The Prediction of Flood Monitoring for Image Satellite Using Artificial Neural Networks

Flooding poses significant risks for property and human life, particularly in compactly settled city areas like Khartoum city in Sudan. Remote sensing technologies offer valuable tools for monitoring and mitigating flood impacts. This study investigates flood dynamics in part of Khartoum city using land cover classification of Landsat 7 and 8 satellite imagery and Artificial Neural Networks (ANNs). The radiometric and atmospheric distortions were corrected by applying pan-sharpening to the imagery. Maximum Likelihood algorithm for supervised classification combined with ground survey data, were conducted, followed by post-classification refinements to enhance accuracy. Also, assessment metrics included overall accuracy and kappa coefficient were used. Additionally, we developed a sophisticated ANN model for land cover classification and flood prediction. The input parameters were Red, Green and Blue (RGB) values and the target was the contrast. The model was trained and validated by divided the datasets into training data, validation data, and testing data. The performance for the ANN model was evaluated based on the regression square (R²) values, which were 0.98825, 0.90020, and 0.97032 for training, validation and testing data, respectively. The overall R² value for all the datasets was 0.9596. These high R² values were demonstrate the model's strong predictive capability and generalization power, affirming its effectiveness in accurately predicting land cover and flood dynamics. This makes the ANN model a valuable tool for environmental monitoring and land management applications.