Urban Growth Detection Using Filtered Landsat Dense Time Trajectory in an Arid City

Wednesday, 17 December 2014
Zhiwei Ye and Annemarie Schneider, University of Wisconsin Madison, Center for Sustainability and the Global Environment, Madison, WI, United States
Among all remote sensing environment monitoring techniques, time series analysis of biophysical index is drawing increasing attention. Although many of them studied forest disturbance and land cover change detection, few focused on urban growth mapping at medium spatial resolution. As Landsat archive becomes open accessible, methods using Landsat time-series imagery to detect urban growth is possible.

It is found that a time trajectory from a newly developed urban area shows a dramatic drop of vegetation index. This enable the utilization of time trajectory analysis to distinguish impervious surface and crop land that has a different temporal biophysical pattern. Also, the time of change can be estimated, yet many challenges remain. Landsat data has lower temporal resolution, which may be worse when cloud-contaminated pixels and SLC-off effect exist. It is difficult to tease apart intra-annual, inter-annual, and land cover difference in a time series.

Here, several methods of time trajectory analysis are utilized and compared to find a computationally efficient and accurate way on urban growth detection. A case study city, Ankara, Turkey is chosen for its arid climate and various landscape distributions. For preliminary research, Landsat TM and ETM+ scenes from 1998 to 2002 are chosen. NDVI, EVI, and SAVI are selected as research biophysical indices. The procedure starts with a seasonality filtering. Only areas with seasonality need to be filtered so as to decompose seasonality and extract overall trend. Harmonic transform, wavelet transform, and a pre-defined bell shape filter are used to estimate the overall trend in the time trajectory for each pixel. The point with significant drop in the trajectory is tagged as change point. After an urban change is detected, forward and backward checking is undertaken to make sure it is really new urban expansion other than short time crop fallow or forest disturbance.

The method proposed here can capture most of the urban growth during research time period, although the accuracy of time point determination is a bit lower than this. Results from several biophysical indices and filtering methods are similar. Some fallows and bare lands in arid area are easily confused with urban impervious surface.