We investigate the impact of climate changes on regional to the global scale water cycle with remote sensing, land surface models, and AI/machine learning technics.
Our goal is to understand how water movement across the land surface relates to broader-scale behavior, and to improve the accuracy of natural disaster and water resource predictions by applying remote sensing and machine learning techniques in the context of the climate crisis.
우리 수문 원격탐사 인공지능 연구실(HydroAI)은 기후변화/인간 활동에 의한 환경의 변화와 자연재해/수자원 예측 및 분석을 위해 인공위성 및 수치모델 자료와 AI/기계학습 기술을 적극 활용합니다. 전 지구적 규모로 발생하는 자연재해(가뭄, 홍수, 산불, 황사 등) 및 수자원을 위성 영상을 통해 더욱 신속하고 정확하게 예측하는 것은 기후변화에 대응한 가장 중요한 연구 분야로 주목받고 있습니다. 우리 연구실에서 현재 진행하고 있는 프로젝트는 여기에서 확인이 가능합니다.
• 위성 원격 탐사, 수치모델, 기후변화, 환경 분야 Domain Scientist로 AI/기계학습 방법론 적용 연구에 대한 열정을 가지고 있는 학부생 및 대학원 진학 예정 학생을 환영합니다.
• (2025년도 하반기) 차세대 위성/드론을 이용한 환경 원격탐사에 경험 및 관심이 있는 학생을 모집합니다.(석/박 통합 TO 메일로 문의)
• 딥러닝 방법론 중 physics-informed neural networks (PINNs), partial convolution neural network (PCNN), super-resolution GAN (SRGAN), reinforcement learning (RL) 등을 활용한 위성 이미지 분석 및 정확도 향상, long short-term memory (LSTM), gated recurrent unit (GRU), Transfer Learning (TL) 을 활용한 각종 자연재해 예측, 지표 모델링 위한 loss function 개발에 관심이 있는 학생들을 환영합니다.
• 석사/박사 (한국)학생의 경우 여름/겨울 방학 기간에 미국/유럽 연구 기관 및 대학교에서 방문 연구 기회를 제공합니다. 배움에 열정이 있는 모든 학생을 환영합니다. 관심 있는 학생은 Contact 페이지를 통해 연락 바랍니다.
• If you are passionate about hydrology, remote sensing, and machine learning applications in Earth sciences, please feel free to contact me.
• Particularly, we are looking for students interested in satellite image analysis and accuracy improvement using deep learning (DL) methodologies such as PINN, PCNN, GAN, SRGAN, predictions of various natural disasters using LSTM, GRU, and the development of loss functions for environmental data analysis. We welcome students who have always been interested in applying AI/ML/DL methodologies in the civil/environmental field but couldn't learn them, as well as students who majored in AI/ML/DL but wanted to contribute to society through research in climate change and natural disaster prediction.
• We are currently interveiwing MS and one Ph.D. students who are interested in utilizing machine learning techniques to address water-related and climate change issues. Before reaching out to me, you can visit this page for what opportunities are currently available. I'm excited to work with the next generation of researchers in this exciting and rapidly evolving field.
We use shallow machine learning and deep learning approaches, along with the application of Bayesian theorem, to predict natural phenomena and improve the accuracy of our data.
We utilize various Earth observing satellite systems to monitor the dynamics of Earth's systems and gain insight into the global water cycle.
We utilize land surface models at various scales to comprehend physical processes and examine how the land surface and atmosphere interact with each other on the ground.
We integrate data from ground, satellites, and land surface models to improve the quality and quantity of the data, leading to a better understanding of Earth's processes.
Our research projects involve the study of the Earth's complex processes, and the interactions between the Earth's various components. The complexity, scale, and nonlinearity of these processes make machine learning (ML) an ideal tool for advancing our understanding and predictive capabilities. Earth science generates massive amounts of data from various sources, including satellite observations, ground-based measurements, and numerical simulations. ML algorithms can efficiently process, analyze, and extract valuable insights from these large, diverse, and complex datasets. In addition, we are devoted to understanding the uncertainty of the data using a Bayesian ML approach.
We use various shallow and deep ML approaches to predict water resources, natural disasters (e.g., floods and droughts), data assimilation, and explainable and Bayesian ML for the uncertainty analysis. Please check our relevant research projects below.
Bayesian machine learning
Streamflow prediction over ungaged regions
Floods and droughts prediction
Explainable machine learning for the climate change analysis
We use Earth observing satellite systems to collect comprehensive data sets, allowing us to monitor global changes in the environment and improve our understanding of climate, land use, natural disasters, and other environmental processes. We are currently focused on the following research projects.
Global-scale water cycle analysis
Spatial and temporal resolution improvments
Algorithm developments for new data
Data uncertainty quantification and prediction
Land surface models (LSMs) are computer models that simulate land surface processes, such as energy and water fluxes, carbon and nutrient cycles, and vegetation dynamics. They help researchers understand and predict how land surface processes respond to environmental changes like climate change and land use change. They operate at different scales: microscale, mesoscale, and global scale.
We use the Land Information System (LIS) to run various land surface models and combine them with other types of observational data to provide a comprehensive view of land surface processes. Please check our relevant research topics below.
Impact of human activities on the water cycle
Data assimilation (integration)
Global-scale water balance analysis
Our paper, titled “Assessing Land Cover-Specific Responses to Flash Droughts in the Mississippi River Basin,” published in Journal of Environmental Management (IF: 8, top 9%) by Ms. Bakar (visiting scholar) investigates how different land cover types respond to rapidly developing drought events from 2000 to 2022. By utilizing the Standardized Anomaly-based Precipitation Evapotranspiration Index (SAPEI) for drought detection and Gross Primary Production (GPP) data to assess vegetation recovery, we identified 315 flash droughts and analyzed the 10 most severe cases. Recovery times varied significantly, ranging from 8 to 120 days, with the most prolonged recovery periods observed during extreme drought years such as 2006, 2012, and 2022. Our results show that forested areas recover faster, whereas rain-fed agricultural lands exhibit the highest vulnerability to sudden moisture loss, particularly in the drier Upper Mississippi River Basin (MRB), where intensive farming practices exacerbate drought impacts. These findings underscore the necessity for improved drought management strategies, including optimized water resource allocation and the development of drought-resistant crops, to enhance resilience in vulnerable regions.
Link
Our paper, titled “Assessing Land Cover-Specific Responses to Flash Droughts in the Mississippi River Basin,” published in Weatehr and Climate Extreme (IF: 6.1, top 9%) by Ms. Bakar (visiting scholar) examines how sudden and intense droughts influence river discharge across the region from 1980 to 2022. By applying the Standardized Anomaly-based Precipitation Evapotranspiration Index (SAPEI), we identified over 1,000 flash droughts and revealed distinct regional patterns in their occurrence. The eastern Mississippi River Basin (MRB) experiences frequent but short-lived droughts, while the northwest endures fewer but more prolonged events. The southern MRB faces the most severe drought conditions, largely influenced by upstream water use. Our findings establish a strong relationship between flash droughts and reduced river flow, demonstrating SAPEI’s effectiveness in monitoring these hydrological impacts. This research provides valuable insights for improving water resource management and developing strategies to mitigate future drought risks.
Link
Our paper, titled “Quantifying Spatiotemporal Satellite Soil Moisture Error for Improved Land Data Assimilation,” pbulisehd in GIS&RS (IF:6, top 8%) by Ms. Subin Kim (MS/PhD combined program) presents a novel framework that integrates triple collocation analysis (TCA) with advanced machine learning techniques, such as Light Gradient Boosting Machine (LGBM) and Deep Neural Networks (DNN), to estimate time-variant satellite-based soil moisture (SM) error dynamics on a global scale. By developing stand-alone TCA-based SM error prediction models that rely solely on data used for Soil Moisture Active Passive (SMAP) retrievals, we demonstrate the ability to recover error information in regions where SMAP SM error data were previously unreliable due to uncertain error characteristics. The results indicate that our approach outperforms models that incorporate external datasets, such as the Global Land Data Assimilation System (GLDAS), highlighting its potential for improving land data assimilation (LDA) applications. Furthermore, the methodology can be extended to other satellite-based geophysical datasets, providing broader applications for Earth system modeling and climate monitoring.
Link
Our paper, titled “Over 60% precipitation transformed into terrestrial water storage in global river basins from 2002 to 2021,” published in Communications Earth & Environment (Nature) (IF:9, Top 4%), provides a novel quantitative assessment of the transformation of daily precipitation into terrestrial water storage across 121 global river basins over a two-decade period. The study introduces the average daily fraction of precipitation transformed into terrestrial water storage, leveraging enhanced terrestrial water storage statistical reconstruction and water storage data from the Gravity Recovery and Climate Experiment (GRACE) and its follow-on mission. We reveal that approximately 64% of land precipitation contributes to terrestrial water storage, with notable variations across different climatic and geographical regions. These findings offer critical insights into the interactions between precipitation, land surface processes, and climate change, providing valuable implications for hydrological modeling and future water resource management.
Link
Our paper, titled “From theory to hydrological practice: Leveraging CYGNSS data over seven years for advanced soil moisture monitoring,” published in Remote Sensing of Environment (IF;11, Top 1.7%), provides a systematic review of the CYGNSS mission’s seven-year data record for soil moisture (SM) monitoring. The study underscores the significant potential of the Cyclone Global Navigation Satellite System (CYGNSS) mission, initially designed for tropical cyclone monitoring, in capturing fine-scale SM variability through its high revisit frequency and unique L-band bistatic radar technology. By analyzing real-world data, we identify and address critical challenges related to retrieval algorithms and the distinctive data characteristics of CYGNSS.
Link
Our paper, titled “Improving weather forecast skill of the Korean Integrated Model (KIM) by assimilating SMAP soil moisture anomalies,” published in the Quarterly Journal of the Royal Meteorological Society, demonstrates how incorporating SMAP soil moisture data enhances global soil moisture estimates and improves KIM weather forecasts. By using advanced bias correction methods, the study shows significant improvements, particularly in northern Africa and West-Central Asia, with benefits extending up to the 700 hPa atmospheric level.
Link
We are thrilled to announce that Ms. Aigerim Bolatbekkyzy (undergraduate-master combiend student at HydroAI lab) has been selected as one of the winners of the prestigious AOGS2024 Best Student Poster Award. Her exceptional research, titled “Comparative Analysis of the 2021-2022 Droughts in Kazakhstan, South Korea, and the USA Using Remote Sensing and Reanalysis Data,” stood out among numerous entries for its insightful analysis and contribution to the field. This well-deserved recognition highlights Aigerim’s dedication and hard work in advancing our understanding of drought impacts across different regions!
Link
Dr. Wagner's recent paper titled "Global Scale Mapping of Subsurface Scattering Signals Impacting ASCAT Soil Moisture Retrievals" in the IEEE Transactions on Geoscience and Remote Sensing. This study identifies subsurface scattering as a significant and previously misunderstood source of error in ASCAT soil moisture retrievals, revealing its widespread occurrence in both arid and humid regions with dry seasons, and recommends masking affected data to improve accuracy.
Link
Dr. Xing's recent paper titled "Seasonal-scale intercomparison of SMAP and fused SMOS-SMAP soil moisture products" in the Frontiers in Remote Sensing. This study evaluates the performance of the Fused-IB and SMAP-E soil moisture products, showing that Fused-IB generally outperforms SMAP-E, especially in forests, across various seasons from 2016 to 2020, with seasonal variations in accuracy linked to vegetation growth and rainfall effects, providing insights for improving soil moisture monitoring algorithms.
LinkDr. Spennemann's recent paper titled "Evaluation of a Combined Drought Indicator against Crop Yield Estimations and Simulations over the Argentine Humid Pampas" in the Theoretical and Applied Climatology. This study evaluates the Combined Drought Indicator (CDI) in Argentina’s Humid Pampas, showing its effectiveness in tracking drought impacts on agriculture despite some limitations, and underscores the importance of improving drought early warning systems given the increasing drought frequency in southern South America.
LinkDr. Le's recent paper titled "A Framework on Utilizing of Publicly Availability Stream Gauges Datasets and Deep Learning in Estimating Monthly Basin-scale Runoff in Ungauged Regions" in the Advances in Water Resources. This study presents a framework using LSTM models for monthly runoff prediction in South Africa and Central Asia, showing that models trained on data from other continents can outperform GLDAS datasets, with prediction accuracy depending on hydrological similarities between regions.
Link
Ms. Pavur's recent paper titled "Spatial Comparison of Inland Water Observations from CYGNSS, MODIS, Landsat, and Commercial Satellite Imagery" in the IEEE Geoscience and Remote Sensing Letters. This study compares inland waterbody extent and location data across the contiguous United States in 2019 using CYGNSS, MODIS, and Landsat, revealing that CYGNSS data, when refined with binary thresholds, improves accuracy and shows water area estimates within 2.3% of those from MODIS and Landsat, highlighting similarities and discrepancies between optical and radar-based satellite observations.
Link
Dr. Crow's recent paper titled "Systematic modelling errors undermine the application of land data assimilation systems for hydrological and weather forecasting" in the Journal of Hydrometeorology (IF:4.9; Q1). This study reveals that biases in the coupling strength between water state and water flux in land data assimilation systems can significantly impact the accuracy of water flux estimates, even when soil moisture precision improves.
LinkDr. Zhang's recent paper titled "Temporal Gap-filling of 12-hourly SMAP Soil Moisture over the CONUS using Water Balance Budgeting" in the journal Water Resources Research (IF:6.2; Q1). The study presents a method to accurately fill gaps in satellite soil moisture data using observed precipitation and a hydrologic loss function.
Link
Dr. Kim's recent paper titled "Interpreting Effective Hydrologic Depth Estimates Derived from Soil Moisture Remote Sensing: A Bayesian Non-Linear Modelling Approach in the journal Science of The Total Environment (IF:9.8; Q1). This study examines the challenges in deriving physically interpretable ΔZ values from the water balance equation using remotely sensed data, emphasizing the potential biases and recommending ΔZ be treated as an effective parameter reflective of the fitted datasets.
LinkDr. Kim's recent paper titled "True Global Error Maps for SMAP, SMOS, and ASCAT Soil Moisture Data Based on Machine Learning and Triple Collocation Analysis" has been published in the journal Remote Sensing of Environment (IF:13.8; Q1). This study employs machine learning to fill spatial gaps in assessing satellite-based soil moisture data accuracy, providing complete error maps for SMAP, SMOS, and ASCAT systems, and revealing the influence of environmental conditions on retrievals, ultimately enhancing our understanding of soil moisture dynamics across these satellite missions.
LinkDr. Kim's recent paper titled "Analyzing Error Characteristics of Satellite-Derived Soil Moisture Data Using Bayesian Inference" has been published in the journal Remote Sensing of Environment (IF:13.8; Q1). The study introduces a Bayesian approach to examine error factors and highlights the importance of considering multiple environmental factors and human activities in assessing the accuracy of satellite-based soil moisture data.
LinkDr. Kim has been invited to become an associate editor for the prestigious hydrological journal, Vadose Zone Journal. This is an incredible opportunity to contribute to the scientific community by helping to review and publish the latest research in hydrologic research field. As an associate editor, he will have the chance to work with esteemed colleagues, stay up-to-date on the latest advancements in this field, and help guide the direction of scientific research.
Link
Dr. Zohaib's paper (which Dr. Kim corresponded on) characterizing errors in satellite-based soil moisture data now out in IEEE Transactions on Geoscience and Remote Sensing (IF: 8.2; Q1).
LinkDr. Kim has been recognized as one of the top 20 reviewers for Remote Sensing of Environment (RSE) for the calendar year of 2022. He has been recognized for his contributions to advancing scientific knowledge in the field of remote sensing.
Link
Mr. Saeedi paper (which Dr. Kim collaborated on) estimating precipitation from microwave satellite systems now out in Remote Sensing of Environment (IF: 13.8; Q1).
LinkOur research team is committed to staying up-to-date in the field of Earth science by attending at least two international conferences each year, while also contributing to hydrologic research fields through publishing in top-tier journals.
The most recent symposium paper (HydroML 2023 at Lawrence Berkeley National Laboratory)
This study highlights the significance of characterizing errors in satellite-basedsoil moisture (SM) data, crucial for numerous Earth Science and EnvironmentalEngineering applications.
The most recent symposium paper (HydroML 2023 at Lawrence Berkeley National Laboratory)
Human activity impacts the water cycle speed, altering precipitation patterns and water availability, which is crucial to inform sustainable water management practices and mitigate climate change impacts for the well-being of our planet and its inhabitants.
The most recent peer-reviewed publications
This study compared soil moisture estimates using NASA's Aqua satellite with the AMSR2 instrument across different conditions. Results suggest selective use of data from different wavelengths can improve signal-to-noise ratio, with C-band products outperforming X-band products in vegetated areas and vice versa in barren lands.
The most recent conference abstracts
We participated in the 2022 AGU Fall Meetings and delivered successful presentations on diverse research topics, such as machine learning applications in hydrology, natural disaster prediction, and biases in land surface modeling. If you are interested in learning more about these topics or our research, feel free to contact us.
If you have a keen interest in the intersection of climate change and its impact on hydrological research fields, I encourage you to consider pursuing a Master's, PhD, or postdoctoral position. By delving deeper into this critical area of study, you can play an essential role in addressing the world's most pressing environmental challenges and help safeguard our water resources, ecosystems, and communities. Your dedication and expertise can significantly contribute to the development of sustainable solutions and innovative approaches to hydrological research. Embark on this exciting journey and become part of the passionate community of scientists working towards a more resilient and environmentally responsible future.
