Integrating earth observation in hydrological modeling: A data and computing-intensive approach

As climate change accelerates, inducing urgent and unprecedented natural challenges, the significance of hydrology in addressing societal and environmental issues—such as food security, water resource management, and natural disasters—has increasingly come to the fore. However, many widely used hydrological models were developed in an era of less sophisticated technology, highlighting the need for upgrading. Recent decades have seen significant advancements in data measurements and production (e.g., satellites and UAVs), computing hardware (e.g., CPU/GPU), and algorithms (e.g., machine learning and parallel computing), offering substantial potential to revolutionize hydrological analysis. Yet, to fully leverage these advancements, current hydrological analysis tools face several limitations. First, our knowledge still has many gaps, which is inadequate for understanding and representing complex phenomena with intertwined processes like soil erosion and fate and transport of nutrients in agricultural fields. Second, there is a challenge in integrating diverse, long-term Earth observation datasets into existing models due to the inherent limitation of model configurations, limiting the potential of enhancement in model accuracy. Furthermore, analyses are constrained by scaling issues, necessitating compromises between model resolutions (spatial or temporal), computational time, and physical processes parameterization and scaling. To assess these challenges, this dissertation focused on integrating and implementing data-intensive and computing-intensive approaches into two cross-related hydrological responses: (1) soil erosion and (2) modeling hydrologic fluxes in the vadose zone. Specifically, it applied a maximum entropy machine learning approach to predict gully erosion susceptibility using LiDAR, high-resolution aerial imagery, and seasonal environmental variables to improve the susceptibility prediction of this phenomenon. Furthermore, an explainable stacking ensemble model to enhance gully erosion susceptibility predictions was developed using the Shapley additive explanation method (SHAP) to resolve model parameter interpretability. Also, the DyLEMa algorithm was developed targeting (1) the spatio-temporal data completeness on the State of Illinois scale of satellite-derived evapotranspiration data, (2) the reduction of uncertainty in ET prediction based on seasonal environmental variables, and (3) the computational scaling of the DyLEMa algorithm on a high-performance computing environment. Lastly, a physically based, fully distributed model has been devised to simulate infiltration and runoff generation in the vadose zone, designed to ingest long-term Earth observation data at moderate spatial resolution as inputs while using parallel computations to achieve feasible modeling performance.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Agricultural & Biological Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Han, Jeongho

Contributors dc:contributor

• Chu, Maria L.
• Guzman, Jorge A.
• Kumar, Praveen
• Malek, Keyvan

• Hydrological Modeling
• Earth Observation
• Machine Learning
• Gully Erosion
• Evapotranspiration
• Explainable Artificial Intelligence
• Remote Sensing

Statement dc:rights

• Copyright 2024 Jeongho Han

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125669

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125669