Table 1.
List of spectral vegetation indices proposed for water detection.
Table 2.
Field data characteristics and cardinality of available spectra for each data set (#).
Table 3.
Study areas: satellite data available and selected location used to study the temporal series of NDSIs.
Figure 1.
Study areas and satellite data analysed for the four case studies (ITA, IND KHM, VNM).
Black and red boxes indicate, respectively, the MODIS tile (H18v04,H25v07 and H27v07) and Landsat (194-28/29, 125-53, 142-48 and 127-51) extent. Black lines indicate the analysed area for which a Landsat – RGB 543- and MODIS-RBG 621– colour composite is provided.
Figure 2.
Comparison of different surface responses in different spectral ranges: full range 350–2500 (a), visible [0.35–0.75] (b), near infrared [0.75–1.35] (b) and shortwave infrared [1.4–2.5] (b).
The position of MODIS spectral response function of bands 1–7 is overlayed on the field data.
Figure 3.
Behaviour of spectral indices (NDSIs) in different paddy rice conditions from bare, dry soil to dense vegetation.
Table 4.
Results of ANOVA between flood condition (W) and other classes: wet soil (WS), dry soil (DS), sparse canopy (V1) and full canopy cover vegetation (V2) and post-hoc Tukey test.
Table 5.
Separability (S) between flood condition and other classes: wet soil (WS), dry soil (DS), sparse canopy (V1) and full canopy cover vegetation (V2).
Table 6.
Separability (S) score between pure water MODIS pixels and other LCs (soil and vegetation) for the different NDSIs and combined index at all test sites.
Figure 4.
Boxplots of analysed indices for pure water pixels (blue), pure soil pixels (orange) and pure vegetation pixels (green).
Every boxplot has the same cardinality. The vertical dotted lines show the calculated threshold. Panel (a) shows the NDSIs and Panel (b) the combined B2B6-EVI index.
Table 7.
Threshold values (T) for the NDSI/combined index and detection accuracy (%) on pure pixel validation data set.
Figure 5.
Correlation between the 11 NDSIs and water percentage in MODIS pixels.
Panel (a) shows the NDSIs and Panel (b) the combined B2B6-EVI index.
Table 8.
Regression parameters and performances (r2 and RMSE) for the NDSI/combined index vs water fraction relation.
Figure 6.
Noise equivalent (see Eq. 4) as a function of water percentage for the 11 NDSIs and the combined B2B6-EVI index.
Figure 7.
The upper section shows high resolution LC maps (blue: Water; green: vegetation; orange: bare soil; white: clouds) from Landsat analysis and water maps from MODIS for VIS/SVIR indices using global thresholds at the four test sites.
The lower section shows the performance of VIS/SWIR, VIS/NIR, NIR/SWIR indices and the combined B2B6-EVI index in the Omission/Commission Error space. Pareto boundaries in the Omission/Commission Error space for each site are provided.
Table 9.
Overall Accuracy (%) of water mapping for the 11 NDSIs/combined index at the four test sites (IND, ITA, VNM, KHM).
Figure 8.
NDSI and combined B2B6-EVI index temporal series for 2000–2003 for four selected rice cultivation areas in Italy (ITA), India (IND), Cambodia (KHM) and Vietnam (VNM).
For each site, the top panel shows smoothed profiles of EVI, B1B7, B1B6, B4B7 and B2B6-EVI with black, red, dark green, orange and purple continuous lines respectively. Horizontal lines show the threshold for each index. Grey points indicate cloud contamination, black stars indicate dates of Landsat image acquisition. For each site, the bottom panel shows the flood detection by each index for each MODIS composite date.