By Tim Harris
“But how do you know that tree belongs to that species?”
This is a common question received by botanists around the world.
They might answer that they typically identify trees according to the shapes of their flowers, fruits and leaves. But just as we see and identify trees using reflected rays of sunlight captured by our eyes, satellites also detect sunlight reflecting from the tops of trees. Incredibly, these images are sometimes sufficient to identify individual trees.
Scientists have a long history of using the chemistry of trees for their identification. For instance, research has shown that the exact proportions of pigment chemicals within the leaves of a particular tree species determines the light that is reflected into the sensors of orbiting satellites1.
The ability to routinely monitor tree species diversity across the globe using satellites would be a huge step forward in assessing the state of terrestrial ecosystems. Scientists have shown that more than 20% of all plant species are at risk of extinction, largely due to the influence of humans. Real-time assessments of species diversity from satellites could revolutionise our response to this extinction crisis.
But not all satellites are the same. Satellites differ in their sensitivity to variations in the wavelength of reflected light. Species-level tree identification from space is largely done with the most sensitive equipment: satellites with multispectral sensors.
There are satellites with sufficient sensitivity to the different parts of the light spectrum to tell apart different species of tree. But there remains another problem to solve if we are to use satellite imagery to identify individual trees: establishing the boundary between one tree and its surroundings.
Just defining which plants are trees can be problematic. For the purposes of observing trees from satellites, it is the full extent of their branches, that is to say the size of their crown, which is important. While experts can assess this visually, based on the shape of each tree crown in an image, the aim of many scientists is for this procedure to be automated using a computer algorithm.
Modern satellites typically have a sensor with pixels comparable to those in a digital camera, but the area of the earth’s surface that is represented by one pixel in the satellite can vary considerably. One pixel in Landsat 8 for example, corresponds to an area of land surface up to 30m2. In contrast, one pixel in the QuickBird satellite, now decommissioned, corresponded to an area of just 2.4m2, and the current satellite WorldView has pixels that correspond to an area less than 2.08m2.
Only trees that cover an area greater than the area sensed by one pixel can be identified individually. These large individual adult plants therefore have an important role in the remote sensing of biodiversity.
When Dan Li and Yinghai Ke used imagery from satellites called WorldView2 and WorldView3 to automate the recording of street tree species diversity in Beijing, they found that the detail in the imagery was sufficient for an algorithm to define the boundaries of individual trees and then use the wavelengths of light within those boundaries to differentiate between eight different species of tree2. They also used satellite imagery from different times of the year, combined with knowledge how leaf chemistry varies with the changing seasons, to further enhance their accuracy in discriminating between species. However, trees that happened to be in the shadow of buildings when the satellite passed overhead could not be identified in this way.
Other scientists have used additional sources of data to solve the problem of finding individual trees in a satellite image. Data from LiDAR has proved particularly useful when combined with satellite imagery. LiDAR measurements (Light Detection And Ranging) from aircraft can record fine scale ground elevation (used in flood risk mapping) and even vegetation structure.
LiDAR works by measuring the time it takes for a laser beam to hit the ground surface and be reflected back to a receptor. This information is used to measure the altitude of the ground surface so accurately that it can be used to measure the height and width of individual trees.
LiDAR data can thus be used as an additional data source in an algorithm that calculates the boundaries of individual trees. Scientists can make measurements of tree height and structure using LiDAR data, which can also aid the process of discriminating different species of tree. For instance, one study combined data from the satellite QuickBird with airbourne LiDAR data to assess the relative abundance of different forest tree species near the Great Lakes in Ontario, Canada3.
Ground-truthing is also a key element when interpreting satellite data. One New Zealand study tracked down individual Pohutukowa trees (Metrosideros excelsa), to provide training data for an algorithm to ‘learn’ the characteristic light reflectance, size and structure of this culturally important species. The team were then able to successfully predict the occurrence of the tree elsewhere in New Zealand using QuickBird satellite data combined with airbourne LiDAR data4.
Of course, it is not only botanists who can identify trees; many of us know some of our local tree species and are able to identify them based on their general appearance, and citizen science is greatly increasing the data on plant species distributions.
Uniform, systematic data collection for species identification is now being called for in international agreements. Scientists monitoring the world’s climate already have clear standards for collecting climatic data using satellites, and now the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services is calling for standardised measurements of biodiversity, too.
Trees may well be a good place to start. Clearly there are challenges due to the sheer number of tree species across the globe, and currently hyperspectral imagery is only available for a fraction of the earth’s surface, but collaborations between scientists studying leaf chemistry and scientists analysing satellite imagery and LiDAR data might just allow the automated monitoring of tree species diversity… even the trees at the end of your street!
1. Ustin, S.L. (2013) Remote sensing of canopy chemistry. Proc. Nat. Acad. Sci. USA 110 804-805.
2. Li, D., Ke, Y., Gong, H., Li, X. (2015) Object-Based Urban Tree Species Classification Using Bi-Temporal WorldView-2 and WorldView-3 Images. Remote Sensing 7 16917-16937.
3. Van Ewijk, K.Y., Randin, C.F., Treitz, P.M., Scott, N.A. (2014) Predicting fine-scale tree species abundance patterns using biotic variables derived from LiDAR and high spatial resolution imagery. Remote Sensing of Environment 150 120-131.
4. Pham, L.T.H., Brabyn, L., Ashraf, S. (2016) Combining QuickBird, LiDAR, and GIS topography indices to identify a single native tree species in a complex landscape using an object-based classification approach. International Journal of Applied Earth Observation and Geoinformation 50 187-197.