Comparing Landsat water index methods for automated water classification in eastern Australia

Three decades of Landsat-derived spring surface water dynamics in an agricultural wetland mosaic; Implications for migratory shorebirds

“…These water indices have variable performance depending on scene conditions and study area extent (Fisher et al, 2016;Yang et al, 2015); thresholds also vary in time and space (Friedl and Brodley, 1997). Thorough comparisons of multiple indices have been limited to a handful of studies (Campos et al, 2012;Fisher et al, 2016;Ji et al, 2009) with no clear best index for global application.…”

“…Thorough comparisons of multiple indices have been limited to a handful of studies (Campos et al, 2012;Fisher et al, 2016;Ji et al, 2009) with no clear best index for global application. Previous research indicates that thresholds should be modified for local conditions (Feyisa et al, 2014;Xu, 2006), yet few studies provide guidance for appropriate threshold optimization methods.…”

“…Previous research indicates that thresholds should be modified for local conditions (Feyisa et al, 2014;Xu, 2006), yet few studies provide guidance for appropriate threshold optimization methods. A recent study by Campos et al, 2012 used the mean and standard deviation of values across 12 monthly images to select a threshold, while others have used omission and commission error as the basis of threshold selection (Fisher et al, 2016;Yang et al, 2015). Sheng et al (2016) used iterative segmentation to identify NDWI thresholds for mapping lakes in Oceania.…”

“…Ji et al (2009) used spectral unmixing of pure pixels derived in a lab setting to suggest default thresholds for different fractional water coverages (starting points for 25, 50, 75, 100% water cover). Some of these studies remain untested outside of lab conditions (Ji et al, 2009), while others incorporate more field validation points than is normally practical (Fisher et al, 2016). Most studies do not explicitly evaluate accuracy outside of the source regions and dates used to train water classifications (Campos et al, 2012;Feyisa et al, 2014;Fisher et al, 2016;Sheng et al, 2016), which is of particular import for time series analysis.…”

“…Some of these studies remain untested outside of lab conditions (Ji et al, 2009), while others incorporate more field validation points than is normally practical (Fisher et al, 2016). Most studies do not explicitly evaluate accuracy outside of the source regions and dates used to train water classifications (Campos et al, 2012;Feyisa et al, 2014;Fisher et al, 2016;Sheng et al, 2016), which is of particular import for time series analysis. This study used the entire Landsat thematic mapper time series to evaluate water distribution patterns during spring shorebird migration in the Sacramento Valley of California, a globally important wetland stopover (Western Hemisphere Shorebird Reserve Network, 2009).…”