International Journal of Remote Sensing Vol. 32, No. 24, 20 December 2011, 8935–8948 Classifying tropical deciduous vegetation: a comparison of multiple approaches in Popa Mountain Park, Myanmar NAING ZAW HTUN*†, NOBUYA MIZOUE‡ and SHIGEJIRO YOSHIDA‡ †Graduate School of Bioresource and Bioenvironmental Sciences, Laboratory of Forest Management, Kyushu University, 6-10-1 Hakozaki, Higashi-ku, Fukuoka 812-8581, Japan ‡Faculty ofAgriculture, Kyushu University, 6-10-1 Hakozaki, Higashi-ku, Fukuoka 812-8581, Japan (Received 12 August 2009; in final form 9 October 2010) Although several studies have reported that rule-based methods are better than other image classification methods, no study has quantified their performance for tropical deciduous vegetation classification. We compared rule-based and maxi- mum likelihood classification (MLC) approaches in classifying tropical deciduous vegetation in Popa Mountain Park, Myanmar. Classification was primarily based on Thematic Mapper (TM) bands of multi-season Landsat images, normalized difference vegetation indices (NDVIs), NDVI differences, mean NDVI and ele- vation (advanced spaceborne thermal emission and reflection radiometer digital elevation model (Aster DEM)). We used two main approaches for classification, a single-step approach in which all vegetation types were classified in one procedure, and a two-step approach in which forest and non-forest were discriminated first and then forest was classified into additional classes. Each of those approaches was conducted with and without elevation under the rule-based and MLC approaches, yielding eight separate methods. The two-step approaches generated more accurate results and all classifications improved markedly when elevation was included. The rule-based two-step with elevation approach produced the best overall accuracy and reliability. 1. Introduction Vegetation type maps are essential tools for managing natural terrestrial ecosystems for multiple purposes, especially biodiversity conservation (Trisurat et al . Make et al . 2000, Dymond et al . 2002, Brown de Colstoun et al . 2003, Joy et al . 2003, Singh et al . 2005, Xie et al . 2008). Remotely sensed data are widely used in forest type classification owing to the reduced cost and interpretation time compared with aerial photographs (Ismail and Jusoff 2008). Traditional methods of analysing satellite images for land cover classification include unsupervised algorithms, such as k-means and the itera- tive self-organizing data analysis technique (ISODATA), and supervised classification methods such as maximum likelihood classification (MLC) (Brown de Colstoun et al . 2003, Xie et al . 2008). However, because different vegetation types may have simi- lar spectral values, these traditional methods of vegetation classification are not very *Corresponding author. Email: nzhtun@gmail.com International Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online © 2011 Taylor & Francis http://www.tandf.co.uk/journals http://dx.doi.org/10.1080/01431161.2010.531779