Research & Projects

Objectives

The primary technical objective of this project is to systematically quantify and characterize how uncertainties in the geophysical input data propagate through the PCA‐ and K‐means‐based clustering scheme and ultimately affect runoff prediction for each identified cluster in the Great Lakes region. In particular, we will (1) compile key geophysical datasets (e.g., precipitation, land use, soil characteristics, topography) and quantify their associated uncertainties, whether from measurement error or variability in data sources; (2) propagate these uncertainties through the dimensionality reduction and clustering pipeline by applying methods such as Monte Carlo simulation or bootstrap resampling; (3) evaluate the stability of the cluster assignments under different realizations of the input data, focusing on the likelihood of watersheds changing cluster membership; and (4) assess how these shifts in cluster membership translate into variability in predicted runoff outcomes. Through this process, we aim to identify which input variables most strongly influence cluster formation, determine the level of confidence in cluster‐based runoff predictions, and provide recommendations for improving data collection or refining the cluster analyses to achieve more robust outcomes.

Results

The project will produce a suite of uncertainty‐aware clustering outputs for HUC‐10 watersheds in the Great Lakes region, showing how each watershed’s assignment varies under different input realizations. These results will include probabilistic cluster membership maps that depict which regions have stable groupings versus those with higher classification uncertainty. Accompanying these maps will be an analysis of the sensitivity of the clustering to key geophysical factors, identifying which data sources most significantly impact cluster membership. Finally, a runoff prediction uncertainty report will summarize how cluster‐level runoff estimates shift as a function of input uncertainty and which clusters are most susceptible to large variability. Collectively, these products will inform watershed‐scale management decisions, highlight priorities for future data collection or model refinement, and provide a transparent framework for communicating the reliability of regionalization outcomes in complex hydrological settings.

Objectives

The goal of this project is to generate a new precipitation dataset that can outperform the existing datasets (Table 1). The State of Delaware will serve as the case study site for this research, given the region's susceptibility to climate change impacts and the associated risks of flooding and water quality degradation.

Results

1. Developing WRF-Hydrology models for the four selected domains.

2. Generate a new precipitation product that can outperform the existing five products (potentially leveraging Spatial-Temporal Vision Transformer models).

3. Explore approaches for evaluating the new precipitation product. For example, we can use the precipitation product as forcing to simulate streamflow and evaluate the simulation's goodness-of-fit against observed data.

[1] Fu, Z., Yu, G., Liu, M., Lee, C. T., & Lin, J. E. (2021, May). “Data Analysis of Influencing Factors of Air Quality in Chinese Cities.” 2021 6th International Conference on Big Data and Computing. (pp. 61-68).

[2] Yu, G., Fu, Z., Huang, Y., & Lee, C. T. (2021, August). “Data analysis of Higher Education System Based on TOPSIS and Time Series Model.” 2021 5th International Seminar on Education, Management and Social Sciences (ISEMSS 2021) (pp. 1014-1024). Atlantis Press.