2000 ARC Projects

Nitrogen Stress Detection in Wheat Using Remote Sensing

A quarter-section center pivot in Minidoka, Idaho was used as the study site for this project. The field was divided up with 4 transects running the entire length. Each of the transects had a different rate of nitrogen applied to test an under and over applied plot against a normal rate. Imagery was collected, chlorophyll meter samples and tissue samples were taken, and a simple visual analysis was performed.

It was concluded that IKONOS imagery was indeed as accurate, and cost effective as ground based methods, but was not delivered fast enough for application. Another project is planned for 2001 to test the method by applying a mid season topdressing of nitrogen to improve both yield and protein.

Remote Sensing for Wetland Delineation around Utah's Great Salt Lake

Automated Geographic Reference Center
Final Report

IKONOS imagery was collected over the study area on 7/13/2000. Ancillary data layers and manual digitizing were used to isolate and remove simple known cover type classes including urban, commercial, industrial, and agriculture from the imagery. A Normalized Difference Vegetation Index (NDVI) was also used in the same manor for water bodies and mud flats. The remaining pixels were classified into 20 clusters using ERDAS Imagine's Iterative Self Organizing Data Analysis (ISODATA) clustering algorithm. Points collected in the field representing major land cover types were intersected with the clusters. Clusters with strong correlations toward unique land cover types were labeled and removed from the image. Again, the remaining confused pixels were then clusterbusted into finer groups and intersected with the points and identifiable areas were labeled and removed. In an attempt to label the remaining pixels, Imagine's Expert Classifier (developed for the second ARC/AGRC project) was used. All of the thematic layers were then recoded and mosaicked to create a complete final land cover map of the study area. After reviewing the results of this study, several conclusions regarding the data sets and methods were realized. Although some obvious cover types were easily identified by their spatial context, i.e., Medicago lupulina L. (Alfalfa) from Phragmites australis (Giant Reed Grass) and Mud Flats from rooftops, we were unable to spectrally discriminate between them. The inclusion of a middle infrared band to the satellite platform (with supplied coefficients) would have allowed for a tasseled cap transformation to be applied to the imagery solving some classification problems. Perhaps most evident to the problems encounter was the lack of and adequate number of representative field points. With a larger number of quality sample points, much of the variability between the land cover types would have been reduced allowing for finer and more precise delineations. The results of this research indicate that IKONOS imagery is a viable satellite platform for identifying wetland areas at our particular scale of operation.

Wetland Delineation via Automatic Feature Extraction (AFE)

Automated Geographic Reference Center
Automated Feature Extraction
Final Report

IKONOS imagery was collected over the study area on 7/13/2000. A combination of local statistical operations and enhancement techniques were applied to the imagery and output to a seven-band image file. Points representing major land cover types collected in the field were intersected with the seven-band image and those values were input into a decision tree model to try and identify unique patterns. The Knowledge Base was created using these outputs as the framework for the decision tree and spectral patterns obtained from the imagery using traditional photogrammetry methods. The results of the output classification revealed that while the Expert Classifier performed well when identifying land cover types sampled in the field and it over-classified areas lacking field samples. The ordered approach used by the Expert Classifier in determining how the pixels were to be classified seemed to degrade the functionality of this utility for our particular study. In addition, establishing a basis for generating confidence values seemed arbitrary and was not fully explained in any of the documentation. To fully evaluate the potential of ERDAS Imagine's Expert Classifier for classifying wetlands, an adequate number of representative field samples would be needed.

Wheat Baseline Project