Realistic River Image Synthesis Using Deep Generative Adversarial Networks

Abstract

In this paper, we demonstrated a practical application of realistic river image generation using deep learning. Specifically, we explored a generative adversarial network (GAN) model capable of generating high-resolution and realistic river images that can be used to support modeling and analysis in surface water estimation, river meandering, wetland loss, and other hydrological research studies. First, we have created an extensive repository of overhead river images to be used in training. Second, we incorporated the Progressive Growing GAN (PGGAN), a network architecture that iteratively trains smaller-resolution GANs to gradually build up to a very high resolution to generate high quality (i.e., 1,024 × 1,024) synthetic river imagery. With simpler GAN architectures, difficulties arose in terms of exponential increase of training time and vanishing/exploding gradient issues, which the PGGAN implementation seemed to significantly reduce. The results presented in this study show great promise in generating high-quality images and capturing the details of river structure and flow to support hydrological modeling and research.

Blockchain Applications and Opportunities for Water Resources and Hydrology: A Systematic Review

“…Hydrology and water resources domain followed the same trend with developments in hydroinformatics (Demir et al, 2022). The most notable developments can be listed as novel applications of deep learning in image synthesis and communication (Gautam et al, 2022;Sermet and Demir, 2021), large scale modeling and analysis on client-side systems (Ewing et al, 2022;, virtual and augmented reality for hydrological education and modeling purposes (Sermet and Demir, 2022), and novel programming libraries and data standards (Ramirez et al, 2022;Xiang and Demir, 2022). As a natural outcome of this rapid digital transformation, the water sector has started to generate, process, and store more data (Haltas et al, 2021).…”

TempNet – Temporal Super Resolution of Radar Rainfall Products with Residual CNNs

“…Research studies for better rainfall products can be broadly classified as either quality-or resolution-related. Dataset improvements using deep neural networks include data cleaning to eliminate noise (Lepetit et al, 2021), increasing the resolution or accuracy of datasets by various statistical or data-driven methods (Li et al, 2019;Demiray et al, 2021a;Demiray et al, 2021b), synthetic data generation (Gautam et al, 2020), and bias correction . Resolution-related improvements, on the other hand, either focus on increasing the resolution of two spatial dimensions or the temporal dimension.…”

U-net-based semantic classification for flood extent extraction using SAR imagery and GEE platform: A case study for 2019 central US flooding

“…Although both can cover a large area rapidly, remotely sensed imagery has a significant advantage in terms of data accuracy and temporal and spatial consistency. Furthermore, remote sensing data can be utilized to generate new data, such as super-resolution DEMs (Demiray et al, 2021) and new satellite-based river imagery (Gautam et al, 2022). Additionally, remotely sensed data can provide estimates, such as precipitation (Seo et al, 2019) and sedimentation movements (Xu, Demir, et al, 2019;, that are hard to collect using other approaches.…”