Dissertations and Theses @ UNI

Award Winner

Recipient of the 2008 Outstanding Master's Thesis Award - First Place.

To go to the Graduate Student Award Recipients collection page, click here.


Open Access Thesis


Trees--Iowa--Cedar Falls--Identification; Optical radar--Iowa--Cedar Falls; Remote sensing--Iowa--Cedar Falls;


This thesis explores a data fusion approach combining hyperspectral, LiDAR, and multispectral data to classify tree species in an urban environment. The study area is the campus of the University of Northern Iowa.

In order to use the data fusion approach, a wide variety of data was incorporated into the classification. These data include: a four-band Quickbird image from April 2003 with 0.6m spatial resolution, a 24-band AISA hyperspectral image from July 2004 with 2m spatial resolution, a 63-band AISA Eagle hyperspectral image from October 2006 with lm spatial resolution, a high resolution, multiple return LiDAR data set from April 2006 with sub-meter posting density, spectrometer data gathered in the field, and a database containing the location and type of every tree in the study area.

The elevation data provided by the LiDAR was fused with the imagery in eCognition Professional. The LiDAR data was used to refine class rules by defining trees as objects with elevation greater than 3 meters. Classes included honey locust, white pine, crab apple, sugar maple, white spruce, American basswood, pin oak and ash.

Results indicate fusing LiDAR data with these imageries showed an increase in overall classification accuracy for all datasets. Overall classification accuracy with the October 2006 hyperspectral data and LiDAR was 93%. Increases in overall accuracy ranged from 12 to 24% over classifications based on spectral imagery alone. Further, in this study, hyperspectral data with higher spatial resolution provided increased classification accuracy.

The limitations of the study included a LiDAR data set that was acquired slightly before the leaves had matured. This affected the shape and extent of these trees based on their LiDAR returns. The July 2004 hyperspectral data set was difficult to georectify with its 2m resolution. This may have resulted in some minor issues of alignment between the LiDAR and the July 2004 hyperspectral data.

Future directions of the study include developing a classification scheme using a Classification And Regression Tree, utilizing all of the LiDAR returns in a classification instead of just the first and fourth returns, and examining an additional LiDAR-derived data set with estimated tree locations.

Year of Submission


Year of Award

2008 Award

Degree Name

Master of Arts


Department of Geography

First Advisor

Ramanathan Sugumaran, Chair, Thesis Committee


If you are the rightful copyright holder of this thesis and wish to have it removed from the Open Access Collection, please submit an email request to scholarworks@uni.edu. Include your name and clearly identify the thesis by full title and author as shown on the work.

Date Original


Object Description

1 PDF file (viii, 98 pages)



File Format