| dc.description.abstract | Maintaining the condition of the drainage areas is an important task of the regional water authority “Hoogheemraadschap De Stichtse Rijnlanden”, abbreviated by HDSR. HDSR prefers to monitor the maintenance condition of the drainage areas continuously in order to keep the condition at an acceptable level. However, observing this condition continuously for each drainage area is financially unfeasible. The aim of the study is to develop an approach to measure the maintenance condition of the main waterways of the drainage areas of HDSR in real time. This approach requires the water level difference in the waterways, Δh, as a parameter of the maintenance condition, since Δh is linked to flow resistance. In order to develop the approach, two case studies were performed: one in the Amerongerwetering drainage area and one in the Lange Weide drainage area. During the study, machine learning techniques, such as a linear regression model, a random forest model and a gradient boosting model, were applied. The models required a large input dataset to predict the Δh values. These value were compared to the observed values of Δh. When observing a significant difference between the predicted and the observed values, the date was classified as an anomaly. The data included in the study were provided by HDSR and KNMI. The linear regression model was not suitable for the study, because of insufficient prediction quality. Both the results of the random forest and gradient boosting model showed that most of the anomalies were detected. The anomalies were analysed and were linked to possible explanations. This analysis explained that excessive vegetation has large influence on Δh. The approach proved more promising for the (simpler) Amerongerwetering drainage area, compared to the Lange Weide drainage area. The random forest model proved to be a better performing model, both statistically and visually. The study concludes that machine learning provides opportunities for the water management in the drainage areas of HDSR. However, it is recommended that these opportunities are further examined in future studies. | |
