Geophysical Observations Toolkit for Evaluating Coral Health (GOTECH)

Andrew Wang, Joseph Quinn Stank, and Kareem Naguib

Practicum Project for the Master of Science in Analytics at the Georgia Institute of Technology

The Geophysical Observations Toolkit for Evaluating Coral Health (GOTECH) project seeks to use machine learning models to infer, from CALIPSO imagery, vitality properties upon satellite pass. The key to GOTECH is to establish a combined data source and data fusion algorithms to pull together public reef databases..

Final Report and Presentation:

Final Paper (.pdf)Download
Final Presentation Slides (.pptx)Download

The GOTECH Team is comprised of Georgia Institute of Technology (Georgia Tech) Masters in Analytics students Andrew Wang, Kareem Naguib, and J. Quinn Stank. Both NASA and Coral Vita provided expert mentorship and support to the team to guide analysis.

Background

Marine Experts Foresee numerous challenges which will affect Coral Reef health globally, many include:

coral_health.jpg

What We Did:

GOTECH built a tool that utilizes NASA CALIPSO satellite imagery to early identify coral reef habitats which are diseased or failing in order to inform decisions leading to effective funding and efficient action for organizations such as Coral Vita. The table below explains the 4 Technical Challenges faced by the team.

image-1Results

Technical Area 1 and 2: Cross-Validate Open Source Reef Databases then Time/Geo-Align with Corresponding CALIPSO data

image-2
Open source data was collected from 8 resources and combined into our Fused dataset. Each source was used to answer one of two questions: 1) Where is the coral? 2) What is the health of the coral? Once collected, the data was filtered and aligned by both temporal and locational parameters.

Technical Area 3: Correlate Imagery Backscatter with Coral Vitality

image-3
After testing several different models, the two best initial models were a 1D Convolutional Neural Network and a Feed Forward Neural Network. Please note the best accuracy was only ~75%.

Technical Area 4: Interpret Trends in a Known Global Region based on Backscatter Alone

4
The team then applied model tuning based off the dataset and model parameters to improve results. At the request of NASA, only CALIPSO backscatter data was supposed to be used for modelling.

Key Takeways and Future Work

image-5