Approach

About the satellite data

PaddockTS uses surface reflectance data provided by the European Space Agency (Sentinel-2 satellites). The data is processed and made freely available by Geoscience Australia. Although we show here satellite data as imagery resembling what the human eye would see, Sentinel-2 reflectance data actually comprises reflectance intensity from distinct regions of the electromagnetic spectrum ranging from the visible (what we see) to the infrared. In this way, the satellite data contains more information than what the human eye receives by looking at something. 

Locating paddock boundaries

What defines a paddock? The answer is not always clear, and it can change with time. The AI approach we use for defining paddock boundaries searches for parcels of land that have distinct vegetation dynamics from the surroundings. This characteristic is especially typical for cropping paddocks, where activities like sowing, growing and harvesting tend to result in similar vegetation dynamics within a paddocks that is distinct from surrounding pastures, roads, forest patches and even other cropping paddocks with different regimes. 

It’s important to note that we do not define paddocks necessarily by fence lines that may or may not surround them, which has several consequences for how the approach works. First, when farmers change the area where they plant crops within a paddock over consecutive years, our approach can draw multiple paddocks within that area. Second, our paddock detection methods perform worse in grazing and pasture paddocks where the vegetation dynamics are less stark than in cropping paddocks. This is evident in our mixed-farming case study, where our automated paddock delineation performs much better on cropping than non-cropping paddocks.