A Temporal Deep Convolutional Neural Network Model on Sentinel-1 Image Time Series for Pixel-Wise Flood Classification

Abstract

Accurate and timely flood mapping is important in emergency management during and after extreme flood events which can be greatly served by Synthetic Aperture Radar (SAR). This work is focused on open-land regions where a custom annotation based on expert knowledge and NDFI is used. Research on SAR flood detection is mostly based on histogram thresholding that has low time complexity and seems ideal for emergency response, although human intervention is needed. Machine learning methods have fewer errors and minimize the need for human intervention but their computational complexity is higher. This work aims to provide a lightweight convolutional neural network baseline for pixel-wise time series classification in flood monitoring on SAR satellite image time series. Quantitative and qualitative evaluation of results indicate that the approach is promising.

WQEMS platform for inland surface water bodies' monitoring: serving user communities and supporting expert analyses

“…The flood sub-service is based on a time series of properly preprocessed Sentinel-1 images. A trained temporal deep convolutional neural network model has been adapted to generate a flood map to depict flooded areas 33 for a given date (Figure 19). Moreover, several statistics, which can be related to the intensity of a flood event, are also available to the users.…”