Sub-pixel mapping of urban green space using multiple endmember spectral mixture analysis of EO-1 Hyperion data

Urban green space is an important biophysical component in assessing urban environment. Remote sensing technology offers an alternative method to traditional ground-based survey of these green spaces. However, accurate green space extraction is still a challenge due to the existence of mixed pixels. Traditional methods such as classification and NDVI for deriving green space are found to be inaccurate and unsatisfactory. Multiple endmember spectral mixture analysis (MESMA) models spectra as the linear sum of spectrally pure end members that vary on a per-pixel basis, which is a technique for identifying materials in a hyperspectral image using end members from a spectral library. In this paper, we explore MESMA approach to extract information of green space for the city of Palo Alto, the United States, using Hyperion imagery acquired in 2002. An atmospheric correction was used to convert the radiance to surface reflectance, then a water mask was applied to remove the dark objects from the image. Ater that, A spectral library consisted of three types of image endmembers (green vegetation, non-photosynthetic vegetation and impervious surface) were constructed through the cooperation of the pixel purity index (PPI) method and Endmember Average RMSE (EAR) , then MESMA were applied to 4-endmember model to generate fraction abundance maps for each endmember. Confusion matrix and RMSE were used in evaluating the fraction images. The overall root mean square (RMS) errors for green vegetation and non-photosynthetic vegetation was 0.237% and 0.246% respectively, the overall accuracy for the fraction image is 84.21%, while the Kappa coefficient get a value of 73.93%. The results indicate that MESMA were an effective method for extracting information of green space from hyperspectral remote sensed data in the heterogeneous and spectral mixed urban area . Current and future areas of research are outlined toward incorporating the imaging spectrometry to provide more accurate refe- rence endmembers, additionally, to analyze change detection of the green space or other physical components of the city , a time series of Hyperion data will be analyzed with the application of MESMA.