An evaluation of the effect of terrain normalization on classification accuracy of Landsat ETM+ imagery.

Abstract

More than 60% of land in New Zealand has been converted from native forests to residential areas, agriculture, or forest plantations. Settlers brought many species of plants and animals to New Zealand. Many native species were unable to protect themselves from these new predators, causing numerous extinctions. Due to this rapid decline in biodiversity, the New Zealand government has made it a priority to halt this loss. Restoration of developed land and protection of remaining areas of native forest are two important ways to mitigate the loss of biodiversity. Monitoring of restoration efforts is important to the government and organizations responsible for this work. Using remotely sensed data to perform change analysis is a powerful method for long-term monitoring of restoration areas. However, there is significant terrain variation within many of these areas that may significantly reduce land cover classification accuracy. Landcare Research New Zealand has developed a topographic suppression algorithm that reduces the effects of topography. Landsat ETM+ imagery from November 2000 was processed with this algorithm to produce two images, an orthorectified image and a terrain flattened image of a 50-km by 60-km area near Wanganui, New Zealand. Using GLOBE reference data collected on the ground in September/October 2004 and additional reference data photointerpreted from aerial photography, thematic maps were created using various classification methods. The accuracy of the thematic maps was evaluated using error matrices and the different image processing techniques were statistically compared. It was determined that the topographic algorithm did not significantly improve map accuracy.

Department

Natural Resources and the Environment

Publication Date

5-2-2006

Journal Title

Proceedings of the Annual Meeting of the American Society of Photogrammetry and Remote Sensing, Reno, NV.

Publisher

ASPRS 2006 Annual Conference

Document Type

Article

Share

COinS