Extended Data Fig. 2: Example of bimodal histograms used to calculate adaptive thresholds for water classifications that approximate, on average, the standard versions of the water classification thresholds.
From: Satellite imaging reveals increased proportion of population exposed to floods
a, b, Example bimodal histograms (left axes) with interclass variance (ICV; blue lines, right axes) extracted from MODIS imagery used to determine optimal thresholds for B2B1ratio (K1; a) and DNSWIR (K2; b). The dashed red and black lines reflect the estimated Otsu and standard thresholds, respectively. c, d, Distribution of estimated Otsu thresholds calculated for each flood event across the Global Flood Database (n = 913), for B2B1ratio (K1; c) and DNSWIR (K2; d). The average Otsu threshold across the Global Flood Database for B2B1ratio (K1 = 0.77; dashed red line in c) and DNSWIR (K2 = 599; dashed red line in d) are comparable to the standard thresholds (K1 = 0.70, dashed black line in a; K2 = 675, dashed black line in b).