Towards high-resolution flood monitoring at daily scale: fusing Suomi-NPP/VIIRS and Sentinel-1 SAR based inundation maps by adversarial learning
Abstract
Remote sensing-based flood monitoring is limited by the low spatiotemporal resolution or contamination (e.g., cloud and shadows), which results in miss coverage or low-quality images, especially for flash flood events. Spatial downscaling1 (SD) and missing value filling2 (MVF) are two common techniques to improve the inundation maps, which relies only on the topography characteristic or water occurrence3. In this study, we introduce a deep learning based end-to-end method to map the water bodies distribution from Suomi-NPP/VIIRS (~375m)4 to Sentinel-1 SAR (~10m)5, with conditional predictors including HAND, DEM, land cover, and surface water occurrence. Specifically, we utilize the framework of Bicycle-GAN6 to enable probability mapping. We also regularize the generator based on the cycle-consistent statistical7. After training with over half-million patch samples, we test the model at various regions across the contiguous United States. Results show that flood inundation estimates from the proposed model outperform the traditional SD+MVF or convolutional neural network. Based on the model structure, we further evaluate the impact of patch size and the missing pixel ratio on the performance. The accuracy and practicality of the method renders it promising for daily high-resolution flood inundation monitoring at both the regional and continental scale.
Reference: Aires F, Miolane L et al. Journal of Hydrometeorology 2017; Zhao G, Gao H. Geophysical Research Letters 2018; Pekel JF, Cottan A et al. Nature 2016. Li S, Sun D et al. Remote Sensing of Environment 2018; Yang Q, Shen X et al. Bulletin of American Meteorological Society 2021; Zhu J, Zhang R et al. Conference on Neural Information Processing Systems 2017; Pan B, Anderson G et al. Journal of Advances in Modeling Earth Systems 2021.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H53D..04Y