An innovative method based on cloud model learning to identify high-risk pollution intervals of storm-flow on an...

Identifying high-risk storm-flow pollution intervals in an urban watershed is critical for watershed pollution control decision-making. High-risk pollution intervals of storm-flow are defined as storm-flow intervals that contribute more than the background pollutant load, and whose load contribution rank in the top 20%. However, the identification of high-risk pollution intervals is difficult due to variations in the flow-concentration relationship among rain events, uncertainty inherent in stormwater quality data, and physically-based stormwater models requiring a substantial number of parameters. A new method for identifying high-risk pollution intervals during different rain events is proposed. A dataset of the urban watershed located in Shenzhen, southern China, was used to demonstrate the proposed method. A "cut-pool" strategy was initially used to pre-process the dataset for maximizing valuable information hidden in existing datasets and to investigate the impact of rainfall on flow-concentration relationships. Gaussian cloud distribution was then introduced to capture the trend, dispersing extent and randomness of stormwater quality data at any flow interval. Interval Overlapping Ratio (IOR) and Load contribution of storm-flow high-risk pollution intervals was used to assess the performance of the method. Results show that storm-flow high-risk Chemical Oxygen Demand (COD) pollution intervals of the Shiyan watershed was 0.5-1.5 mm under light rain (0-13 mm), 1-3 mm under moderate rain (13-27 mm) and 5-7 mm under heavy rain (27-43 mm). The accuracy of the identified high-risk pollution intervals (IOR) was 63-66% under light rain, 64-67% under moderate rain. Moreover, COD load can be reduced by 44-48% with high-risk storm-flow under light rain; 43-49% under moderate rain; 32% under heavy rain. This method is very useful for effectively controlling storm-flow pollution on an urban catchment scale.