Cheang et al. (2025) RUSH: Rapid Remote Sensing Updates of Land Cover for Storm and Hurricane Forecast Models

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-09-12
• Authors: Chak Wa Cheang, Kristin B. Byrd, Nicholas M. Enwright, Daniel Buscombe, Christopher R. Sherwood, Dean B. Gesch
• DOI: 10.3390/rs17183165

Research Groups

• U.S. Geological Survey Western Geographic Science Center, Moffett Field, CA, USA
• U.S. Geological Survey Wetland and Aquatic Research Center, Lafayette, LA, USA
• Applied Coastal Research and Engineering, Washington State Department of Ecology, Olympia, WA, USA
• U.S. Geological Survey Woods Hole Coastal and Marine Science Center, Woods Hole, MA, USA
• U.S. Geological Survey Earth Resources Observation and Science (EROS) Center, Sioux Falls, SD, USA

Short Summary

This study developed the RUSH (Rapid Remote Sensing Updates of Land Cover for Storm and Hurricane Forecast Models) tool, an open-source application that generates high-resolution (3 meter) coastal land cover maps from Planet SuperDove imagery. The tool provides near-real-time or historical land cover data with overall accuracies of 93% to 94%, crucial for improving hydro-morphological models used in hurricane impact forecasting.

Objective

• To develop a rapid-repeat, user-friendly, and open-source Jupyter Notebook application and graphical user interface (GUI) called RUSH, capable of delivering high-resolution (3 meter) maps of coastal vegetation for near-real-time conditions.
• To generate coastal land cover products tailored to the needs of advanced hydro-morphological models for forecasting coastal storm and hurricane impacts.
• To determine if a single, scalable remote sensing model of coastal land cover can be developed for multiple coastal regions.
• To streamline the retrieval and processing of satellite images for land cover mapping.
• To optimize the delivery of land cover mapping results to users.

Study Configuration

• Spatial Scale: 3 meter spatial resolution for land cover maps. User-defined Areas of Interest (AOIs) up to approximately 100 square kilometers. Study regions included the Outer Banks of North Carolina, the Mississippi River Delta in Louisiana, and a portion of the Florida Gulf Coast near Apalachicola.
• Temporal Scale: Near-daily coverage provided by Planet SuperDove imagery. Seasonal models (Cool Season and Warm Season) were developed to account for year-round vegetation phenology. Case study involved pre- and post-hurricane comparison (e.g., Hurricane Ida, August to October 2021).

Methodology and Data

• Models used: Random Forest machine learning model (Python scikit-learn package) for image classification. Two seasonal models: Cool Season Model and Warm Season Model.
• Data sources:
  • Planet Labs SuperDove multispectral imagery (3 meter spatial resolution, near-daily coverage, 8 spectral bands: red, green, blue, near-infrared, red edge, green 1, coastal blue, and yellow).
  • Reference data: 1331 manually labeled points (401 in Florida, 469 in North Carolina, 461 in Louisiana) derived from visual interpretation of U.S. Department of Agriculture National Agriculture Imagery Program (NAIP) orthoimagery (0.6 meter resolution) and the 2021 Coastal Louisiana Vegetation Types map.
  • National Oceanic and Atmospheric Administration (NOAA) Coastal Change Analysis Program (C-CAP) data (30 meter resolution) for defining AOI masks.
  • Calculated vegetation indices from Planet SuperDove bands: NDVI, NDWI, WDRVI2, WDRVI5, SR, DVI, GDVI, GNDVI, GRVI, IPVI, TBVI81, TBVI82, TBVI41, TBVI64, NDCI.

Main Results

• The RUSH tool, an open-source Jupyter Notebook application and GUI, was successfully developed to generate 3 meter resolution near-daily coastal land cover maps.
• The tool classifies land cover into five classes relevant to coastal settings: open water, emergent wetlands, dune grass, woody wetlands, and bare ground.
• The Cool Season Model achieved an overall accuracy of 93%, and the Warm Season Model achieved an overall accuracy of 94% when tested with independent reference points.
• Producer's and user's accuracies for vegetated classes ranged from 88% to 93%, while non-vegetated classes ranged from 93% to 98%. Dune grass had the lowest accuracy among vegetation classes (82%).
• Variable importance analysis showed NDVI as the most important variable for the Warm Season Model and Simple Ratio (SR) for the Cool Season Model.
• The random forest models demonstrated transferability across the three study areas (North Carolina, Louisiana, Florida) and different seasons, with location having low variable importance.
• A case study for Hurricane Ida (2021) in Grand Isle, LA, demonstrated the RUSH tool's capability to detect pre- and post-storm land cover changes, including shifts from vegetated areas to bare ground.
• Final outputs include raster land cover classification maps (traditional and cloud-optimized GeoTIFF) and accuracy assessment tables.

Contributions

• Developed RUSH, a novel, rapid-repeat, user-friendly, and open-source software solution for generating high-resolution (3 meter) coastal land cover maps from Planet SuperDove satellite imagery.
• Provided tailored land cover products with specific classes (open water, emergent wetlands, dune grass, woody wetlands, bare ground) directly applicable as inputs for advanced hydro-morphological models, addressing a critical need for up-to-date surface roughness data in hurricane and storm forecasting.
• Demonstrated the high accuracy (93-94%) and transferability of a single machine learning model across diverse U.S. coastal regions and varying seasonal phenologies, enhancing the scalability and utility of coastal remote sensing.
• Streamlined the complex process of satellite image retrieval and processing through an automated workflow, making high-resolution remote sensing capabilities accessible to both non-expert users via a GUI and advanced users via a Jupyter Notebook.
• Enabled near-real-time and historical assessments of coastal geomorphic and ecosystem changes at fine spatial and temporal scales, overcoming limitations of existing coarser resolution or less frequently updated land cover products.

Funding

• National Oceanographic Partnership Program (NOPP)
• U.S. Department of the Navy Office of Naval Research
• USGS Community for Data Integration
• USGS Coastal/Marine Hazards and Resources Program Remote Sensing Coastal Change project
• USGS National Land Imaging Program
• NASA’s Commercial Satellite Data Acquisition Program (for Planet satellite images)

Citation

@article{Cheang2025RUSH,
  author = {Cheang, Chak Wa and Byrd, Kristin B. and Enwright, Nicholas M. and Buscombe, Daniel and Sherwood, Christopher R. and Gesch, Dean B.},
  title = {RUSH: Rapid Remote Sensing Updates of Land Cover for Storm and Hurricane Forecast Models},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17183165},
  url = {https://doi.org/10.3390/rs17183165}
}