What is ClearSky?
ClearSky is an AI algorithm that gives you invaluable unobscured Earth Observation data intelligence in near time. It’s like looking at the world as though there were no clouds at all.
JARGON → PLAIN ENGLISH
Synthetic Aperture Radar (SAR)
Satellites use SAR to bounce micro- and radio waves off the Earth’s surface and detect their echo. Technically, SAR sees through clouds, but the images produced are challenging to interpret.
Optical-Infrared Spectrum
The wavelengths of light spanning roughly 400-2300 nanometres, essential for many types of Earth Observation (EO) analysis.
When cloud, smoke or pollution obscures a satellite’s view of the ground, intelligence from Earth Observation images is lost. ClearSky solves this problem using a unique combination of deep learning and SAR to predict imaging across the full optical-infrared spectrum. Even when there’s 100% cloud cover.
ClearSky cuts through the cloud in this RGB composite image of Cornwall, UK.
How does ClearSky work?
ClearSky translates SAR data into cloud-free multispectral optical-infrared images. Each pixel in a ClearSky image is an AI prediction, mimicking traditional ‘direct view’ optical and infrared images. Because it’s delivered in a standard raster format, ClearSky results can be handled, analysed and interpreted in the same way as regular Earth Observation imagery.
Our current offering is based on ESA Sentinel imagery. ClearSky predicts Sentinel-2 imagery, giving you:
10-20m resolution
Global land coverage
Bottom-of-atmosphere reflectance values
10 spectral bands: Blue, Green, Red, Red Edge 1—4, Near-IR, SWIR 1, SWIR 2
ClearSky predicts Sentinel-2 equivalent imagery using Sentinel-1 C-band SAR. Depending on geographic location, Sentinel-1 acquires a new image every 3-12 days. For every Sentinel-1 SAR acquisition, we can produce a ClearSky prediction of what Sentinel-2 would see at optical-infrared wavelengths. The result is a reliable sequence of cloud-free optical and infrared imagery.
Exploded view with all 10 spectral bands.
ClearSky In Action
Tracking Normalised Difference Vegetation Index (NDVI) at 10m level
The Normalised Difference Vegetation Index (NDVI) is a well-established remote sensing index sensitive to the density of photosynthesising vegetation. It can easily be calculated from ClearSky imagery using the Red and Near-IR channels in the standard way: NDVI = (Near-IR – Red) / (Near-IR + Red). Since ClearSky provides cloud-free imagery, we can track the evolution of NDVI at the 10m level in single fields at high time resolution.
ClearSky prediction
Direct Sentinel-2 view
Mapping NDVI variations on a field-by-field basis, and their evolution over time.
ClearSky In Action
Flood mapping through the Normalised Difference Water Index (NDWI)
The Normalised Difference Water Index (NDWI) compares the reflectance in the short-wave infrared band to the green band to detect surface water. So NDWI can easily be calculated from ClearSky imagery to map and track flooding events.
Flooding in Northern England, November 2019
How accurate is ClearSky?
We validate ClearSky by comparing the algorithm’s prediction to actual cloud-free views on a pixel-by-pixel, band-by-band basis. This way we can quantify the accuracy over billions of pixels and recommend, if necessary, statistically robust confidence intervals for any ClearSky pixel right across the optical-infrared spectrum. The average accuracy per pixel is currently plus or minus 12% for any predicted value. The more ClearSky collects image data, the greater its performance. Over time, its accuracy keeps increasing and we anticipate accuracy to within plus or minus 5% in the near future. Aspia Space updates ClearSky every six months to ensure predictions are the best they can be.
Benefits
Regularity: Receive cloud-free optical-infrared multispectral imagery at the same cadence as Sentinel
Reliability: Get consistent, clean images with no intermittency. Introduce the time-domain to your image analysis
Flexibility: ClearSky produces information-rich multispectral imagery that can be used in myriad ways
Multi-scale: Analyse individual fields or entire continents, on scales of 10 metres to 1000s of kilometres
