69 Geocarto International, Vol. 14, No. 3, September 1999 Published by Geocarto International Centre, G.P.O. Box 4122, Hong Kong. Vegetation Identification and Biomass Estimation Using AIRSAR Data David Pairman, Stephen McNeill, Neal Scott and Stella Belliss Landcare Research, PO Box 69, Lincoln 8152, New Zealand Abstract Monitoring of vegetation biomass is becoming more important as countries attempt to preserve major carbon reservoirs. We examine the potential use of radar information to quantify forest biomass and classify vegetation type based on data from the New Zealand segment of the 1996 Pacific Rim AIRSAR campaign. Two different types of vegetation were measured: plantation forestry and indigenous scrub. A very significant statistical relationship was found between plantation forestry biomass and measurements derived from AIRSAR data. However, for scrub vegetation, no relationship was found, possibly due to the very small number of ground samples used. In order to separate these vegetation classes from other cover types, two classification procedures, based on the polarimetric characteristics of the backscatter, were trialed. Both approaches produced good classification accuracies for the cover types of interest. methodologies for inclusion in future monitoring programmes. Consequently, one of the projects undertaken in the New Zealand segment of the AIRSAR campaign investigated the relationship between backscatter and biophysical parameters, including forest biomass. Areas of regenerating scrub were of particular interest as it is more difficult to obtain biomass estimates for this type of vegetation than for exotic plantation forestry, where systematic records are generally kept by forestry companies. In this context scrub refers to co-dominant m¯ anuka, (Leptospermum scoparium), and k¯ anuka, (Kunzea ericoides), two similar indigenous species ranging in height from 3 to 6 metres. This vegetation plays an important role in natural regeneration; it is also significant as a carbon sink due to its high wood density. Methodology and results of the biomass estimation study are given for both exotic forest and indigenous scrub sites. Following this, two approaches to classifying vegetation cover based on polarimetric features are presented. Finally, conclusions and future work are discussed. Biomass Estimation Data were collected over two regions of the central volcanic plateau, North Island, using the POLSAR mode, providing full polarimetric data in P,L, and C bands. The two 10 km x 10 km images, identified as CM5231 (scrub) and CM5232 (forest), were both taken on 5 November 1996 from the same flight line at 196 degrees with the radar instrument looking to the left. Data were processed to a compressed Stokes matrix format by NASA / JPL before being supplied to New Zealand on CD-ROM. Imagery was geocoded by Landcare Research to the New Zealand Map Introduction Monitoring forest biomass is a significant global issue, with the adoption of treaties limiting the emission of CO 2 and other greenhouse gasses and the possibility of industries and even countries trading carbon credits. In New Zealand this issue is particularly important as the national strategy is to use tree planting to help balance our total carbon budget. As a consequence, scientifically sound methods must be identified to monitor the biomass carbon not only in exotic forests, but also in large areas of regenerating indigenous scrub and other indigenous forest types. Tate et al. (1997) estimate that 80% of New Zealand’s above ground organic carbon occurs in indigenous forest ecosystems, including scrub, on less than 26% of the land area. Radar remote sensing is one potential method to cost- effectively monitor forest biomass over large areas. There has been a world-wide interest in the use of radar for this purpose in recent years (cf. Dobson et al., 1995; Lucas et al., 1998). However, relationships found generally apply to a specific set of conditions and vegetation type. Factors such as tree structure, wood density, and moisture content affect the backscatter. In addition, non-vegetative factors such as temperature, soil type, ground moisture, surface roughness, and slope can also influence radar backscatter. For these reasons the methodology must be examined in our local environment for the vegetation types of interest. The 1996 Pacific Rim AIRSAR campaign was a unique opportunity to obtain full polarimetric multi-wavelength radar imagery of local test sites. Although operational availability of similar imagery from satellite sources is still some years away, AIRSAR imagery is invaluable to assess