Research Papers

Relation of preclassification class likelihood to postclassification accuracy in an Australian multitemporal sequence of imagery

[+] Author Affiliations
Kim Lowell

Cooperative Research Centre for Spatial Information, P.O. Box 672, 204 Lygon Street/Level 5, Carlton, Victoria 3053, Australia

University of Melbourne, Department of Resource Management and Geography, Parkville, Victoria, Australia

J. Appl. Remote Sens. 8(1), 083555 (Sep 05, 2014). doi:10.1117/1.JRS.8.083555
History: Received June 26, 2014; Revised August 10, 2014; Accepted August 13, 2014
Text Size: A A A

Abstract.  Using 20 time slices of imagery from 1972 to 2010 (e.g., 1972, 1977, 1980, etc.), the relationship between the preclassification probability of a pixel being forest—p(Forest)—and its classification correctness (forest versus nonforest) was examined. Classification accuracy was determined by human interpretation of high resolution IKONOS and SPOT digital imagery. The study area comprised the approximately one-third of the Australian continent that is relatively densely populated—i.e., not the interior. Preclassification p(Forest) values for individual pixels averaged over three, nine, and all 20 time slices were considered. Postclassification lineages—e.g., “always forest, regeneration”—for the 10-year period (2000 to 2010) covered by nine time slices were also examined. Chi-square analysis showed that regardless of the time period over which p(Forest) was calculated, a statistically significant relationship exists between p(Forest) and the classification accuracy, and between the 2000 and 2010 lineage and classification accuracy. Pixels having an intermediate p(Forest) value (0.20 to 0.80) were more likely to be erroneously classified than pixels having higher (greater than 0.80) or lower (less than 0.20) p(Forest) values. Pixels with a 10-year lineage of “always forest” were most likely to be correctly classified while those having experienced land cover change (deforestation or regeneration) were the least accurately classified. Logistic regression models reinforced these findings, although the relationship between preclassification p(Forest) values and classification correctness was weak.

Figures in this Article
© 2014 Society of Photo-Optical Instrumentation Engineers

Citation

Kim Lowell
"Relation of preclassification class likelihood to postclassification accuracy in an Australian multitemporal sequence of imagery", J. Appl. Remote Sens. 8(1), 083555 (Sep 05, 2014). ; http://dx.doi.org/10.1117/1.JRS.8.083555


Access This Article
Sign in or Create a personal account to Buy this article ($20 for members, $25 for non-members).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Related Book Chapters

Topic Collections

Advertisement
  • Don't have an account?
  • Subscribe to the SPIE Digital Library
  • Create a FREE account to sign up for Digital Library content alerts and gain access to institutional subscriptions remotely.
Access This Article
Sign in or Create a personal account to Buy this article ($20 for members, $25 for non-members).
Access This Proceeding
Sign in or Create a personal account to Buy this article ($15 for members, $18 for non-members).
Access This Chapter

Access to SPIE eBooks is limited to subscribing institutions and is not available as part of a personal subscription. Print or electronic versions of individual SPIE books may be purchased via SPIE.org.