Close Window

LC-20 Abstract

Hyperspectral analysis of landcover in Rondonia

Lenio Soares Galvao — IAI - Inter-American Institute for Global Change Research (SA-PI)
Dar A. Roberts — University of California (US-PI)

Remote sensing is a critical component of LBA

ecology, contributing to the basin wide study primarily through land-cover

classification, land-cover change, and biophysical retrievals. Accurate maps of

land-cover, land cover change and canopy biophysical properties are needed as

inputs for hydrological and biogeochemical models, as a means for quantifying

above ground carbon stocks, scaling up flux tower measurements beyond tower

sites and to direct field sampling.  Currently a wide array of

spaceborne image data are available to the project including multitemporal Landsat TM,

AVHRR, some SAR and more recently, MODIS, SeaWiFS, Ikonos and ASTER. While many

of these data sets are invaluable, they do not have the capability of

separating many important land-cover classes, nor the capability of more accurate

biophysical retrievals offered by hyperspectral systems. Examples of

land-cover classes that cannot be distinguished reliably with image data available to the

project include green pasture, early regeneration and tree crops. Past

research in Eastern Amazonia, and more current research in Rondonia has shown that

some of these ambiguities are non trivial. For example, in the vicinity of Ji

Parana, spectral ambiguity between green pasture and second growth forest results

in interannual fluctuations between these two cover classes of up to

10%, with the highest proportion of second growth occuring in early dry

season images.  No such ambiguity will occur in hyperspectral data. Tree crops,

which cannot be separated from some second growth forests without

extensive supporting field data, should also be spectrally distinct in

hyperspectral data. Improved biophysical retrievals include LAI retrievals for

forested areas at ranges that typically saturate NDVI.  LBA ecology is also limited by the availability

of high quality spectral reflectance data at branch to canopy scales. In order

to best interpret canopy reflectance, it is critical that data exist at an

appropriate spatial scale for remotely sensed data. Because of the difficulty of

acquiring a sufficient number of spectra from towers, an airborne campaign is

virtually the only way to develop such a library. Not only can such a library

improve the quality of land-cover maps, but it can be convolved to other

sensors, and thereby used to improve analysis using broad band systems that cover a

larger area and include archived data sets.  For this research, we propose to address a number

of critical problems in land-cover mapping and biophysical retrievals through

the use of high resolution AVIRIS. We propose a 4 to 8 week deployment,

starting in late June/early July, at the start of the dry season and

prior to the onset of significant burning. We propose target sites in

Rondonia that include well characterized sites such as Fazenda Nova Vida, tower

sites such as Jaru, and a selection of targets that cover a range of land-cover

categories (crop, pastures of ranging quality, second growth at different

ages and primary forest). Site selection would be guided by extensive

GIS layers and Landsat time series we have developed for the state of Rondonia.

A key focus will be on well characterized sites. Expected products include a

regionally specific spectral library for Rondonia that includes all major

dominant cover types, land-cover maps using techniques developed at UCSB, and

new biophysical maps such as LAI retrievals. UCSB will work closely with

Brazilian collaborators and other LBA researchers to develop optimal test sites

throughout the state. This project contributes to LBA educational initiatives

through graduate student support of a Brazilian PhD and a remote sensing short

course on hyperspectral remote sensing, to be taught in Rondonia and

potentially at INPE.    

Close Window