Near-Real-Time Monitoring of Key Reservoir Variables by Integrating Wide-Swath SWOT Altimetry and Deep Learning Techniques
Research Team Members
Objective:
Provide high-temporal-resolution estimates of key reservoir variables in a near-real-time framework across the contiguous United States to support NOAA's national reservoir operation models: water levels, surface areas, storage volumes, inflows, and outflows.
Approach:
The project employs Wide-Swath SWOT Altimetry, Multi-sensor Satellite Observations, and Deep Learning Techniques to derive high-temporal-resolution estimates of reservoir water levels, surface areas, storage volumes, inflows, and outflows.
1. Reservoir Inventory Creation: We will compile a consistent, georeferenced national inventory of reservoirs using Sentinel-1 SAR data.
2. Reservoir Rating Curve Construction: We will construct reservoir-specific surface area-water level rating curves based on the SWOT satellite's unique capability for simultaneous measurements.
3. Water Extent and Width Derivation: Algorithms and software tools will be developed to derive reservoir water extent, inflow, and outflow river widths from SAR (Sentinel-1A, Sentinel-1C, and NISAR) and optical (Landsat-8/9 and Sentinel-2A/B) satellite imagery.
4. Deep Learning Model Development: A Graph Neural Network (GNN) deep learning model will be developed. This model will estimate, hindcast, and forecast daily reservoir water level, surface area, and volume change.
Impact:
This project will substantially enhance the calibration and accuracy of NOAA’s national reservoir operation models and extend the applicability of these models to encompass all reservoirs in the contiguous United States, including the vast number of ungauged ones, thereby broadening their utility.Abstract:
This project focuses on developing and implementing cutting-edge algorithms, software tools, and corresponding data products for monitoring key reservoir variables across the contiguous United States. Our goal is to provide high-temporal-resolution estimates of reservoir water levels, surface areas, storage volumes, inflows, and outflows in a near-real-time framework. These data products will substantially enhance the calibration and validation processes of NOAA's national reservoir operation models. Crucially, this project will enable the expansion of these operation models to a vast number of ungauged reservoirs throughout the country, thus broadening their applicability and utility. To realize this goal, we will initially compile a consistent, georeferenced reservoir inventory using Sentinel-1 SAR data at 10 m spatial resolution. Leveraging the SWOT satellite’s unique capability, we will construct reservoir surface area-water level rating curves and river width-water level rating curves for the reservoir’s inflow and outflow rivers. With these established rating curves, we will be able to convert reservoir surface area measurements and river width measurements obtained from regular SAR (Sentinel-1A, Sentinel-1C, and NISAR) and optical (Landsat-8/9 and Sentinel-2A/B) satellite image observations to corresponding water level estimations, thereby greatly improving the temporal resolution of water level measurements for both reservoirs and their inflow/outflow rivers. We will also use SWOT wide-swath altimetry observations, nadir-looking satellite altimetry satellites (Jason-3, Sentinel-3A/B, Sentinel-6, and ICESAT-2), and indirect water level measurements from SAR and optical satellite image data to develop an innovative graph deep learning model. This model will be capable of estimating, hindcasting and forecasting the water level, surface area, and volume change of reservoirs and the water levels and discharges of their inflow and outflow rivers. The accuracy and reliability of these estimates and forecasts will be rigorously evaluated.