Data Resolution
Introduction to spatial, temporal, and spectral resolution
Remote sensing datasets have three main types of resolution associated with them: spatial, temporal, and spectral.
Spatial resolution refers to the size of the ground element represented by a single pixel. Generally, this is expressed as the size of a single pixel along one side, e.g. the spatial resolution of the Sentinel-2 red band is 10 meters. Most imagery in the Landsat archive has a spatial resolution of 30 meters. At this resolution, one pixel is the size of a baseball infield, however, to reliably detect an object in an image it’s often recommended that the object should occupy at least 3-9 pixels, depending on the sensor. Spatial resolution has major implications on the size of the data stored on disk, legality of use, ease of access and cost, and most importantly, what you can distinctly identify from the imagery. If a user needs to count cars, Landsat will not meet their spatial resolution needs, and they may need to save room in their project’s budget for high resolution data.
Spectral resolution refers to the sensitivity of the sensor to resolve fine ranges in the EM spectrum. A sensor with high spectral resolution will contain information about very narrow spectral ranges (e.g. each band may cover a 10nm range), while a sensor with a coarse spectral resolution could contain a bands with information about a wide spectral region (e.g. the entire visual range, about 400nm). Differences in spectral range and resolution between satellites can have a meaningful impact. It is particularly important for applications where the feature of interest has a narrow absorption feature, such a mineral mapping, meaning that signal of the feature is only distinct in a narrow range of EM radiation.
Temporal resolution refers to the frequency of image updates for any given location on the Earth. This is also called the satellite “revisit rate”. Satellites that are collecting in sun-synchronous orbit will have a regular revisit rate over a given location on Earth (e.g. Sentinel-2 will revisit a location either every 5-10 days depending on the location), while satellites that must be tasked or aerial datasets will have more irregular collection patterns. Temporal resolution is important for monitoring applications, as well as for applications where the temporal pattern of the signal is informative. For example, high temporal resolution is often important for agricultural applications, where the satellite can be used to track the growth phases of a crop over the growing season.
Generally speaking, users have to compromise on some element of resolution. Technical limitations prevent a single sensor from capturing data at high resolution in all three areas. Increasingly, with access and the ability to take advantage of more diverse and larger datasets, a user may achieve high resolution across all three types by combining data from different sensors.
The choice of which resolutions are the most important is dependent on the user’s application.
-
Computer vision (CV): A user is interested in counting the number of swimming pools in southeastern Florida. This will require images that can resolve relatively small objects to allow the CV model to identify the pools (spatial resolution), but likely will not require information from bands beyond the visible range (spectral resolution), or require multiple images of the study area (temporal resolution). Overall, the required spatial resolution will be high, while the spectral and temporal resolution requirements will be low. An ideal sensor here may be Airbus SPOT or Pléiades imagery.
-
Vegetation Monitoring: A land manager is interested in monitoring vegetation growth over the course of the year. In order to separate healthy vegetation from areas with little to no growth, the sensor needs to measure bands covering various parts of the EM spectrum that are sensitive to vegetative growth, from the red to near/mid infrared regions. As the user is interested in getting regular updates throughout the year, the revisit rate (temporal resolution) should be relatively high. However, since the user is interested in broader spatial trends, very high spatial resolution is less important. In this case, Sentinel-2 would work well.
-
Wildfire Detection: A user is interested in detecting fires across North America. In this case, there is a need to get very frequent updates (temporal resolution). The spectral range needs to include bands that are sensitive to high temperatures (midwave-IR, thermal-IR, and shortwave-IR). Given the large area of interest, the user is satisfied using data without very high spatial detail. For this application the GOES sensor, which images the western hemisphere every 15 minutes, may be a good fit.