Dataset: Manually annotated reef halos based on sattelite imagery from 6 study areas as training and test data for a deep learning model

Final no updates expectedDOI: 10.26008/1912/bco-dmo.932211.1Version 1 (2024-12-18)Dataset Type:model results

Principal Investigator, Contact: Elizabeth Madin (University of Hawaiʻi at Mānoa)

Scientist: Simone Franceschini (University of Hawaiʻi at Mānoa)

BCO-DMO Data Manager: Karen Soenen (Woods Hole Oceanographic Institution)


Project: CAREER: Decoding seascape-scale vegetation patterns on coral reefs to understand ecosystem health: Integrating research and education from organismal to planetary scales (Coral Reef Halos)


Abstract

Reef halos are rings of bare sand that surround coral reef patches. Halo formation is likely to be the indirectly result of interactions between relatively healthy predator and herbivore populations. To reduce the risk of predation, herbivores preferentially graze close to the safety of the reef, potentially affecting the presence and size of the halo. Reef halos are readily visible in remotely sensed imagery, and monitoring their presence and changes in size may therefore offer clues as to how ...

Show more

Our study area included 20 areas of interest) from 6 countries.

SkySat satellite images were acquired through Planet Inc. Planet Explorer Catalogue. Obtained as a SkySat Collect product, each image was roughly 20km x 5.9km, with a spatial resolution ranging between 0.5m and 0.8m. Before download, all images were orthorectified, radiometrically calibrated, and atmospherically corrected to surface reflectance following Planet’s standard procedures. All surface reflectance products are orthorectified using fine digital elevation maps (30-90m posting) and ground control points. Planet conducts atmospheric correction with the 6SV2.1 radiative transfer model which accounts for atmospheric absorption and scattering, with aerosol optical depth, water, vapor and ozone inputs from MODIS near-real-time data (MOD09CMA and MOD09CMG). All calculations were done in R. Satellite images cover a time interval from March 2019 to June 2021. The satellite images can not be shared due to file size issues and data-sharing policies, below are the Skysat unique identification numbers for each satellite image used in the project:

  • 20201107_025354_ssc9_u0001
  • 20190510_152255_ssc4_u0003
  • 20190625_182659_ssc7_u0001
  • 20190309_183114_ssc11_u0001
  • 20190531_182931_ssc6_u0001
  • 20200919_182915_ssc7_u0002
  • 20190626_182402_ssc8_u0003
  • 20201129_191646_ssc8_u0001
  • 20201022_191415_ssc6_u0001
  • 20201022_191415_ssc6_u0002
  • 20200626_192104_ssc6_u0001
  • 20200626_192104_ssc6_u0002
  • 20200627_190037_ssc7_u0001
  • 20200627_191435_ssc8_u0001
  • 20190413_081243_ssc13d3_0004
  • 20190413_081243_ssc13d3_0005
  • 20190413_081243_ssc13d3_0006
  • 20200202_110140_ssc7_u0001
  • 20210627_034355_ssc19_u0001
  • 20210127_154738_ssc12_u0001

 

All non-overlapping halos in the AOIs were labeled, resulting in 4,127 manually annotated halos. Halos were labeled using ArcGIS software (ver. 2.9.1), allowing the geo-referenced information for all objects to be retained. To avoid biases due to a single user's perception of halos size, five users were trained to the same labeling procedure – zooming into each halo at a 1:600 scale and tracing light contours with a mouse – and labeled the same AOIs. The dataset generated from the imagery annotation was divided into training and test sets (~70% and ~30%, respectively). The training set was used for model implementation and optimization, while the test set was used for comparing model-predicted vs. manually annotated halos. In addition, we selected independent areas (i.e., AOIs) where no halos were used for the training process) for the test set to estimate the model generalization properties better.

Simone Franceschini and Amelia C. Meier downloaded the satellite images used for this project. Halos data were labeled by Simone Franceschini, Amelia C. Meier, Aviv Suan, Kaci Stokes, and Elizabeth M.P. Madin. Simone Franceschini developed the model and estimated performance metrics.

 


Related Datasets

IsSourceOf

Dataset: Mask R-CNN and U-Net models and reef halo ouput calculations
Relationship Description: Mask R-CNN and U-Net model using the manually annotated reef halos as input.
Madin, E., Franceschini, S. (2024) Mask R-CNN and U-Net model and output of coral reef halo measurements based on global multispectral satellite imagery. Biological and Chemical Oceanography Data Management Office (BCO-DMO). (Version 1) Version Date 2024-12-18 doi:10.26008/1912/bco-dmo.943698.1

Related Publications

Results

Franceschini, S., Meier, A. C., Suan, A., Stokes, K., Roy, S., & Madin, E. M. P. (2023). A deep learning model for measuring coral reef halos globally from multispectral satellite imagery. Remote Sensing of Environment, 292, 113584. https://doi.org/10.1016/j.rse.2023.113584