Landsat Collections - RMSE
Landsat Collections: Providing a Stable Environment Record for Time Series Analysis
This is the second of a three-part video series describing the new U.S. Geological Survey (USGS) Landsat Collection 1 inventory structure. Collection 1 required the reprocessing of all archived Landsat data to achieve radiometric and geometric consistency of Level-1 products through time and across all Landsat sensors. This video specifically explains the radial root mean square error (RMSE), and how it is used to define the spatial accuracy of a Landsat scene, and whether it is organized into Tier 1 or Tier 2 of the Landsat Collection 1 inventory structure. To learn more about Landsat Collections please visit https://landsat.usgs.gov/landsat-collections.
The Landsat Program is a series of Earth-observing satellites co-managed by USGS and NASA, and offers the longest continuous space-based record of Earth’s land in existence. Every day, Landsat satellites orbit Earth and provide essential information to help land managers and policymakers make informed decisions about our natural resources and environment. All Landsat data are distributed by the USGS at no charge from EarthExplorer, GloVis NEXT and the LandsatLook Viewer. To learn more about the Landsat Program please visit https://landsat.usgs.gov/ or follow us on Twitter @USGSLandsat or Facebook @NASA.Landsat.
Image Dimensions: 1920 x 1080
Location Taken: Sioux Falls, SD, US
Content created by Andrew Dykstra and Linda Owen (Contractors to USGS EROS)
[Gene Fosnight] RMSE is the Root Mean Square Error, and it’s used to define how spatially accurate an image is.
[Narration] Landsat Collections data is processed to minimize this error, in most cases to less than a third of a pixel.
[Brian Sauer] When we process a satellite image, the first thing we do is we generate a satellite model. Landsat 8 has GPS built in it, so we know we can get the accuracy of that model very tight. For the older sensors, we didn’t have that luxury. So when you do a systematic model of where the satellite is pointing on the ground the scene can be shifted some. So what we’ve done is built the Ground Control Library. We have this stack of GCPs that we correlate each image to.
[Gene Fosnight] Ground Control Points are chips which are pulled from a reference grid. Each of these chips are small images in their own rights which we have precisely located to the earth’s surface. We take the images, we compare them against those hundreds of Ground Control Points for each image, and we shift the datasets so they align with the Ground Control Points.
[Narration] After the image is shifted, an RMSE can be determined. 90% of Collections scenes are within a 12 meters RMSE goal.
[Brian Sauer] The Pixel that we’re working with is a 30 meter pixel. So when we say RMSE, we’re comparing that image to those points on the ground to ensure that the image actually aligns to that 12 meter specification of each pixel. That way when you’re doing land change monitoring and you’re stacking scenes over time, you know that each pixel isn’t going to shift more than 12 meters either way.
[Narration] Each Landsat scene’s metadata file contains its RMSE. Users can also choose an RMSE range in EarthExplorer to narrow their search and get only the scenes that fit their needs.