• Create a water only image by thresholding the band(s) which best separate the water from the land mass. This wavelength range corresponds to bands 6 & 9 of the ALI image. Otsu thresholding is performed on the image formed by taking the average of bands 6 & 9. This is illustrated in figure A-3.
• Cloud Shadows and interference is removed. Some sections of the image with cloud coverage, and other smaller water bodies which are not of interest are masked out. Done using morphological operations image fill and closing. This is illustrated in figure A-4.
• Unsupervised Image Classification is done. k-means clustering done with number of clusters set to 10. The number of clusters was chosen arbitrarily at this stage. The resulting maps are color coded for display. This is illustrated in figure A-5.
• Remove classes potentially affected by shallow water, shore effects, and submerged aquatic vegetation. These are not suitable for further processing. This is done by removing classes showing elevated activity in specific bands.
• Obtain the collected samples from Maryland DNR/other sources. Match the sample locations with image locations. Extract the selected spectra from the imagery.
•
Check the spectra for corrupted, shallow water and other anomalies. Remove the affected spectra for the set.
•Run multiple regression analysis based model equations. Identify Outliers samples which significantly deviate from the model and remove them. Repeat the regression analysis to obtain the regression coefficients.
•
Using the regression coefficients obtained to all the pixels in the image, to obtain a pixel level SDT/Chl-a map.
