Publications

•This study tackles the challenge of unbalanced stream gauge distributions by leveraging publicly available datasets and deep learning to enhance hydrological modeling in under-gauged regions.

•This study introduces a novel framework combining LSTM-based deep learning with GLDAS and GSIM data for effective runoff prediction across continents.

•This study assesses the sensitivity of LSTM models using data from diverse regions, emphasizing the importance of hydrological similarities for accurate runoff predictions in ungauged basins.

•LSTM models demonstrate superior runoff prediction skills over traditional GLDAS datasets, influenced by the hydrological similarities between training and test regions.

Accurate and timely information about the extent and location of inland water bodies is crucial for various tasks related to managing water resources. A promising tool for identifying these water bodies is the Cyclone Global Navigation Satellite System (CYGNSS). CYGNSS uses eight microsatellites with special radar technology to observe surface reflectivity characteristics over both dry and wet land. This study compares data from CYGNSS with data from two other Earth observation systems—Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat Global Surface Water product—for the contiguous United States.

We conducted a 1-kilometer comparison of water masks (representations of where water is located) for the year 2019, focusing on the area between latitudes 24°N and 37°N. To assess how well these water masks performed, we used statistical measurements called confusion matrices and high-resolution satellite images.

By using specific thresholds for different sub-regions observed by CYGNSS, we improved the performance of our water mask data, with up to a 34% increase in a measurement called the F1-score. We also calculated a metric that compares the amount of inland water to the size of the surrounding area, and found that CYGNSS, MODIS, and Landsat provided very similar estimates, differing by less than 2.3% for each sub-region.

In summary, this study sheds light on how well water masks derived from optical (visible light) and radar-based satellite observations compare in terms of their accuracy and spatial representation of inland water bodies.

Recent advancements in land data systems and satellite soil moisture data show that using this data can help improve surface condition estimates in land models. However, this improvement in estimating soil moisture doesn't necessarily translate to better predictions of water movements like evaporation and runoff. This issue might be due to inaccuracies in how these models link soil moisture with water movements. Experiments suggest that even with better soil moisture data, these inaccuracies can lead to poor water movement predictions. Adjusting satellite data to fit these models isn't always effective and could sometimes worsen predictions. This highlights the need for more accurate models that can better link soil moisture with water movements.

The use of satellite-based soil moisture (SM) data in Earth system science can be hindered by gaps in observations. To address this, our study introduces a method to fill in these gaps in the Soil Moisture Active Passive (SMAP) dataset. We utilized a basic water balance equation, considering factors like precipitation and soil water loss, to estimate soil water content over 12-hour periods. This led to the development of a new, continuous SM product for the contiguous United States. This product, named the SMAP-based 12-hourly SM product, showed promising results, aligning well with on-the-ground measurements and effectively capturing soil moisture variations, especially those related to heavy rainfall events. This continuous dataset, with its detailed insights into soil water dynamics, offers valuable contributions to our understanding of land-surface hydrology.

The Water Balance Equation (WBE) is like a budget for water in a specific area, tracking how much water comes in, goes out, and how much is stored over time. It's crucial in understanding water availability and the water cycle, especially on land. The WBE relates rainfall, evaporation, and runoff to changes in soil moisture.

Recently, there's been interest in using satellite data to guess some WBE parts, but this can be complex. Challenges include simplifications in the equation, missing water cycle elements, and limits of satellite data. Our study used advanced modeling to show these issues can affect the reliability of WBE estimates. We suggest treating these estimates as general guides, reflecting the satellite data used, rather than exact measurements. Better remote sensing and improved WBE will enhance our understanding of the water cycle.

This study aims to address gaps in assessing the accuracy of satellite-based soil moisture data by utilizing machine learning techniques to generate spatially complete error maps for Soil Moisture Active Passive (SMAP), Soil Moisture and Ocean Salinity (SMOS), and Advanced Scatterometer (ASCAT) systems, and by examining the influence of environmental conditions on satellite-based soil moisture retrievals, revealing that a significant portion of missing error information from triple collocation analysis (TCA) can be reconstructed using ensemble prediction mean of machine learning models, contributing to a more comprehensive understanding of soil moisture dynamics across the three satellite missions.