Abstract
Our research aims on predicting the volume of roadway runoff volume based on rainfall. This study is significant because it allows us to estimate the quantity of pollutants, including chemicals from brake wear, vehicle emissions, and notably, polyfluoroalkyl substances (PFAS). PFAS are known for their persistence in the environment and resistance to conventional water treatment methods. By understanding and quantifying runoff, we can assess the impact of these pollutants as they are transported into larger bodies of water, posing risks to ecosystems and human health.
In this study, we employed a multivariate time series transformer model [1] to predict runoff data based on rainfall. We explored two versions of the model: one with pretraining and one without pretraining. Our research utilized a rainfall-runoff dataset provided by the California Department of Transportation. The results demonstrated that the pretrained model significantly outperformed all baseline methods in terms of RMSE (Root Mean Square Error). We've also developed a user-friendly web interface using Gradio. This interface enables users to effortlessly upload rainfall runoff data files and generate graphs comparing predicted and runoff amount.