Keeping greed good: sparse regression under design uncertainty with application to biomass characterization