Can machine learning improve the accuracy of water level forecasts for inland navigation? Case study: Rhine River Basin, Germany

In a context where the German Federal Ministry of Transport and Digital Infrastructure (BMVI) expects a 23% growth of traffic in inland navigation in Germany by the year 2030, a higher efficiency and an optimized logistics in traffic management are required. The BMVI-funded project Digital Skipper Assistant (DSA) had the objective to develop a cloud-based application to support inland navigation, able to calculate best routes and Estimated Times of Arrival (ETA). Contributing to the DSA, this study investigates the competence of artificial neural networks (ANNs) to predict water levels up to 10 days ahead in some crucial gauges of the Rhine River Basin in Germany. A multiple-outputs model based on long short-term memory (LSTM) networks was implemented, adopting as inputs firstly the water level measurements at specific gauges. In a second work phase, the water level forecasts of the hydrological model chain of the German Federal Institute of Hydrology (BfG) were included as additional predictor into the model, highly improving the results (the coefficient of determination R2 increased of about 20%). The LSTM model has been trained, validated and tested (respectively with 80%, 10% and 10% of the dataset) using the historical data and the BfG hindcasts from January 2008 until December 2015. The results of the model evaluation were very good (i.e. R2 around 90% for 7-days prediction). Several tests have been run during the DSA field test (from July 2018 to December 2018) and the results were promising