We assessed the capability of airborne hyperspectral imagery for detecting and mapping the early invasion by Solidago altissima of understorey vegetation in moist tall grassland. Two sources of ground truth Solidago data were utilized: one, density records in fifteen classes for 10,937 quadrats (10x10 m), and another, density/cover records in 91 quadrats (5x5 m). For each data source, half of the samples, chosen randomly, were used to develop generalized linear models predicting Solidago abundance from hyperspecral data, and the other half were used to validate the models. Original Solidago data and transformed ones for presence-absence and three-class density/cover were utilized as dependent variables, and original hyperspectral bands and MNF (Minimum Noise Fraction) transformed ones were utilized as independent variables. Models constructed from the 5x5 m quadrat data had higher accuracy for the validation dataset than those from the 10x10 m quadrat data to estimate Solidago abundance, while model accuracy was in the descending order of the presence-absence, three-class density/cover and original data for both quadrat data. The MNF bands also improved model accuracy compared to the original bands. Our results suggest that the use of the 5x5 m quadrat data and the MNF bands is effective for the early detection and mapping of S. altissima.