Flood classification and prediction in South Sudan using artificial intelligence models under a changing climate

Abstract

This study used Artificial Intelligence (AI) techniques as a modeling tool to estimate the risk of Nile flooding in the cities of southern Sudan. Climatic records, and precipitation, from stations along the area were used between 2010 and 2019. To test how well the models worked, the forecast was done using a variety of stations. To determine the flood rate in southern Sudan with the highest degree of accuracy, various artificial neural network techniques were investigated. Six artificial neural network (ANN) models were created and compared to show flood prediction to reach the maximum level of accuracy and to improve the results (NN, GRNN, RNN, CFNN, PNN, FFNN). The artificial neural network (FFNN) produced the best results in the first test, reaching a 95 % accuracy rate. Three further strategies were evaluated by increasing the neural network's hidden layer count to ten. Tests with 15 and 25 hidden layers also showed that the accuracy changes with the increase of hidden layers. Also, six other algorithms were applied to reach the highest value expected from Using one of the artificial intelligence techniques (AI), in predicting the flood by machine learning methods (ML). The highest expected value of flooding was reached through the (Gradient Boosting) model, where it was Classification Accuracy (CA) 0.937, followed by (AdaBoost), (CA 0.916).