Detecting settlement expansion in South Africa using a hyper-temporal SAR change detection approach

• SAR data have recently become significantly more readily available.
• Hyper-temporal SAR change detection is thus becoming more feasible.
• This paper presents a SAR Temporal Autocorrelation Change Detection framework.
• Results indicate improvement when compared to multi-temporal change detection.

Recent times have seen a significant increase in the amount of readily available SAR data, with many current and historic SAR data holdings now adopting an open distribution policy. As more regular SAR observations are becoming available, the use of a hyper-temporal SAR change detection framework (utilizing a stack of potentially hundreds of SAR images) is now becoming significantly more feasible. A relevant use case is the detection of new informal settlements in South Africa. Here, hyper-temporal change detection has been shown to be very effective but has been limited to coarse resolution optical satellite imagery only. In particular, it has been found that for optical data the Temporal Autocorrelation Change Detection (TACD) method is able to effectively detect the formation of new informal settlements using hyper-temporal MODIS time-series data. In this paper, the TACD is modified for the use of coarse resolution hyper-temporal SAR data for the detection of new informal settlements. It is shown that by using a hyper-temporal approach to detecting these new informal settlements, a higher overall accuracy was achievable when compared to standard bi-temporal change detection. A dataset of ENVISAT Advanced Synthetic Aperture Radar images over the study area was used to create a hyper-temporal time-series of backscatter values for each of the pixels in the study area. It was found that the proposed method achieved change detection accuracies of 87% at a false alarm rate of less than 1% with bi-temporal SAR change detection achieving a change detection accuracy of 70% at an approximate 1% false alarm rate.