SPECIAL SECTION BIOTROPICA 37(4): 508–519 2005 10.1111/j.1744-7429.2005.00069.x The Utility of Spectral Indices from Landsat ETM+ for Measuring the Structure and Composition of Tropical Dry Forests 1 Kenneth J. Feeley 2 Department of Biology, Duke University, Durham NC 27708 Thomas W. Gillespie Department of Geography, University of California–Los Angeles, Los Angeles, CA 90095-1524 and John W. Terborgh Nicholas School of the Environment, Duke University, Durham NC 27708 ABSTRACT There is a growing emphasis on developing methods for quantifying the structure and composition of tropical forests that can be applied over large landscapes, especially for tropical dry forests that are severely fragmented and have a high conservation priority. This study investigates the relationships between various measures of forest structure (annual woody increment, canopy closure, stand density, stand basal area) and composition (tree species diversity, tree community composition) measured in semi-deciduous tropical dry forests on islands in Lago Guri, Venezuela and three spectral indices derived from Landsat ETM+: Normalized Difference Vegetation Index (NDVI), Infrared Index (IRI), and Mid-Infrared Index (MIRI). Even though there were significant autocorrelations among spectral indices, there were significant differences in the relationships between spectral indices and forest attributes. IRI was not significantly correlated with any of the structural variables while MIRI was correlated with canopy closure and NDVI values were correlated with canopy closure as well as annual woody increment. NDVI and MIRI were both related to relative tree diversity and all three indices were associated with aspects of tree species composition. Based on the results of this study, it appears that spectral indices, and in particular NDVI, may be useful indicators of forest attributes in tropical dry forest habitats. Further research needs to be undertaken to identify if the results of this study can be applied to other tropical dry forests at a global spatial scale. RESUMEN Existe un inter´ es cada vez mayor en desarrollar m´ etodos para cuantificar la estructura y composici´ on de los bosques tropicales que se puedan aplicar a nivel de paisajes grandes, especialmente para los bosques secos tropicales que est´ an severamente fragmentados y tienen una alta prioridad para su conservaci´ on. Este estudio investiga las relaciones entre las diferentes medidas de la estructura forestal (crecimiento anual, densidad de la copa, densidad de ´ arboles, y ´ area basal) y la composici´ on (diversidad de especies de ´ arboles y composici´ on de la comunidad de ´ arboles) en bosques secos tropicales semi-deciduos en las islas en el Lago Guri, Venezuela, con tres ´ ındices espectrales provenientes de Landsat ETM+: Normalized Difference Vegetation Index (NDVI), Infrared Index (IRI), and Mid-Infrared Index (MIRI). Aunque se obtuvieron autocorrelaciones significativas entre los ´ ındices espectrales, tambi´ en se obtuvo diferencias significativas en las capacidades de proyecci´ on de cada ´ ındice para las cualidades del bosque que se midieron. IRI no fue correlacionado con las variables estructurales, pero MIRI fue correlacionado con, y los valores de NDVI eran correlacionados con la densidad de la copa y tambi´ en el crecimiento anual de los ´ arboles. NDVI y MIRI fueron relacionados con la diversidad relativa de especias y los tres ´ ındices fueron asociados a aspectos de la composici´ on forestal. De acuerdo con los resultados de este estudio, sugerimos que los ´ ındices espectrales y en particular el NDVI, pueden ser indicadores ´ utiles de las caracter´ ısticas generales de los bosques secos tropicales. Se necesitar´ a llevar a cabo mayores investigaciones para determinar con precisi´ on si los resultados de este estudio pueden ser aplicados a los bosques secos tropicales en una escala espacial global. Key words: IRI; Lago Guri; Landsat; MIRI; NDVI; productivity; remote sensing; spectral indices; tree diversity; Venezuela. THE SUCCESSFUL DESIGN AND IMPLEMENTATION OF CONSERVATION STRATEGIES requires information about tropical dry forest fragments, yet field inventories of tropical forests are often very time consuming and costly due to difficult access and the large number of difficult to identify tree species (Ruokolainen et al. 1997, Gillespie et al. 2004). Therefore, there is an increasing emphasis on developing methods of rapidly characterizing the structure, diversity, and composition of these forests using remote sensing techniques (Innes & Koch 1998; 1 Received 15 October 2004; revision accepted 10 April 2005. 2 Corresponding author. Current address: Center for Tropical Forest Science- AA, Harvard University Herbarium, 22 Divinity Ave, Cambridge, MA 02138; e-mail: kfeeley@oeb.harvard.edu Foody et al. 2001; Bawa et al. 2002; S´ anchez-Azofeifa et al. 2003, 2005; Turner et al. 2003). Since the mid-1960s, a number of multispectral spaceborne sensors have been used to quantify various biophysical attributes associated with tropical forests. The Landsat satellite series, initially launched in 1972, provided scientists with a unique tool to measure and monitor tropical forests at a high spatial and spectral resolu- tion (Jensen 2000). The most recent satellite in the Landsat series, Landsat 7 Enhanced Thematic Mapper Plus (ETM+) contains six spectral bands with a spatial resolution of 30 × 30 m, one panchro- matic band (15 × 15 m), and one thermal band (60 × 60 m) and has been used to successfully map the extent of tropical forests (Rey- Benayas & Pope 1995, Gao 1999, Hill 1999, Southworth 2004). 508