Tree species identification on large-scale aerial photographs in a tropical rain forest, French Guiana - application for management and conservation

Management and conservation planning of any ecosystem requires knowledge of species composition. This is a real challenge in tropical rain forests that are characterised by very high species richness and canopy access limitations. The possibility of approaching trees from remote sensing on large-scale aerial photographs, takes on its full significance in this context. Results of tree species identification by photo-interpretation in a French Guianan forest canopy are discussed, as well as an overview of the part of the forest accessible from the photographs. Two sets of aerial photographs were used. One set (1:3700 colour slides) covers 15 ha of primary forest, divided into a training set (TS, 5 ha) and a validation set (VS 1: 10 ha). Another validation set, taken in different conditions of acquisition, scale and season, is available for an adjacent area (VS 2: 6.5 ha). Aerial photographs captured a quarter of the tree community (dbh >= 10 cm) on average, and about 45% of the SGS (Species or Group of Species) on the training set. The crown appearance of 12 major canopy SGS, including commercial species and species of ecological interest, had been described in a previous work on the same training set. Following these descriptions, two photo-interpreters separately identified 309 tree crowns overall on VS 1, with a good agreement in their respective judgements. After their interpretations were checked in the field, the overall average identification success was high (87%) but the results varied according to the SGS. The results on VS 2 showed that some species displayed major seasonal and scale variations and were hardly recognized, whereas some others could be identified without modifying the learning process. The results are encouraging and this work will be extended as the identification of tropical rain forest trees from remote sensing has many applications, ranging from fundamental ecological knowledge of canopy species to the management and conservation of such highly diverse and hardly inventoried ecosystems. (c) 2005 Elsevier B.V. All rights reserved.