Positional uncertainty and quality assurance of digital elevation change detection (DECD)

Chang Li,Qi Meng,Dong Wei,Wenzhong Shi,Ming Hao

DOI: https://doi.org/10.48550/arXiv.2206.00765

2022-05-15

Abstract:Studies on rapid change detection of large area urgently need to be extended from 2D image to digital elevation model (DEM) due to the challenge of changes caused by disasters. This research investigates positional uncertainty of digital elevation change detection (DECD) caused by different degrees of DEM complexity and DEM misregistration. Unfortunately, using three-sigma rule (3{\sigma}R) for DECD is disturbed by accuracy of parameter estimation, which is affected by the outliers (i.e., varied DEM) from DEM differencing samples. Hence, to reduce the aforementioned uncertainty of DECD, we propose a new strategy of quality assurance, adaptively censored three-sigma rule (AC3{\sigma}R), in which with the samples censored, outliers of global DEM differencing samples outside the standard deviations of the mean calculated by moment estimation are iteratively removed. Compared with the 3{\sigma}R and censored three-sigma rule (C3{\sigma}R) that is similar to AC3{\sigma}R but without iteration for both simulation and real-world data experiments, the proposed global AC3{\sigma}R method always exhibits the highest accuracies of DECD in terms of both the overall accuracies 0.99967, 0.98740 and kappa coefficients 0.99598, 0.81803 respectively, and the strongest robustness with a large convergence interval [0, 0.30010] under the simulated maximum registration error and most complex terrain complexity conditions.

Geophysics