Brazil, Rondonia; Temporal dependences; Probabilistic change thresholds; Multi-temporal segmentation; Change detection; Mahalanobis distance; time series analysis; Tracking; Probabilistic approach; object oriented; remote sensing; human activity; Change detection;Multi-temporal segmentation;Temporal dependences;Mahalanobis distance;Probabilistic change thresholds; evaluation; automation; Limitation; Dependence; imagery; Environment; Procedure; spatial resolution; segmentation; Plant cover; Multidate observation; land cover; Anthropogenic factor
• ISSN: 0034-4257; 1879-0704
• DOI: 10.1016/j.rse.2008.03.013

Tracking land cover changes using remotely-sensed data contributes to evaluating to what extent human activities impact the environment. Recent studies have pointed out some limitations of single-date comparisons between years and have emphasized the usefulness of time series. However, less effort has hitherto been dedicated to properly account for the temporal dependences typifying the successive images of a time series. An automated change detection method based on a per-object approach and on a probabilistic procedure is proposed here to better cope with this issue. This innovative procedure is applied to a tropical forest environment using high temporal resolution SPOT-VEGETATION time series from 2001 and 2004 in the Brazilian state of Rondônia. The principle of the method is to identify the objects that most deviate from an unchanged reference defined by objective rules. A probabilistic changed-unchanged threshold provides a change map where each object is associated with a likelihood of having changed. This improvement on a binary diagnostic makes the method relevant to meet the requirements of different users, ranging from a comprehensive detection of changes to a detection of the most dramatic changes. According to the threshold value, overall accuracy indices of up to 91% were obtained, with errors involving change omissions for the most part. The isolation of changes within objects was made possible through a segmentation procedure implemented in a temporal context. In addition, the method was formulated so as to differentiate between inter- and intra-annual vegetation dynamics. These technical peculiarities will likely make this analytical framework suitable for detecting changes in environments subject to a strongly marked phenology.