Date of Award


Degree Name

Master of Arts



First Advisor

Dr. Charles Emerson

Second Advisor

Dr. Lisa DeChano

Third Advisor

Dr. James Biles

Access Setting

Masters Thesis-Open Access


In this study, Landsat imagery from June (summer) and October (autumn senescence) 2000 of the area of Fort Custer Training Center in Kalamazoo and Calhoun Counties, Michigan, were analyzed for forest classification accuracies. The use of fractal analysis to improve forest classification, particularly to distinguish among northern hardwood species, was examined. Using moving windows with different sizes, measurement of local fractal dimension and spatial autocorrelation (Moran's I) were performed on the NDVIs and the panchromatic images. The measurement products were combined with Landsat TM bands as additional layers in supervised maximum- Iikelihood multispectral classifications. The accuracy for multispectral classification of the October scene is 43.21 percent. The addition of the fractal dimension measurement on the October panchromatic image results in 45.06 percent accuracy, showing a slight improvement of 1.85 percent. The accuracy for the multispectral classification of the June scene is only 30.25 percent. However, the fractal dimension seems to work better on the June scene, since it gives a higher percentage of improvement. The addition of fractal dimension measurements on the June NDVI and panchromatic image gives 36.42 percent of accuracy, an improvement of 6.17 percent compared with using the spectral bands alone.

Included in

Geography Commons