Not so picture perfect: common problems with satellite imagery

  • The delivery of less perfect than expected imagery is a common problem for many clients. Imagery providers are affected by a multitude of naturally occurring phenomena that have a detrimental impact on the quality of collected images. 

    In this short blog, we address some of the barriers to the provision of “perfect” imagery; and the processes TCarta employ to mitigate these issues. 

    QA/QC processes:

    TCarta employ a variety of rigourous internal Quality Assurance / Quality Checking processes prior to imagery collection and delivery. This enables us to deliver the best possible quality imagery to our end clients: 

    Check coverage of customer area of interest;
    Check general accuracy by comparing to existing data / freely available imagery;
    Check resolution and number of multispectral bands is as ordered;
    Check for cloud cover / other artefacts if a customer has a specific demand.

    Common phenomena affecting satellite image quality 

    Cloud Cover

    When collecting new imagery, it is possible to state particular parameters, such as “less than 5% or 10% cloud cover” of the Area of Interest. These requests do come with uplifts and may prolong the collection time. Certain geographical regions will be affected by cloud cover, such as equatorial regions. Cloud cover can be penetrated by some types of imagery, such as radar, but these may not provide the resolution or information that the client requires.

    Nadir angle

    The nadir angle of the satellite has a direct effect on the resulting resolution of the image due to the “lean angle” of the satellite which can mean that the distance from the camera to the ground is extended, resulting in poorer resolution. This can be mitigated by specifying a particular angle that will not impact on the resolution. Uplifts to pricing will apply, and this will also impact the revisit time of the satellite.

    Dust in the air

    In certain geographical (largely desert regions), dust can impact upon the visibility of ground based features. This cannot always be avoided if new collection of imagery is required on a specific date. Therefore it is sometimes recommended to be flexible with collection dates so that the effects of dust can be mitigated against.

    Revisit time

    Revisit time refers to an interval between multiple collections of the same area. This can be impacted by weather, competition for the satellite from other demands, technical restrictions, etc. In general, most satellites have a revisit interval of one day or more, but customers should be aware that if a small interval is required then nadir angles may vary (meaning inconsistent resolutions between collections), and weather conditions may also vary.

    Time of day

    Most passive (optical) satellites pass over the equator at the same time every day due to a regular orbit. This means that it is not possible to acquire an image of a specific time over an area of interest. Some satellites, such as Landsat 8 follow a designed acquisition plan to collect images over the same specific location every 16 days. 

    In addition, optical satellites can only collect images during day time since they rely on the light from the sun, like a traditional camera. Radar satellites can collect imagery during night conditions.

    Accuracies 

    The accuracy of a satellite image is generally dictated by both the GPS (Global Positioning Satellite) and resolution of the image. It can be improved by ortho-rectifying the image and removing terrain distortions, together with using surveyed Ground Control Points to pinpoint the exact location on the ground. The best possible horizontal accuracy from the satellite is approximately 3 metres CE90 (Circular Error with 90% confidence). TCarta can discuss various options to ensure the best possible accuracy achievable.

    Resolution 

    The best possible resolution that can be acquired from a satellite is 30 centimetres. Meaning that the pixel size on the ground is the size of a ruler. However, it is normally advised that features are identifiable with 2 by 2 pixels, so this would mean that an object the size of a desk would generally be identifiable. It is not possible to zoom in and read vehicle registrations from commercially available imagery as normally referenced in the movies!