Figure 1
High-level overview of the SAS workflow and corresponding interface components. The user begins with (1) hyperspectral image input, then (2) image pre-processing and filtering, followed by (3) spectral library input in the “Select input data” tab. In the second “Run SMASH” tab, the user (4) sets up a MESMA run by specifying parameters and then (5) performs SMASH and examines the results [10].
Figure 2
The interactive threshold masking method involves three new windows: one displaying the NDWI image, another that contains the histogram of pixel values within this image and allows the user to adjust the contrast by moving the red slider bars, and a smaller dialog box for specifying the minimum and maximum band threshold values used to produce the initial mask [10].
Figure 3
The masking tools in SAS allow the water body of interest to be isolated and spectral and spatial filters enhance the smoothness of the image [10].
Figure 4
SAS includes a spectral library viewer that allows the user to highlight specific endmembers in bold, or to remove them from the plot [10].
Figure 5
SAS provides a tool for calculating normalized spectral separability scores (NS3) that can help guide endmember selection [10].
Figure 6
One of the SMASH outputs is a classified map that illustrates the dominant cyanobacterial genera identified within each pixel of the image [10].
Figure 7
The distribution of dominant cyanobacterial genera can also be summarized via a histogram and a table listing the area and percentage of the water pixels identified in the masking step assigned to each endmember [10].
Figure 8
SAS can produce a map displaying the actual endmember fractions for a genus selected by the user, such as this example for Aphanizomenon in Upper Klamath Lake [10].
Figure 9
A false color composite can provide an effective means of visualizing the spatial distribution of and interactions between two selected endmembers, such as this representation of Aphanizomenon and Gloeotrichia in Upper Klamath Lake [10].
Figure 10
SAS can also produce a larger figure in which the endmember fractions for all endmembers, including water, are arranged as the tiles of a mosaic [10].
Figure 11
The RMSE calculated for each image pixel provides a convenient summary of the uncertainty associated with the MESMA output based on the mismatch between the observed and modeled spectral mixtures [10].
Figure 12
Averaging the RMSE values over all the pixels for which a particular taxon was identified as the dominant endmember can provide insight as to which types of cyanobacteria were not modeled well by the MESMA algorithm [10].
Figure 13
SAS provides a tool for conducting an analysis of the sensitivity of the MESMA output to the maximum RMSE constraint. The MESMA is repeated for, in this case, 20 different values of the maximum RMSE between 0.001 and 0.02, with the proportion of the image assigned to each endmember, or left unclassified, plotted as a function of the maximum RMSE [10].
Table 1
Output data products available for export from SAS. nem denotes the number of endmembers. See text for further details on image geo-referencing, file naming conventions, and the organization of the bands in the fraction mosaic image.
| SAS OUTPUT DATA PRODUCT | FILE FORMAT | NUMBER OF BANDS |
|---|---|---|
| Classified map | GeoTIFF | 1 |
| Class mask | GeoTIFF | 1 |
| Fraction image for selected endmember | GeoTIFF | 1 |
| Fraction color composite for two selected endmembers plus water | GeoTIFF | 3 |
| Fraction mosaic | GeoTIFF | nem + 1 |
| Root mean squared error (RMSE) image | GeoTIFF | 1 |
| Taxa distribution | ASCII csv | N/A |
| Endmember mean RMSE | ASCII csv | N/A |