ABSTRACT:

This resource contains the Jupyter Notebook for a surrogate model (SM) designed to improve the accuracy of conceptual flood inundation mapping (FIM) by mimicking the behavior of a high-fidelity hydrodynamic model.
The study leverages the National Oceanic and Atmospheric Administration (NOAA) Office of Water Prediction's (OWP) Height Above Nearest Drainage-Flood Inundation Mapping (HAND-FIM) method as the baseline framework, which enables large-scale flood mapping at low computational costs.
The surrogate model (SM) employs machine learning techniques, specifically a Random Forest algorithm, to replicate critical hydrodynamic characteristics derived from a 2D Hydrologic Engineering Center-River Analysis System (HEC-RAS) model, a high-fidelity representation of flood dynamics. This approach enhances the fidelity of the simpler HAND-FIM by infusing it with insights from the more detailed HEC-RAS model. The model development is applied across 14 carefully selected sub-watersheds in the Amite River Basin.

Key features of the resource include:

- Surrogate Model: Demo code built using the Random Forest algorithm to predict flood extents based on various input features.
- Input Features: These include the HAND-FIM generated for a historic flood event in August 2016 using the National Water Model (NWM) streamflow and the Office of Water Predictions (OWP) HAND-FIM Synthetic Rating Curves (SRC). Other inputs, such as the Digital Elevation Model (DEM), slope, aspect, land use/land cover (LULC) data, impervious surface data, and river network information (RiverNet).
- Target Data: High-resolution flood extent data from the HEC-RAS model (HEC-RAS-FIM), serving as the ground truth for training the surrogate model.

Contribution:
This work demonstrates how a data-driven, machine-learning approach can bridge the gap between conceptual flood models and hydrodynamic models, improving the accuracy and reliability of large-scale flood mapping while maintaining computational efficiency. The surrogate model mimics complex relationships between terrain, hydrology, and flood extents, making it a powerful tool for flood prediction in regions where detailed hydrodynamic modeling may be too costly or time-consuming.

This research was conducted as part of the NOAA National Water Center Innovators Program, Summer Institute 2023.

ABSTRACT:

This is for the manuscript "Assessing the Effectiveness of Reservoir Operation and Identifying Reallocation Opportunities under Climate Change". Climate change will alter hydroclimatic variability, bringing a set of challenges to existing water management. It remains unclear if current water infrastructure and operational strategies will still be effective in the future. In this study, using 21 federal reservoirs in Texas as examples, we develop data-driven models to represent current reservoir operations and assess their effectiveness under future scenarios. We further explore adaptive strategies for improving water supply reliability without increasing flood risk.

ABSTRACT:

Vegetation plays a crucial role in atmosphere-land water and energy exchanges, global carbon cycle and basin water conservation. Land Surface Models (LSMs) typically represent vegetation characteristics by monthly climatologic index (e.g., green vegetation fraction GVF, leaf area index). However, static vegetation parameterization does not capture dynamic-varying vegetation characteristics, such as responses to climatic fluctuation, long-term trend and interannual variability. This study developed a machine learning accelerated approach to quantify the impacts of dynamic-varying vegetation on the magnitude, seasonality, and biotic and abiotic components of hydrologic fluxes. A deep learning-based surrogate of Noah provided a rapid diagnostic tool to fuse GVF from seven remotely sensed products into LSM. Using the Upper Colorado River Basin (UCRB) as a test case, we found that dynamic-varying vegetation provides more buffering effect to climate fluctuation than the static vegetation configuration, leading to higher total evapotranspiration (thus lower water yield) and smaller evapotranspiration interannual variability. In addition, dynamic-varying vegetation from multi-source remote sensing products consistently predicts larger evaporation abiotic components (e.g., soil evaporation), which are partially compensated by smaller evaporation biotic components (e.g., transpiration). Based on the hydrologic sensitivity analysis to vegetation, we found that vegetation removal in the sparsely vegetated sandy soil regions of the UCRB would lead to the most effective water yield increase. This study highlights the importance of explicit representation of vegetation dynamics in climate change and land management assessment.

ABSTRACT:

Reservoirs are the key hydraulic infrastructure that regulates natural streamflow variability to fulfill various operation targets, including flood control, water supply, hydroelectricity generation and sustaining environmental flow. As an important anthropogenic interference in the hydrologic cycle, reservoir operation behavior remains challenging to be properly represented in hydrologic models, thus limiting the capability of predicting streamflow under the interactions between hydrologic variability and operational preferences. Data-driven models provide a promising approach to capture relationships embedded in historical records. This dataset contains historical daily operations of over 300 major reservoirs across the Contiguous United States with a wide range of streamflow conditions, including inflow, release, storage, elevation, etc. The eastern reservoir data is collected by Duke University (https://nicholasinstitute.duke.edu/reservoir-data/, Patterson et al., 2018. The western reservoir data is accessed via the United States Bureau of Reclamation (https://water.usbr.gov/api/web/app.php/api/).