Forest classification of Southeast Asia using NOAA AVHRR data

Abstract

Methodologies using the 1 km resolution data of the NOAA AVHRR instrument were developed for tropical forest spectral discrimination and mapping at a regional scale. Tropical Southeast Asia was selected as a case study using a multitemporal AVHRR data set of 1990–1992. This study documents first the relevance of AVHRR data to assess the extent of seasonal and dense forest and, moreover, reports on the derivation of a specific fragmented/disturbed forest class. A geographically dependent methodology is developed: For continental Southeast Asia, where generally good cloud-free images are available during the dry season and seasonal vegetation formations are present, multitemporal AVHRR mosaics were produced before the classification process. For insular Southeast Asia, which is particularly affected by the cloud cover and where only humid vegetation formations are present, a multitemporal set of single-date AVHRR images was first classified, and then the classifications were mosaicked together using a combination of two criteria (image quality and maximum occurrence). Unsupervised classifications using NDVI and Channel 3 radiance were processed in both cases. The classifications were labeled using the spectral typology derived from the spectral analysis of the dataset. For evaluation of the labeled classifications, comparisons were made at a regional level with a digital forest map database. Verification of the AVHRR class assignment was carried out locally using a few high spatial resolution satellite images. It highlights the sources of misclassification.