Detection of new settlements using 250 m MODIS time-series data in the Gauteng province of South Africa

Abstract

Human settlement expansion is one of the most prominent types of land cover change in South Africa. These changes typically occur in areas that are covered by natural vegetation. It is important for regional and local government to detect and map these settlements regularly (at least every two years) to support spatial planning but this is often not achievable due to constraints in cost and resources related to manual updating of maps. Methods that can rapidly indicate areas having a high probability of change are very valuable to analysts as this can be used to direct their attention to high probability change areas for further evaluation using, for example, higher resolution imagery of the area. MODIS time-series data (8-daily composite) at a resolution of 500 m has been proven to be an effective data source for detecting these human settlements in South Africa using a Temporal Autocorrelation Change detection method (TACD). In this paper, this method is adapted to be usable with variable sampled temporal resolutions by using a novel framework for parameter selection and a comparison of change detection accuracy vs. false alarm rate is done between 8-daily 500 m and daily 250 m MODIS data. A critical evaluation of the performance of the TACD method was also done as a function of the temporal resolution showing little difference in performance between daily sampled and 2-monthly sampled time-series data for the use case evaluated in this paper.