Reference data ("ground truth") maps traditionally have been used to assess the accuracy of imaging spectrometer classification algorithms. However, these reference data can be prohibitively expensive to produce, often do not include sub-pixel abundance estimates necessary to assess spectral unmixing algorithms, and lack published validation reports. Our research proposes methodologies to efficiently generate, validate, and apply abundance map reference data (AMRD) to airborne remote sensing scenes. We generated scene-wide AMRD for three different remote sensing scenes using our remotely sensed reference data (RSRD) technique, which spatially aggregates unmixing results from fine scale imagery (e.g., 1-m Ground Sample Distance (GSD)) to co-located coarse scale imagery (e.g., 10-m GSD or larger). We validated the accuracy of this methodology by estimating AMRD in 51 randomly-selected 10 m x 10 m plots, using seven independent methods and observers, including field surveys by two observers, imagery analysis by two observers, and RSRD using three algorithms. Results indicated statistically-significant differences between all versions
of AMRD, suggesting that all forms of reference data need to be validated. Given these significant differences between the independent versions of AMRD, we proposed that the mean of all (MOA) versions of reference data for each plot and class were most likely to represent true abundances. We then compared each version of AMRD to MOA. Best case accuracy was achieved by a version of imagery analysis, which had a mean coverage area error of 2.0%, with a standard deviation of 5.6%. One of the RSRD algorithms was nearly as accurate, achieving a mean error of 3.0%, with a standard deviation of 6.3%, showing the potential of RSRD-based AMRD generation. Application of validated AMRD to specific coarse scale imagery involved three main parts: 1) spatial alignment of coarse and fine scale imagery, 2) aggregation of fine scale abundances to produce coarse scale imagery-specific AMRD, and 3) demonstration of comparisons between coarse scale unmixing abundances and AMRD. Spatial alignment was performed using our scene-wide spectral comparison (SWSC) algorithm, which aligned imagery with accuracy approaching the distance of a single fine scale pixel. We compared simple rectangular aggregation to coarse sensor point spread function (PSF) aggregation, and found that the PSF approach returned lower error, but that rectangular aggregation more accurately estimated true abundances at ground level. We demonstrated various metrics for comparing unmixing results to AMRD, including mean absolute error (MAE) and linear regression (LR). We additionally introduced reference data mean adjusted MAE (MA-MAE), and reference data confidence interval adjusted MAE (CIA-MAE), which account for known error in the reference data itself. MA-MAE analysis indicated that fully constrained linear unmixing of coarse scale imagery across all three scenes returned an error of 10.83% per class and pixel, with regression analysis yielding a slope = 0.85, intercept = 0.04, and R^2 = 0.81. Our reference data research has demonstrated a viable methodology to efficiently generate, validate, and apply AMRD to specific examples of airborne remote sensing imagery, thereby enabling direct quantitative assessment of spectral unmixing performance.
Imaging Science (Ph.D.)
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Jan van Aardt
Williams, McKay D., "Generation, validation, and application of abundance map reference data for spectral unmixing" (2017). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus