We tested and compared four different methods for collecting spatial information about onion crops and their development, looking to identify variability.
The tools used were:
- GreenSeeker NDVI (supported by AgriOptics) (Time series here)
- CoverMap canopy cover (supported by ASL) (Time series here)
- UAV imagery (supported by Altus UAS)
- Satellite imagery (obtained 23 November 2016)
Captured data need processing to create viewable maps. When individual point measurements are captured we estimate what is happening between them. The wider apart the points, the bigger the guess. When images are captured, the size of each pixel determines the detail we can see.
We compared the results from each method at the end of November after the satellite image capture. An image section is shown below, with a white grid to help locate same points in each of the images.
We mapped the paddock many times using both CoverMap and GreenSeeker. CoverMap captures more points than GreenSeeker so the map appears more detailed. GreenSeeker also captures many points, but the software averages a number of readings then gives that result at wider spaced points.
We found the 2015-16 version of CoverMap app was very sensitive to brightness. Initially we were mapping sun and shade variation depending on which way the tractor faced and if cloud cover changed. By wrapping the system so only diffuse light reached the target area we did obtain much more stable images.
Part of the issue was a change from one version of iPhone to the next! (ASL worked on image stabilisation under varying light conditions and the 2016 version is good.)
The GreenSeeker software limited the number of points we could record. The image below is a canopy map created from the GreenSeeker data. It shows variation in NDVI, the normalised difference vegetation index.
The map above is a corresponding map created from CoverMap data collected at the same time as the GreenSeeker NDVI. The map looks more “speckly”. This is because it was made from many more data points so identifies variation in more detail.
We think the GreenSeeker and CoverMap maps do show a similar story. We attempted a statistical comparison by “overlaying” the maps and comparing points. The results are shown in the graph below.
We see some agreement shown by this analysis. How much is expected?