With the rising need for water resource management tasks, such as lake coastal zone management, rising seas border shift detection, and erosion monitoring, there is a rising necessity in human effort to provide precise water body metadata. The currently deployed satellites provide complementary data that are either of high spatial or high temporal resolutions. As a result, there is a clear trade-off between space and time when considering a single data source. For efficient monitoring of various environmental resources, several earth science applications require data at both high spatial and temporal resolutions. To address this need, a large number of data fusion approaches have been described in the literature, that rely on combining data snapshots from multiple sources. Most of the fusion methods learn a mapping of data at different resolutions utilizing the local correlation between the pixel values. The main limitations of data fusion methods are their reliability on data correlations that are sensitive to atmospheric disturbances and other climatic factors that result in noise, outliers, and missing data. To address this challenge, we propose Hydrological Generative Adversarial Network (Hydro-GAN), a novel machine-learning-based method that maps the available data at low resolution to a high-resolution data counterpart. Our proposed model integrates generative adversarial networks that we modified to better the shape of the water body boundaries. We limited our research scope to mapping water bodies images acquired from Moderate Resolution Imaging Spectroradiometer (MODIS) at low resolution, and Land Remote-Sensing Satellite (LANDSAT) at high resolution.

Figure:1 -> Architecture of HydroGAN model used for the training and evaluation of the results.