Satellite platforms

Since Landsat 1, the first Earth-observing satellite, was launched into orbit in 1972, a wide range of other satellite Earth observation data is currently available. The satellite image data spans free-to-use imagery to commercial very high resolution (VHR) imagery.

Satellites provide radiometric and structural information consistent with the colour and geometry of the plant canopy they observe from orbit altitudes typically ranging between 450-800 km. The suitability and performance of a satellite remote sensing system for weed detection, to a considerable extent, is a function of the system’s resolution.

The resolution tells the user the extent of detail about the weed the satellite can provide, and this information is crucially important for the analyst in making a decision around the satellite imagery to use for the detection of LGW. The satellite image data is pre-evaluated based upon the different types of the system’s resolution- spatial resolution, radiometric resolution, spectral resolution and temporal resolution.

Fig. 4.  Sentinel-2 satellite in orbit (Source: Copernicus ONDA DIAS, accessed on 6 September 2023).

Spatial resolution

The satellite imagery is composed of pixels with each pixel representing the smallest area of the weed that the sensor aboard the satellite can detect from the given orbital altitude. The size of the pixel of the digital image describes the spatial resolution of the satellite.

The spatial resolution of existing public satellites range between 6 m to 1000 m, while the pixel size of commercial satellites range between 0.3 m to 3 m. Satellites imaging the target at a lower orbit height (e.g., 450 km) capture more spatial detail about the target while satellites farther away in space from the target produce low spatial resolution imagery (e.g., 500 m) but offer larger spatial coverage and higher revisit time.

Radiometric resolution

A satellite system records the radiance or brightness level of the target. Radiometric resolution describes the discernibility of the system to subtle changes in the brightness levels of the target. The radiometric resolution specifies the number of shades of grayscale and is expressed in bits.

Other terms used to describe the radiometric resolution include bit depth, quantization level, and dynamic range. The radiometric resolution of the early SRS systems were typically 8-bit (i.e., 2⁸), making a pixel’s potential brightness value range from 0 and 255.

Every pixel of a digital image is associated with a brightness value in that brighter pixels have high values while darker pixels show low radiant energy. Current satellites, such as Landsat 8, Sentinel-2 have 12-bit while Planet Labs’ SkySat produces 16-bit image data. The larger the dynamic range the lower uncertainty around the variability in the radiance detected by the system, hence, improving the discrimination of weeds in heterogeneous landscapes.

Spectral resolution

Satellite systems record the reflected or emitted light that comes off targets, this light is displayed on our computers as colours. The return light measured by the sensor does not only include visible blue, visible green, and visible red, but can also include infrared light.

Spectral resolution denotes the number of light the sensor is sensitive to, which could be a single broad-band monochrome image (also known as panchromatic) or narrow-band multicolour image. A light is described by a range of discrete wavelengths referred to as a spectral band, and a band can be similar or different between SRS. Although SRS can provide imagery that contains both panchromatic band and multispectral bands, existing optical satellites favour collecting multispectral visible and infrared image data.

A multispectral system can offer between 3 and 24 spectral information about the weed while a hyperspectral system offers more than 24 spectral bands, usually between 100 and 1000 bands. Compared to the multispectral, the hyperspectral systems have high spectral resolution as the wavebands are narrower and it is more useful for discriminating between weed species that have very similar physical and phenological attributes to native plants and/or other invasive plants (e.g., serrated tussock vs African lovegrass).

Currently, there are very limited existing satellite hyperspectral systems, but this is expected to change in the future. Satellite hyperspectral data are expensive to acquire, store, share and process.

Temporal resolution

Temporal resolution is the length of time a satellite takes to revisit and repeat measurement of an area of interest. The temporal resolution differs between satellites as is a function of the satellite’s orbit, swath width, and tilt angle and the location of the target. Landsat takes nominally 16 days while Sentinel-2 can take 5 days to revisit a field.

Commercial satellites, including WorldView-3, can provide daily images of the invasive plant. The temporal resolution of the satellite is more crucial if the physical characteristics of the weed under investigation changes rapidly.

Satellites that have large swath width would have high temporal resolution and vice versa. The size of the swath width is dependent upon the orbital height of the satellite as low altitude satellites correlates with low swath width and high orbital altitude means large swath coverage. However, a high temporal resolution data is obtained at the expense of high spatial resolution.

Earth observation satellites that offer very high resolution imagery, sub-meter pixel size, are more useful for the detection of invasive plant species with large infestation coverage (Robinson et al., 2016). Table 1 shows the satellite remote sensing systems useful for the mapping of environmental and agricultural weeds.

Table 1: The spatial, radiometric, spectral and temporal resolutions of satellite platforms relevant for operational detection and management of invasive plants of large growth and infestations.

PAN = panchromatic, NIR = near infrared, SWIR = short wave infrared