Development of Machine Learning Flood Model Using Artificial Neural Network (ANN) at Var River

Abstract

Data-driven flow forecasting models, such as Artificial Neural Networks (ANNs), are increasingly used for operational flood warning systems. In this research, we systematically evaluate different machine learning techniques (random forest and decision tree) and compare them with classical methods of the NAM rainfall run-off model for the Vésubie River, Nice, France. The modeled network is trained and tested using discharge, precipitation, temperature, and evapotranspiration data for about four years (2011–2014). A comparative investigation is executed to assess the performance of the model by using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and a correlation coefficient (R). According to the result, Feed Forward Neural Network (FFNN) (a type of ANN) models are less efficient than NAM models. The precision parameters correlation coefficient of ANN is 0.58 and for the NAM model is 0.76 for the validation dataset. In all machine learning models, the decision tree which performed best had a correlation coefficient of 0.99. ANN validation data prediction is good compared to the training, which is the opposite in the NAM model. ANN can be improved by fitting more input variables in the training dataset for a long period.