Coastal Erosion Worknight Check-in April 10th

Check in on Project Goals, Short Term Goals for the Hackathon, and the Resources Available.

Summary of Short Term Goals

Original Research Project had Two Parts:

1. Build a Classifier to classify habitat types in the AOI
2. Use NDRI to assess flooding in the AOI

DataKind’s Next Steps Have Two Directions:

Option 1: Solutionize existing work to prep for coming hurricane season.

• Skips the classification step and relies on the existing AOI

Option 2: Expand to Additional Areas

1. Requires rebuilding the classifier model so that we can use it to classify new zones.

Data Scoping Progress

1. Investigating Satellite imagery sources
1. Planet Imagery
2. Other Options: Sentinel-2 :
2. Investigating Toolkits available to work with Satellite Imagery
• ArcGIS Pro
• Other Options: ArcGIS Online
• Other Options: Planet explorer
• Other Options: Python Integrations
3. Investigating Toolkits for Image Classifiers
• Azure GeoAI
• Other Options: TBS
4. Storing Data
• Azure Blob
• Other Options: TBS

1. Investigating Satellite imagery sources

Planet Data

The original report used 3m resolution Planet Data for building the classifier.

Trial Account:

• Does not support API of AOI
• Limits downloads

​

Not Feasible for DataKind Volunteers to Work with for Classifiers.

Sentinel Data

Resolution and frequency are not as strong as Planet

Still supports the bands to perform NDRI analysis

Monthly Accounts allow for,

• API access
• Need to investigate limits. (300 req/min (50.000 per month , need to understand the size limiting)
• Cost is 20$ a month for basic user

Strong contender for source of flooding information in existing AOIs for the coming hurricane season.

2) Investigating Toolkits available to work with Satellite Imagery

1. ArcGIS Online
• Super easy Interface, great tool for delivery.
• Online so no hardware limitations,
• 30 day trial
• Large Shapefiles need to be hosted publicly
2. ArcGIS Pro through Microsoft DSVM
• Need to Continue to investigate Azure DSVM, but requires a Credit Card for the trial
3. QGIs
• Current being scoped, very promising.
4. Planet explorer
• Still in Beta
• I did not have success adding my own shapefiles
5. Python Integrations with Ipython Notebooks
• Not GUI for viewing basemaps or searching for Imagery (So not great for beginners in satellite imagery)
• Most DataKind members are familiar with it

3) Investigating Toolkits for Image Classifiers and

4) Where to Store this Data

*Work in Progress*

1. Azure GeoAI using Azure Blob Storage
1. Requires paying for Servers
2. Other Classifiers
• There are many Datakind Members with experience here, but We have not scoped with this crew
3. Need to scope where data would be stored for QGIS

​

Summary of Short Term Goals for DataDive

DataKind’s Next Steps Have Two Directions:

Option 1: Solutionize existing work to prep for coming hurricane season.

• Skips the classification step and relies on the existing AOI
• Relies on Existing Classification Work
• Assumes Sentinel will not cap image downloads

Option 2: Expand to additional Areas

• Requires rebuilding the classifier model so that we can use it to classify new zones. Because Planet Data is not available, it is an open question on if Sential Data can be used
1. Original Team deemed that the frequency and resolution were not ideal
2. Requires an more in depth knowledge of satellite data

Questions for Audubon:

1. What is your highest priority- New Area’s Classified into Habitats or Knowing Where Flooding is in existing habitats
2. Are there existing relationships with Planet Imagery that would allow Audubon / Datakind continued access to imagery data?
3. Are there existing relationships with Sentinel Hub that would allow Audubon / Datakind continued access to imagery data?
4. What deliverable would make the adoption of this technology the easiest for your organization? Who is the intended user?

​