Bridging the gap: USUS computer scientists develop model to enhance water data from satellites
Pursuing NSF-funded research, Utah State University researchers publish findings in AGU's 'Water Resources Research' journal
LOGAN, UTAH, USA -- Satellites encircling the Earth collect a bounty of water data about our planet, yet distilling usable information from these sources about our oceans, lakes, rivers and streams can be a challenge.
“Water managers need accurate data for water resource management tasks, including lake coastal zone monitoring, rising seas border shift detection and erosion monitoring,” says Utah State University computer scientist Pouya Hosseinzadeh. “But they face a trade-off when reviewing data from currently deployed satellites, which yield complementary data that are either of high spatial or high temporal resolutions. We’re trying to integrate the data to provide more accurate information.”
Varied data fusion approaches present limitations, including sensitivity to atmospheric disturbances and other climatic factors that can result in noise, outliers and missing data.
A proposed solution, say Hosseinzadeh, a doctoral student, and his faculty mentor Soukaina Filali Boubrahimi, is the Hydrological Generative Adversarial Network – known as Hydro-GAN. The scientists developed the Hydro-GAN model with USU colleagues Ashit Neema, Ayman Nassar and Shah Muhammad Hamdi, and describe this tool in the March 13, 2024, online issue of the American Geophysical Union journal Water Resources Research.
The team’s research is supported by the National Science Foundation.
Hydro-GAN, says Filali Boubrahimi, assistant professor in USU’s Department of Computer Science, is a novel machine learning-based method that maps the available satellite data at low resolution to a high-resolution data counterpart.
“In our paper, we describe integrating data collected by MODIS, a spectroradiometer aboard the Terra Earth Observing System satellite, and the Landsat 8 satellite, both of which have varied spatial and temporal resolutions,” she says. “We’re trying to bridge the gap by generating new data samples from images collected by these satellites that improve the resolution of the shape of water boundaries.”
The dataset used in this research consists of image data collected during a seven-year span (2015-2021) of 20 reservoirs in the United States, Australia, Mexico and other countries. The authors present a case study of Lake Tharthar, a salt water lake in Iraq, comparable in size to Great Salt Lake and facing similar climate and usage pressures.
“Using seven years of data from MODIS and Landsat 8, we evaluated our proposed Hydro-GAN model on Lake Tharthar’s shrinking and expansion behaviors,” Hosseinzadeh says. “Using Hydro-GAN, we were able to improve our predictions about the lake’s changing area.”
Such information is critical for the region’s hydrologists and environmental scientists, he says, who need to monitor seasonal dynamics and make decisions about how to sustain the lake’s water supply.
The scientists demonstrate Hydro-GAN can generate high-resolution data at historical time steps, which is otherwise unavailable, for situations where a large amount of historical data is needed for accurate forecasting.
“We think this will be a valuable tool for water managers and, moving forward with similar models, we can employ a multi-modal approach to provide data in addition to images, including information about topology, snow data amounts, streamflow, precipitation, temperature and other climate variables,” says Hosseinzadeh, who presents the research during USU’s 2024 Spring Runoff Conference March 26-27 at the Cache County Fairgrounds and Utah State’s Logan campus.
JOURNAL
Water Resources Research
METHOD OF RESEARCH
Computational simulation/modeling
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Spatiotemporal Data Augmentation of MODIS-Landsat Water Bodies using Adversarial Networks
Estimating coastal water depth from space via satellite-derived bathymetry
Researchers test a novel machine learning-based depth estimation technique on Korean coastal regions with unique characteristics
Since ancient times, knowing the depth of coastal waters has been key to safe and successful navigation and to exploit the sea’s resources. Today, bathymetry—the measurement of sea depth—is even more important, playing essential roles in our understanding of marine environments and the development of large marine structures.
With the development of shipborne echo sounders in the early 20th century, bathymetric surveys saw massive leaps in both accuracy and convenience. However, even with modern echo sounders, there are still many hardships to overcome when conducting bathymetric surveys. These include high cost, unpredictable weather, high ship traffic, and potential geographic or diplomatic issues, to name a few.
To address these issues, scientists around the world have been developing satellite-derived bathymetry (SDB) techniques, which estimate water depth from multispectral satellite images. These methods can sometimes produce accurate results, especially for depths up to 20 meters. Unfortunately, most SDB models were developed using data from coastal regions with clear waters and a uniform distribution of seabed sediment. Since light reflects differently depending on water turbidity and the composition of the seabed, developing SBD models with consistent performance throughout different coastal environments has proven challenging.
Against this backdrop, a research team from Korea has been developing a new SDB model that leverages machine learning to shed light onto the various factors that can compromise accuracy, thus paving the way to potential solutions. Their latest study, which included Dr. Tae-ho Kim from Underwater Survey Technology 21 (UST21), was published in the Journal of Applied Remote Sensing on March 12, 2024.
One of the main goals of this study was to analyze how the model trained on different coastal regions would be affected by each region’s unique characteristics. To this end, they selected three areas around the Korean Peninsula: Samcheok, characterized by its clear waters, Cheonsuman, known for its turbid waters, and Hallim, where the seabed contains various types of sediments.
The team obtained multispectral satellite data of these regions from the Sentinel-2A/B missions, openly provided by the European Space Agency, and selected multiple images of these areas at different time points with clear skies. To train the SDB model on these data, they also acquired echo sounder-derived nautical charts from the Korea Hydrographic and Oceanographic Agency (KHOA); these charts were used as ground truth.
The SDB model itself was based on a well-established theoretical framework that links how light coming from the Sun is reflected by the atmosphere, the sea, and the seabed before reaching a satellite. As for the machine learning part of the model, the team employed a random forest algorithm because of its ability to adjust to multiple variables and parameters while handling large amounts of data.
Upon training and testing region-specific instances of the SDB model, the researchers found that accuracy was generally acceptable for Samcheok, with a root-mean-squared error of about 2.6 meters. In contrast, accuracy was markedly lower for both Cheonsuman and Hallim, with satellite-based depth predictions deviating significantly from KHOA measurements.
To understand these discrepancies better, the researchers first tried correcting the predictions by including a turbidity index in the calculations. This improved results mainly for Cheonsuman. Then, to further investigate the sources of error, the team acquired high-resolution satellite images from the WorldView-3 mission, as well as on-site photos. Analyses revealed the reflectance characteristics of the seabed sediments had a large impact on depth estimations, with dark-colored basalt leading to a consistent overestimation. “If we incorporate additional seabed spatial data into the training dataset in the future, we anticipate enhancements in model performance,” said Dr. Kim. “A sediment distribution map, created from airborne hyperspectral imaging, is scheduled to be provided by R&D project.”
Finally, the researchers then tested the generalization capability of their approach by applying region-specific SDB models on other coastal areas with similar characteristics. “Unlike previous studies that presented SDB model results for waters with high transparency only, we developed individual SDB models that can be applied to waters with various characteristics, and suggested methods for obtaining improved results,” Dr. Kim said.
With any luck, these efforts will lead to improvements in SDB technology and pave the way for more convenient coastal depth mapping. Satisfied with the results, Dr. Kim concludes: “Ultimately, SDB results will be applied as depth monitoring data to facilitate safe ship passage in coastal areas, as well as input data for numerical ocean models, contributing to various scientific fields.”
Read the Gold Open Access paper by Jae-yeop Kwon et al., “Estimation of shallow bathymetry using Sentinel-2 satellite data and random forest machine learning: a case study for Cheonsuman, Hallim, and Samcheok Coastal Seas,” J. App. Rem. Sens. 18(1) 014522 (2024) doi 10.1117/1.JRS.18.014522.
JOURNAL
Journal of Applied Remote Sensing
METHOD OF RESEARCH
Observational study
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Estimating coastal water depth from space via satellite-derived bathymetry
ARTICLE PUBLICATION DATE
21-Mar-2024
Satellite data assimilation improves forecasts of severe weather
UNIVERSITY PARK, Pa. — In 2020, a line of severe thunderstorms unleashed powerful winds that caused billions in damages across the Midwest United States. A technique developed by Penn State scientists that incorporates satellite data could improve forecasts — including where the most powerful winds will occur — for similar severe weather events.
The researchers reported in the journal Geophysical Research Letters that adding microwave data collected by low-Earth-orbiting satellites to existing computer weather forecast models produced more accurate forecasts of surface gusts in a case study of the 2020 Midwest Derecho. Derechos are lines of intense thunderstorms notorious for their damaging winds.
“The computer model is able to produce a series of forecasts that consistently emphasize the most powerful storms and strongest wind damage at where it happened,” said Yunji Zhang, assistant professor in the Department of Meteorology and Atmospheric Science at Penn State and lead author. “If we have this kind of information in real time, before the events occur, forecasters might be able to pinpoint where the strongest damage is going to happen.”
The technique could be especially useful, the scientists said, in areas that lack ground-based weather monitoring infrastructure — like radars traditionally used in weather forecasting. In the study, the researchers only used data available from satellite observations.
“In regions where there are no surface observations, or basically no radar, we show that this combination of satellite observations can generate a decent forecast of severe weather events,” Zhang said. “We can probably apply this technique to more regions where there are no radar or dense surface observations. That’s the fundamental motivation behind this study.”
The research builds on the team’s prior work using data assimilation, a statistical method that aims to paint the most accurate picture of current weather conditions. This includes even small changes in the atmosphere as they can lead to large discrepancies in forecasts over time.
In prior work, scientists with Penn State’s Center for Advanced Data Assimilation and Predictability Techniques assimilated infrared brightness temperature data from the U.S. Geostationary Operational Environmental Satellite, GOES-16. Brightness temperatures show how much radiation is emitted by objects on Earth and in the atmosphere, and the scientists used infrared brightness temperatures at different frequencies to paint a better picture of atmospheric water vapor and cloud formation.
But infrared sensors only capture what is happening at the cloud tops.
Microwave sensors view an entire vertical column, offering new insight into what is happening underneath clouds after storms have formed, the scientists said.
“Just based on the cloud tops, it’s more difficult to infer what the convection of these storms looks like underneath,” Zhang said. “So that’s one of the benefits of adding in the microwave observations — they can provide information on where the strongest convections are.”
By combining assimilated infrared and microwave data in the study of the derecho, the researchers were able to predict surface gust locations and maximum wind values more accurately.
In future work, Zhang said he plans to apply the method to regions that lack the resources and infrastructure to support high-spatiotemporal-resolution weather observations.
“We know that there have been several times in the past several years in West Africa where very strong torrential rainfall events have brought on a lot of precipitation to those countries,” Zhang said. “And one thing about these countries is that they are also the places that will likely be impacted most by global warming. So I think if we can use these available satellite observations to provide better forecast for those regions, it will be really beneficial for the people there as well.”
Also contributing from Penn State were David Stensrud and Eugene Clothiaux, professors, and Xingchao Chen, assistant professor, all in the Department of Meteorology and Atmospheric Science.
NASA and the U.S. Department of Energy provided funding for this work.
JOURNAL
Geophysical Research Letters
METHOD OF RESEARCH
Computational simulation/modeling
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Enhancing Severe Weather Prediction With Microwave All-Sky Radiance Assimilation: The 10 August 2020 Midwest Derecho
ARTICLE PUBLICATION DATE
22-Mar-2024
No comments:
Post a Comment