Biomass, harvestable area, and forest structure estimated from commercial timber inventories and remotely sensed imagery in southern Amazonia Ted R. Feldpausch a, * , Andrew J. McDonald a , Carlos A.M. Passos b , Johannes Lehmann c , Susan J. Riha a a Department of Earth and Atmospheric Sciences, Cornell University, 1126 Bradfield Hall, Ithaca, NY 14853, USA b Departamento de Engenharia Florestal, Universidade Federal de Mato Grosso, Cuiaba ´, Mato Grosso, Brazil c Department of Crop and Soil Sciences, Cornell University, Ithaca, NY 14853, USA Received 4 November 2005; received in revised form 2 June 2006; accepted 4 June 2006 Abstract The purpose of this study was to determine if spatially-explicit commercial timber inventories (CTI) could be used in conjunction with satellite imagery to improve timber assessments and forest biomass estimates in Amazonia. As part of a CTI, all commercial trees 45 cm DBH were measured and georeferenced in 3500 ha of a logging concession in NW Mato Grosso, Brazil. A scientific inventory was conducted of all trees and palms 10 cm DBH in 11.1 ha of this area. A total of >20,000 trees were sampled for both inventories. To characterize vegetation radiance and topographic features, regional LANDSAT TM and ASTER images were obtained. Using a stream network derived from the ASTER-based 30 m digital elevation model (DEM), a procedure was developed to predict areas excluded from logging based on reduced impact logging (RIL) criteria. A topographic index (TI) computed from the DEM was used to identify areas with similar hydrologic regimes and to distinguish upland and lowland areas. Some timber species were associated with convergent landscape positions (i.e., higher TI values). There were significant differences in timber density and aboveground biomass (AGB) in upland (6.0 stems ha 1 , 33 Mg ha 1 ) versus lowland (5.4 stems ha 1 , 29 Mg ha 1 ) areas. Upland and lowland, and timber and non-timber areas could be distinguished through single and principal component analysis of LANDSAT bands. However, radiance differences between areas with and without commercial timber on a sub-hectare scale were small, indicating LANDSAT images would have limited utility for assessing commercial timber distribution at this scale. Assuming a 50 m stream buffer, areas protected from logging ranged from 7% (third order streams and above) to 28% (first order and above) of the total area. There was a strong positive relationship between AGB based on the scientific inventory of all trees and from the commercial timber, indicating that the CTI could be used in conjunction with limited additional sampling to predict total AGB (276 Mg ha 1 ). The methods developed in this study could be useful for facilitating commercial inventory practices, understanding the relationship of tree species distribution to landscape features, and improving the novel use of CTIs to estimate AGB. # 2006 Elsevier B.V. All rights reserved. Keywords: Primary forest; Reduced impact logging (RIL); Selective logging; Biomass; GIS; LANDSAT; ASTER; PCA; Deforestation; Amazon; Brazil 1. Introduction Record rates of deforestation in 2004 in the frontier regions of Amazonia, and an increasing trend since 1990 (INPE, 2005), indicate a need for management alternatives for forest resources. Selective logging, a compromise between preserva- tion and complete deforestation, provides opportunities to sustain forest resources while encouraging economic develop- ment. Advances in selective logging through reduced impact logging (RIL) and certification of timber harvests can substantially reduce stand damage and carbon loss (Putz and Pinard, 1993; Johns et al., 1996; Pinard and Putz, 1996; Pereira et al., 2002; Feldpausch et al., 2005). Certification requires a commercial timber inventory (CTI) prior to initiating logging operations. Including the forest of this study, 17 Forest Stewardship Council (FSC) certified RIL operations are currently active in native forests in Amazonia, for a total of 12,812 km 2 (FSC, 2004). These georeferenced CTIs could potentially be used in combination with remotely sensed data and limited, more intensive in situ inventories to improve www.elsevier.com/locate/foreco Forest Ecology and Management 233 (2006) 121–132 * Corresponding author. Tel.: +1 607 255 1729; fax: +1 607 255 2644. E-mail address: trf2@cornell.edu (T.R. Feldpausch). 0378-1127/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.foreco.2006.06.016