It’s well understood that controlling fallow weeds to preserve moisture and nutrition for the following crop is essential for success. What is less known is that the 10m resolution Sentinel 2 imagery is quite capable at detecting the presence of weeds in fallows. You’re not going to detect individual plants, but rather as clusters develop the overall reflectance in the pixel will change, allowing us to map it.
In this example, I have applied the MCARI2 spectral index instead of the normal NDVI. As I’ve mentioned in the past, I find it does a better job and not saturating and less affected by soil color (which will help here). The data has an interpolation applied to smooth out the blocky pixels to make it easier on the eye. The background is a Google Maps high resolution true color satellite image.
This paddock was 2022 wheat harvested in November. After harvest, some residuals + paraquat were applied followed by a Weedit camera spray to get us on top of the fleabane and other scattered weeds.
Following some rain in early January there was a germination of volunteer wheat in the header trails (plus other weeds). You can see here some interesting patterns develop across the paddock. Some areas the header trail had more wheat germinate than in others.
The next image (2023-01-29) shows other parts of the paddock start to green up – particularly the heavier soils. The volunteer wheat took longer to ‘get up and go’ here. Furthermore, in many places where it had got away early have now a decreased MCARI2 value as they have been affected trying to grow in summer and pushed roots past the header trail into the soil and therefore affected by the flame residual.
To put the wheat out of its misery we sprayed glyphosate on 27th January. Following the spray, this image (2023-02-03) shows almost all the green taken out of the map except for a few places. This remaining green was of course glyphosate resistant populations of barnyard and feathertop rhodes grass. These grasses were present in a few other places as individual plants as well – not significant enough in size or population to show on the imagery.
The Weedit did another round to tidy up these weeds with that latest image showing a tidy paddock (2023-02-11).
Attempting to detect weeds in fallows is a challenging use case for this imagery. Therefore the color scale is set to be quite sensitive which can lead to some noise, or other issues such as brighter soils or residues seen as weeds.
The way I suggest using imagery for weed monitoring is more as an insurance or an overview rather than a way to task sprays or inspections. Things will definitely start getting out of control if you were to wait until you can see something on a satellite image before you go out and see what it is on the ground. But if you can start to see green areas on imagery – this would indicate higher priority for sure.