Forecasting Next Day River Levels

Abstract

The objective of this study was to tackle the issue of effectively predicting the river levels in the Taff River catchment for the following day. In order to achieve this, the research utilised daily flow data and rainfall data from six gauges located within the catchment area. These data were obtained via the National River Flow Archives, which are managed by the UK Centre for Ecology and Hydrology. The problem was addressed by means of importing, cleansing, and consolidating the data into a unified data frame. Visualisations were employed to examine the distributions of variables, explore their correlations, and detect any potential seasonal trends. The data included skewed distributions and evident seasonal patterns were identified. The initial approach involved constructing a linear regression model that incorporated the seasonal intercept and interactions among the variables. Transformation techniques were applied in order to enhance the precision of the model. Nevertheless, it was discovered that the assumptions underlying the linear model were not fully satisfied. In order to mitigate this constraint, a more extensive implementation of the generalized additive model (GAM) was employed. The GAM demonstrated superior performance compared to the linear regression model, particularly when a log transformation was employed. Nevertheless, the incorporation of the inverse transformation led to reduced accuracy in the model. Furthermore, the utilisation of GAM in conjunction with principal component analysis (PCA) yielded no additional enhancements. Despite the superior performance exhibited via utilising GAM, it was observed that the model’s assumptions were not entirely met. Moreover, a data point with a significant residual was detected, suggesting greater departures during the autumn and winter months. Arising from this, a suggestion for future research would be to undertake an investigation which takes into account the various aspects that contribute to this observed seasonal trend. The research conducted in this study encountered certain constraints, such as the presence of missing data that necessitated the use of imputation techniques, as well as the exclusion of

other possibly important elements. Consequently, the results underscore the intricate nature and difficulties associated with precise prediction of river levels. In summary, this research study aimed to tackle the issue of predicting future river levels in the Taff River catchment area. Both the linear regression model and the more complex GAM were utilised, with the GAM demonstrating improved performance. Nevertheless, it is important to acknowledge that both models possessed certain shortcomings, and both failed to fully adhere to the necessary assumptions for precise forecasting. Conducting additional study and making further adjustments is advisable, in order to enhance the effectiveness and real-world utility of the models utilised in the Taff River basin.