The potential of high-resolution optical satellite images for mapping of ecologically important urban space is investigated in this study. Both a GeoEye-1 and a Landsat 8 scene over central Shanghai were first segmented by two different algorithms and then classified into seven urban classes by SVM. Shadows in the pan-sharpened GeoEye-1 image were masked out and replaced by the corresponding pan-sharpened classified Landsat 8 image. Largest confusions occurred between sealed and permeable but non-vegetated surfaces, and between low-rise residential and high-rise commercial buildings. Based on the classification result, ecosystem service balances, supply and demand was modelled for each particular land cover class. Classification accuracies of 88% and 91% could be reached, indicating the suitability of the underlying data and method for this application domain. The KTH-SEG segmentation algorithm slightly outperformed the one implemented in eCognition. The highest supply of ecosystem services was found in water bodies whereas high-rise built-up areas revealed largest demands.