To take advantage of the capabilities provided by Sentinel 2 and Landsat 8 data and advancements in image processing, with financial support from the University of Minnesota and Legislative and Citizens Commission on Minnesota Resources, and in collaboration with the Minnesota Supercomputing Institute, we have developed a highly automated, state-of-the-art system to download, process, and analyze satellite imagery for surface water quality.
The basic system design, illustrated above, includes:
- Automated downloading of Landsat 8 and 9 and Sentinel-2 imagery.
- Image processing includes atmospheric correction and removal of areas with cloud cover, haze, and smoke, terrestrial areas, and shallow areas where bottom or vegetation would affect the results
- Application of validated water quality models for clarity, chlorophyll, and CDOM.
- Provide the results in the Minnesota LakeBrowser (Established 2002) which is a dynamic database that allows access and visualization of the data agencies, researchers, and the public via an expanded and more user-friendly LakeBrowser. The data are also available on the University of Minnesota GEMS platform and the Minnesota DNR Watershed Health Assessment application.
Machine-to-machine access to ESA Copernicus and U.S. Geological Survey servers allows for the synergistic acquisition of S2/MSI and L8/OLI imagery to supply the demand for near-real-time data. Newly acquired imagery can be immediately sent through multiple scripted processing modules, which include (1) identifying and omitting potentially contaminated pixels caused by clouds, cloud shadow, atmospheric haze, wildfire smoke, and specular reflection, and (2) classification of water pixels through a normalized difference water index (nNDWI) to delineate a scene-specific water mask. The combined masks result in qualified pixels which advance to (3) a modified SWIR-based aerosol atmospheric correction to the retrieval of remote sensing reflectances (Rrs). The atmospheric correction produces a harmonized reflectance product between S2/MSI and L8/OLI pixels in which modeled L-3 type water quality data products are derived.
Calibrated L-3 water quality models including water clarity, CDOM, and chlorophyll-a, rely heavily on field-validated datasets to account for the dynamics of optically complex lake systems of the region. To this extent, sampling efforts in the summer months constrain uncertainties between satellite-derived and surface water properties caused by varying atmospheric conditions and calibrate/validate water quality retrieval algorithms to yield verifiable water products. As new field validation data become available at season-end, scripted modules within the processing chain can be modified accordingly and applied to incoming and previously processed imagery if any resulting water quality product models need improvement.
Finally, the data can be made available to the public in an online map viewer linked to a spatial database that allows for statistical summaries at different delineations and time windows, temporal analysis, and visualization of water quality variables. The Minnesota LakeBrowser provides an example of the data that are being produced through this project. Due to the cloud cover in the Midwest, we determined that monthly open water (May through October) pixel-level mosaics work best for statewide coverage. Lake level data is determined for each clear image occurrence and compiled in CSV files that can be used to calculate water quality variables for different timeframes (e.g., monthly, summer (June-Sept)) and linked to a lake polygon layer that can be used for geospatial analysis and included in a web map interface. For Minnesota, the lake level (2017-2020) data includes 603,678 daily lake measurements of chlorophyll, clarity, and CDOM (1,811,034 total) and will be updated regularly.
This unique data source dramatically improves data-driven resource management decisions and will help inform agencies about evolving water quality conditions statewide. In terms of decision-making, the production of frequent, near-real-time data on water clarity, chlorophyll-a, and CDOM across large regions can enable water quality and fisheries managers to better understand lake ecosystems. The improved understanding will yield societal benefits by helping managers identify the most effective strategies to protect water quality and improve models for increased fisheries production.
Work on this initiative began in 2015 with testing one model for all images in Google Earth Engine. The successful results led to the LCCMR project with the MSI and its high-performance computing capabilities, with full-scale operational capabilities in 2021 for near real-time monitoring with Landsat and Sentinel data. The results for the different water quality properties can be seen on the Metrics pages. The approach is being extended to additional areas including Michigan, and Wisconsin, and can ultimately be extended to the eastern USA where most lakes and reservoirs are located.