Sensitivity analysis of sub-pixel correlation technique for measuring coseismic displacements using a pair of ASTER images

Abstract

Coseismic displacements play a key role in understanding earthquake dynamics. To derive displacement fields from optical and microwave remote sensing datasets, various methods are available. This study evaluated in detail the offset tracking technique on optical ASTER data for 2005 Kashmir earthquake. This technique required input parameters like resampling methods, correlator types, window sizes and step sizes. For accurate displacement field calculation, careful selection of these parameters is imperative which depends on the study area and dataset characteristics. In the study, we made relative comparisons of coseismic displacement fields calculated by using different combinations of input parameters. The results were validated by field based displacement data of vertical separation. Validation was based on the hypothesis that horizontal displacement component may also have vertical component contribution depending upon local characteristics of the fault. Validation results showed that general trend of the measured displacements was in agreement with the field data. Field measurements were bounded within the uncertainty limits of the technique however at some locations significant deviations were also observed. All the coseismic displacement results obtained by using different input parameter were within the uncertainty limit ±1/10 of the pixel size, except for window size 4×4 and 8×8. The measured component of the fault rupture for northwest of Muzaffarabad is irregular. It may be due to rugged topography as compared to southeast part. The measured fault rupture also coincided to surface rupture mapped in the field. Analysis of the results showed that in comparison to standard parameter set, defined in the literature (Sinc resampling method, Frequential correlator with window size 32×32 and step size 8), selection of resampling method and correlator type had no significant effect on the calculated displacement field. However, window size and step size were found to be the most important parameters of the technique. This study showed that changing the window size does not affect the detection limit of the technique but induces bias in the calculations. Step size is used in relation to window size, however relatively larger step size is found to provide better sensitivity and accuracy of the technique.