Development of Three-Dimensional Convolutional Neural Network for Urban Flood Classification Using Synthetic Aperture Radar Multi-Temporal Image

Urban flooding is a significant catastrophe due to its widespread impact on the population. Typically, floods occur concurrently with heavy rainfall, rendering the affected area obscured by clouds when observed through optical sensors on satellites. A novel approach is proposed in this study to address this issue, aiming to classify flooded urban areas using a remote sensing synthetic aperture radar (SAR) sensor on a satellite. Unlike optical sensors, SAR can penetrate clouds. The framework was developed using the 3D convolutional neural network (3DCNN) method to preserve the temporal variability, which processed multi-temporal SAR data from Sentinel-1 (S-1). The dataset used in this research comprised 24 S-1 scenes covering Jakarta Province, Indonesia, with Dual VV and VH polarization, between March 2019 and February 2020, divided into two co-event images, 18 pre-event images, and four post-event images. We employed hyper-parameters of 150 epochs, a batch size of 100, and a learning rate of 0.001, with a training/testing data split of 80/20 in the training phase. The 3DCNN achieved an average overall accuracy of 70.3%, with a maximum accuracy of 71.4%, and each epoch took 113 seconds on average to process. These results demonstrate the potential of the 3DCNN method to accurately estimate the extent of flooding and identify areas at risk of flooding, thereby aiding early detection and flood prevention efforts in other urban areas in the future.