Wind-dependent Variables: Predict Wind Speeds of Tropical Storms

Throughout a tropical cyclone, humanitarian response efforts hinge on accurate storm intensity estimates. Using satellite images assembled by Radiant Earth Foundation and the NASA IMPACT team, can you estimate the wind speeds of storms at different times? #disasters

$13,000 in prizes
feb 2021
727 joined

Open Data

Good news! The project team from this competition has made the challenge data available for ongoing use, practice and learning. This data will be maintained by the project team and released at their discretion.

For additional information about the data, see the Problem Description page.

Please note that we won't be able to field questions about the data, but we wanted to share it here for the benefit of the community. If you have a question, feel free to post it to the forum.

Tropical Cyclone Wind Estimation Competition Dataset

About Radiant MLHub

Radiant MLHub is an open repository for geospatial machine learning training data. Radiant MLHub hosts open training datasets generated by the Radiant Earth Foundation team as well as other training data catalogs contributed by Radiant Earth’s partners. All the data are hosted in a cloud-friendly format, and the API for allows easy discovery and download. Radiant MLHub is open to anyone to access existing training data and/or share their training data for broader impact.

Data Structure

The Radiant MLHub API is a STAC compliant API that serves metadata about label items and source imagery and links to download these items.

A SpatioTemporal Asset Catalog (STAC) is a standardized specification for organizing metadata, making it easy to search for images or labels that match spatial, temporal, or other criteria. At the root level of the STAC API is a list of collections of items. In the Radiant MLHub API, each collection contains items for either source imagery or labels for a dataset. These items are descriptions of source imagery or labels and links to download assets related to these items. Properties found in these item descriptions include spatial extent, temporal extent, band descriptions in the case of optical imagery, label types and label properties in the case of labels, and other information like Digital Object Identifiers (DOIs) and citation examples to reference.

To learn more about Radiant MLHub API, check out this blogpost on Accessing and Downloading Training Data on the Radiant MLHub API.

Authenticating with the API

To access the Radiant MLHub API, you must be authenticated with an API key. Requests made to the API must contain a query parameter where the key is “key” and the value is your API key. For example, a request made to the /collections endpoint would look similar to this:

You can obtain an API key by creating a free account on Radiant Earth Foundation's dashboard and navigating to the “API Keys” tab.

Accessing the data

There are three collections that contain data for this competition:

  • nasa_tropical_storm_competition_train_source contains the train images (jpegs) and metadata (jsons, one per image)
  • nasa_tropical_storm_competition_train_labels contains the train labels (jsons, one per image)
  • nasa_tropical_storm_competition_test_source contains the test images (jpegs) and metadata (jsons, one per image)

To download the tropical storm images, metadata, and labels, you should first crawl the nasa_tropical_storm_competition_train_labels collection and download the labels file located within the label item. Then, to download the associated image, loop through the links array and navigate to the source imagery items. Source imagery items will have the “rel” type of “source”. Once you navigate to a source imagery item you can find a link to the image (image) as well as a link to the metadata json (features) within the assets dictionary.

An example notebook which implements these steps can be found here.