Prediction of Urban Non-point Pollution Load by Statistical Analysis of Data of Published Research and Its Reliability Evaluation –Statistical Analysis of Mean Load and Verification and Modification of Previously Proposed Model Using Newly Obtained Data–

Abstract

To verify a statistical model for predicting urban pollutant runoff developed in our previous research, newly obtained runoff data were compared with those predicted by the model. The proposed model previously was a regression model using parameters representing geological, rainfall, and hydrological characteristics. The targeted pollutants were COD, SS, TN, and TP, and their event mean concentrations (EMCs) for each rainfall were predicted. From the comparison, the model was found to predict the EMC to one order of magnitude. Moreover, the yearly mean EMC was evaluated from only the mean and standard deviation of all data for each index. The error ratio of the prediction of the mean of 50 rainfall events was within 50%. Furthermore, in order to consider the possible differences among different catchment areas, the EMC values for three catchment areas newly obtained were compared statistically with nationwide values obtained previously. Significant differences were found for one area out of the three which thus emphasizes the importance of the consideration of catchment area differences. Thereafter, the number of runoff samples in a specific watershed area required to detect the mean EMC difference from the nationwide database was derived statistically. This number was calculated to be 10 at most for detecting more than double or less than half of the database means. Lastly, a modified model that includes watershed area differences using a generalized linear mixed model was proposed.