A spatially adaptive spectral mixture analysis for mapping subpixel urban impervious surface distribution

Spectral mixture analysis (SMA) has been widely employed in analyzing urban environments, especially for estimating urban impervious surface distribution at the subpixel level. When implementing SMA, endmember selection is considered an important step, and an inappropriate endmember set could severely affect the accuracy of fractional land covers. Although different degrees of success have been achieved in existing SMA approaches, the paradox of endmember selection is still unsolved: theoretically “purest” endmembers that can be selected with relative ease do not always yield optimal results, while the selection of the most “representative” endmembers is very difficult with simple SMA. To address this problem, we propose a spatially adaptive SMA (SASMA) technique to automatically extract and synthesize the “most representative” endmembers for SMA through considering both between-class and within-class variations. In particular, we developed a classification tree method to automatically extract endmember candidates through incorporating spectral and spatial information. In addition, to mitigate the effect of within-class variation, we employed synthetic spectra of neighboring endmember candidates in a local search window as the most “representative” endmember signature. Results indicate that SASMA performs well in estimating subpixel impervious surface distribution with relatively high precision (mean absolute error of 8.50%, root mean square error of 15.25%, and R2 of 0.701) and small bias (systematic error of − 0.93%).

► We develop an automatic spatially adaptive spectral mixture analysis (SASMA) method.
► Endmember candidates were automatically chosen using a classification tree approach.
► Spectral and spatial information was applied to develop the classification rules.
► Endmember spectra were derived as spatially weighted spectra of adjacent candidates.
► SASMA performs well in estimating impervious surface fractions with high precision.