High Resolution Imagery-Based Statistical Analysis for Urban Rivers Extraction and Quality Classification—Case Study in Beijing District of China

The previous work on monitoring the urban water quality using remote sensing focused mainly on the specific spectral/band math and the regression relationship between the spectral signature and water surface’s Chlorophyll-a content, while conducting urban wastewater monitoring on the basis of indirect interpretation indicators is still a gap to be bridged. Here we report a novel hybrid method to extract and classify the urban wastewaters (caused mainly by eutrophication) from 4-band high resolution imagery using a multi-step statistical approach. First, the MTSUWI (Modified Two-Step Urban Water Index) algorithm is presented for extracting the urban water automatically, and the Kappa Coefficient comes up to 0.92. Second, using the Minimum Message Length Criterion-Expectation Maximization algorithm (MML-EM), the histogram of the water-masked 1st principal component image was screened into two subpopulations, which are mainly a re-flection of the existing background of the waters, rather than the pollution. Third, the water floating matters (most of them are green algae) is selected as an indirect interpretation key of the polluted wastewaters, and by gradually reducing the interpretation target area, pixels containing green alga were enhanced and then extracted in the brightness image. Fourth, any river reach containing at least one patch of green alga is labeled as “polluted”, otherwise they are labeled as “clean/fresh”, and finally, eight black-and-odorous wastewaters in the study area were selected out, which are consistent with literatures and observations from the field. This research is one of the first to apply indirect interpretation indicators to 4-band high resolution imagery for the classification of urban polluted rivers.