Land transitions from multivariate time series: using seasonal trend analysis and segmentation to detect land-cover changes


The detection of land-cover change over large areas using short time series is a challenging and important task in global change studies. This paper introduces a novel method, called the Segment-based Detection of Trends and Change (SDTC), to determine areas that are undergoing land-cover changes. The method is illustrated in the State of Alaska over the 2001–2009 time period using three time series derived from Moderate Resolution Spectroradiometer (MODIS) imagery: the normalized difference vegetation index (NDVI), albedo, and land surface temperature (LST). SDTC extends seasonal trend analysis (STA) by using segmentation in the detection process to combine multiple variables and to incorporate the spatial context. Segments labelled as ‘change’ correspond to groups of adjacent pixels with a majority of pixels undergoing significant temporal trends as defined by the Mann–Kendall (MK) procedure and STA. Segments correspond to landscape units in a less arbitrary manner than pixels because they represent groupings of adjacent areas undergoing similar temporal behaviour. Findings indicate that the use of MK in conjunction with segmentation exploits more fully the richness of the spatial context in the process of detection. Results suggest that SDTC is a useful method for detecting and characterizing land-cover change. The technique is conservative and eliminates the ‘salt and pepper’ effect by filtering noise and identifying areas of change. Using SDTC, we found that most areas of change in Alaska undergo between one and three significant changes and that increasing LST seasonality constitutes the most widespread type of change.

In International Journal of Remote Sensing