Deriving Land Cover Fractional Maps with Unmixing Algorithm

Uttam Kumar, NASA Ames Research Center / ORAU, uttam215@gmail.com (Presenter)
Cristina Milesi, NASA ARC, cristina.milesi@nasa.gov
S Kumar Raja, Airbus Engineering Centre, Bangalore, India, sk.kumar.raja@gmail.com
Ramakrishna R. Nemani, NASA ARC, rama.nemani@nasa.gov
Sangram Ganguly, NASA ARC BAERI, sangramganguly@gmail.com

Land cover (LC) features generally occur at finer spatial scales than the resolution of most primary satellites, leading to mixed pixels in the observed data. Hence, LC mapping at subpixel level is required to obtain fractional maps of each category in a given pixel using unmixing algorithms such as linear mixture model, which assumes no interaction between materials and a pixel is treated as a linear combination of signatures with relative concentrations. The model allows different LC types to be present, each contributing a fraction of its (unique, fixed) spectrum corresponding to the area occupied by that LC type in a pixel. The optimal solution of the mixture models can be an unconstrained solution, or a constrained solution when the abundance nonnegativity and abundance sum-to-one constraints are imposed.

Here we demonstrate the performance of Fully Constrained Least Squares (FCLS) algorithm implemented in C++ programming language with OpenCV package and boost C++ libraries in the NASA Earth Exchange. A set of global endmembers generated for three LC classes, namely, SVD (substrate, vegetation and dark objects) were used to unmix a set of synthetic test data and Landsat data. FCLS was first evaluated on computer simulated data with Gaussian noise of different signal-to-noise ratio. In the second set of experiments, a spectrally diverse collection of 11 scenes of Level 1 terrain corrected, cloud free Landsat-5 TM data representing an agricultural scenario in Fresno, California, USA were used and the results were validated with the corresponding ground data. Finally, a pair of coincident clear sky Landsat TM data were unmixed and validated with the World View 2 data (of 2 m spatial resolution) for an area of San Francisco (an urbanized landscape). The results were evaluated using descriptive statistics, correlation coefficient, RMSE, probability of success, boxplot and bivariate distribution function.

Presentation Type:  Plenary Talk

Session:  Poster Speed Talks

Presentation Time:  Mon 4:00 PM  (1 minutes)