Show simple item record

dc.contributor.authorMeng, Qingmin
dc.date.accessioned2014-03-04T02:28:08Z
dc.date.available2014-03-04T02:28:08Z
dc.date.issued2006-12
dc.identifier.othermeng_qingmin_200612_phd
dc.identifier.urihttp://purl.galileo.usg.edu/uga_etd/meng_qingmin_200612_phd
dc.identifier.urihttp://hdl.handle.net/10724/23687
dc.description.abstractThe main objective of forest inventory is to acquire and maintain accurate and up-to-date forest information. Updating forest inventory information also is an important aspect of land use dynamics. In the process of large area forest inventory, the development and application of suitable technologies to estimate forest variables with fine spatial resolution are important for natural resources management and characterizing land use dynamics. Although ground inventory often has higher accuracy, it has two obvious disadvantages, i.e., time consuming and expensive. Combining geographic information systems (GIS), remote sensing, geospatial statistics, and ground inventory data, I develop and apply two up-to-date forest inventory approaches with fine spatial resolution (i.e., a 25-meter cell size) for the state of Georgia. One is a systematic geostatistical approach using remote sensing imagery for prediction. I develop this systematic approach including spatial/aspatial data exploration, semivariogram modeling, and kriging. Four typical kriging methods (i.e., ordinary kriging, universal kriging, Cokriging, and regression kriging) are compared and evaluated for spatially forecasting forest variables. Regression kriging is tested as the best kriging method. The second approach is the popular K nearest neighbor method. I explored and improved two disadvantages (i.e., the selection of K and computation cost) of K nearest neighbor method before using it to estimate forest variables. Another two important aspects of the K nearest neighbor method (i.e., the distance metrics and weight schemes) also are explored and discussed to improve forecast performance. Next, a weighted K nearest neighbor method to forecast the volume of trees for the whole state of Georgia with a 25- meter cell size using 12 scenes of Landsat TM imagery as auxiliary data was used. Forecast evaluation conducted using 10,000 random sample pixels outside the training dataset and the mean estimations of volume compared with the results from US Forest Service indicate that the estimations from this research are reasonable. These estimations also are compatible with other studies for large area forest inventory. I believe the remote sensing based geostatistical modeling and weighted K nearest neighbor are efficient approaches to studying other aspects of land use dynamics and natural resources management.
dc.languageeng
dc.publisheruga
dc.rightspublic
dc.subjectUp-to-date forest inventory
dc.subjectFine spatial resolution
dc.subjectRemote sensing
dc.subjectGIS
dc.subjectGeostatistics
dc.subjectWeighted K nearest neighbor method
dc.subjectHardwood volume
dc.subjectSoftwood volume.
dc.titleFine spatial resolution forest inventory for Georgia
dc.title.alternativeremote sensing based geostatistical modeling and K nearest neighbor method
dc.typeDissertation
dc.description.degreePhD
dc.description.departmentForest Resources
dc.description.majorForest Resources
dc.description.advisorChris J. Cieszewski
dc.description.committeeChris J. Cieszewski
dc.description.committeeMike R. Strub
dc.description.committeeBarry D. Shiver
dc.description.committeeMarguerite Madden
dc.description.committeeBruce E. Borders


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record