Satellite images are available in different spatial resolutions - from a pixel size of less than one m2 to more than one km2. They are also consisted of different spectral bands - apart from basic red, green and blue bands (this is like images we take with a camera or a mobile phone), satellites also acquire additional bands that capture infrared and other lower electromagnetic wavelengths that we cannot perceive with our eyesight. Another thing is the frequency of satellite images - some images are available daily, others every couple of days or even more sporadically.
These three characteristics (spatial resolution, the number of bands and temporal frequency) determine to a certain extent our approach for non-native vegetation detection. If the temporal frequency of the images is high, we can form a time series (on time series read this article) and observe spectral (phenological) signatures of the plant species - we can imagine it as different curves for each band through time. Based on these spectral curves and advanced image processing algorithms we can perform detection of different non-native species in our area. A high number of bands allows us to study each spectral signature of an invasive species in detail so that we can detect slight differences in the shades of green (leaves of species), invisible to the naked eye.
When specialists collect the exact locations of invasive plants in the field, they have to consider a few things when collecting terrain data for remote sensing. First, the population or stands of invasive species must be visible from the air - if the invasive plants grow under a tree or other object, we cannot see them from above because of the satellite view. Second, it is impossible to detect a stand that is too small compared to the spatial resolution of satellite images. If the spatial resolution is 10 m, the stand must be at least 100 m2 in size to be detectable from the satellite images. The third important thing is the collection of "clean", representable terrain data, which means that we have to register sites where only one species is present. If different species of non-native plants are mixed together, it becomes very difficult, if not impossible, to record the individual invasive plants.
Potential location of Japanese/ Bohemian knotweed and Cherry plum species obtained by time series analysis using Sentinel- 2 satellite images (EU Copernicus).
The detection of Cherry plum looks less reliable because we use a much smaller amount of field data for training an algorithm than in the case of knotweed.
Ana Smerdu & Urša Kanjir Slovenian Centre of Excellence for Space Sciences and Technologies Space-SI