Estimation of impervious surfaces of Beijing, China, with spectral normalized images using linear spectral mixture analysis and artificial neural network

The objective of this article is to evaluate the effectiveness of various algorithms for estimating impervious surfaces. Linear spectral mixture analysis (LSMA) and multi-layer perceptron (MLP) network using original and spectral normalized images were applied to two ASTER images acquired on 31 August and 9 April 2004, respectively. Accuracy assessment was performed with a Quickbird image. Root-mean-square errors (RMSEs) were calculated and compared. Results indicated that LSMA with original images provided the poorest results. RMSE was 14.8% for the August image and 22.4% for the April image. Results from LSMA with normalized images improved significantly with RMSE of 12.6% for the August image and 18.9% for the April image. The MLP modelling with original images generated slightly better results with RMSE of 12.2% and 18.4% for each image. The MLP modelling of normalized images provided the best estimation, yielding a RMSE of 12.1% for the August image and 18.2% for the April image.