The following pictures represent work in progress. All maps are preliminary and not to be used in any final reports. Please contact any of the following people with questions or comments.
Marilyn G. Hughes
Rutgers Cooperative Extension & Center for Remote Sensing
Peter V. Oudemans
Rutgers Blueberry & Cranberry Research Center,Chatsworth, NJ
Phone (609)-726-1590 x 20
The following images are color-IR aerial photographs of the Nadine test beds (except for 1951 which is panchromatic). All imagery is taken at a scale of 1:12000.
The following picture is a flow model computed from the
elevation file for Nadine bed3. The arrows indicate the direction
of flow. The underlying grid is the disease noted in the bed.
The actual values for disease need to be checked.
Flow from May 21, 1999
The following image is the result of a supervised classification
of the August, 2 1999 photograph using disease (presence/absence phytophthora)
data collected within the Nadine test bed in August, 1999. Due to
variation within the photograph, the image was preprocessed by sampling
sand pixels throughout and subtracting out the resulting trend from the
image. Disease/healthy/sand categories were digitized from a Krigged
surface interpolated from sampling 216 points throughout the bed.
The classification was applied to all 5 Nadine beds.
The following images compare using three different vegetative indices, the ndvi, savi, and msavi. The beds shown are black rock, nadine, and cedar swamp. The VIs were run in ERDAS IMAGINE on a subset of CIR aerial photography obtained on May, 1999. All DNs are float and rescaled for viewing. The formulae for the vegetative indices were obtained from Qi et. al. (1994) Remote Sens Environ 48:119-126. Differences in VIs for cranberry may not be great, as % canopy cover is great; however, in crops where canopy cover is less than 75-80% the ndvi is influenced by reflectance from the underlying soil.
The method of using an unsupervised classification for delineation of variability, disease, and poor drainage within the beds has been used with success before. The following images show a way for improving the classifications using ERDAS IMAGINE. The images represent the same area as above.
First Cut-->Unsupervised classification (10 classes)
Examining the raster attribute editor after the original unsupervised classification, it is apparent that class 1 refers to water and that classes 9, 10 refer to sand. These classes do not need to be further classified so they are masked out using the mask utility. Masking reuqires the user to recode the values to 0,1 0 indicating the excluded class. A final masked image is shown here. Masked image. The values in the image are 0 where the masked classes intersect the original image and the original image values everywhere else. The following shows an unsupervised classification into 15 classes of this image with the masked classes merged back into the file (using the overlay utility and the recode option in IMAGINE).
The following table compares a 10 class unsupervised classification with the 15 class classification (using masking) for the three sub-study areas. I left the images full size so that you can save them and read them into an image viewer.
|15 classes no masking||15 classes with masking|