Vol. 10, No. 2, 102-109, 2011
Isabelle Champion, Pascale Dubois-Fernandez, and Xavier Dupuis
Abstract
The objective of this paper was to study the use of texture features to retrieve forest biomass, which is of crucial importance to global carbon studies. Texture features are derived from: a) σ° distribution statistics (variance and entropy), and b) the grey level co-occurrence matrix (energy, entropy). Radar images were acquired at P band cross and parallel polarisations using an airborne radar system built at the French aerospace laboratory and flying over an experimental forest at the north of the Landes region (France). This forest is documented by the French National Institute of Agricultural Research (INRA) as a controlled homogeneous test site: a single species, even-aged forest, subject to identical silvicultural treatment and the sampling of ten stands which cover all forest stages from sowing to harvest and provides sampled data on age, height, trunk dbh (diameter at breast height) and density measurements. Biomass was then calculated from ground data by means of allometric equations for these ten stands using stand age, dbh and density estimations.
Forest biomass was inferred from image texture using two methods. The first directly deduced forest biomass from image texture on the basis of biomass/texture regressions. As a second step we tried to avoid the use of structural measurements to calculate the biomass (stand dbh and density), so that dbh was inferred from texture and density was deduced from stand age. Total and stem biomass were then recalculated.
It was concluded that some of the texture indicators were significantly correlated to growth parameters (age, trunk dbh, biomass) displaying regularly linear variations. Moreover, the uncertainty related to these estimation methods was less than 20% of the biomass value, whichever way biomass is retrieved from texture (directly or through a dbh/texture regression). Texture could therefore be used as an alternative characterisation technique for plantation forests where intensity-based relationships are showing decreased sensitivity for mature stands.
Submitted: 27 June 2011
Revised: 23 Aug 2011
Accepted: 24 Aug 2011
Published: 7 Sept 2011
Responsible editor: Ioannis Gitas Citation
Champion I, P Dubois-Fernandez & X Dupuis, 2011. Retrieving forest biomass from the texture of SAR images. EARSeL eProceedings, 10(2): 102-109
 |
EARSeL European Association of Remote Sensing Laboratories, Strasbourg, France |
 |
BIS-Verlag |