Similarity Search is a tool that utilizes a machine learning technique called self-supervised learning to locate similar images/data from image archives. It is intended to augment and make the process of data search and collection for scientific studies more efficient.

One of the initial and most crucial steps for a scientific study related to climate change or events (e.g., wildfires, oil spills, hurricanes, dust storms) is for scientists to gather many relevant examples. Locating these examples requires painstakingly inspecting millions of square miles of satellite imagery each day across several years. While such an effort can produce a valuable trove of data, the act of manually searching is laborious, expensive, and often impractical–grounding many scientific studies before they could ever take off.

The similarity search augments the search process by allowing users to input a sample image and return similar looking images from the entire archive, minimizing the data download and visual comparison for similarity. This process also allows users to explore the spatio-temporal patterns of the features within the sample image.

Based on the image section the user selects on the base map and the number of results they seek, they receive similar image(s) from across the entire multiyear archive.

The tool returns a set of images similar to the input image based on the similarity threshold scale selected by the user. The smaller the range selected on the threshold scale, the more similar the image results are to the input image.

The histogram at the bottom of the page provides an overview of the distribution of the results across time. The heat map over the world map on the left side of the screen shows the coverage across space.