Abstract
Landsat 7 Scan Line Corrector failure compromised 22% of pixels. Novel procedures were developed estimating missing data from means of pixels present within 2(N)-sized neighbouring blocks whose positions were defined by simultaneously minimizing heterogeneity of source imagery within blocks while satisfying need for sufficiently large size that some pixels remained near defects. Regions used to estimate missing data were defined in a stepwise process in which local standard error cut-offs for accepting particular 2(N)-sized regions of source imagery as representing missing pixels were progressively raised over a series of 51 cycles. For Landsat scenes of interest in western Oregon, averages of 34.7, 68.8, 95.4, and 99.0% of gaps in synthetic data were repaired by cycles 1, 6, 16, and 30. Standard deviations of differences between estimated and real values increased from an average of 9.0 digital numbers at cycle 1 to plateaus near 17.0 by cycle 25. This procedure performed well over a wide range of imagery dates, a consequence of its use of source imagery to identify homologous regions within landscapes rather than directly estimating pixels.
NSTL
Elsevier