The problem of “automated” agronomy and in practice Satellite vs. plant sensor

While browsing the LinkedIn platform, I found this interesting picture of Nathan Faleide, an AgTech specialist from the USA, see

Figure 1: From individual data to automated decision-making.

This illustrates my current experience in the creation of management zone maps (“potential maps”) from multi-year satellite data very well. This is a task that is carried out automatically by various providers. But what is the quality of the maps generated? Every year is different, fields are split and merged, irrigation lines are laid, there are processing and application errors, cloud shadows and much more. In my experience, there are areas where only ONE date has provided a usable image for zoning since the start of the Sentinel-2 flights! For such areas, an ” automated” system provides the mixed entity at the bottom left of the image, and in contrast to this, you can’t even guess which data is responsible for this in the generated map!

An AI also relies on data and learnt knowledge; if these are incorrect, the result will be the same. In computer science, programming an access function for a database is labour-intensive for humans, but a trifle for an AI. Assessing the quality of basic data and arriving at a comprehensible result with the ” correct” agronomic knowledge is much more complex. And then there’s the “battle” over which is the ” correct” agronomic knowledge 😉 .

In practice Satellite vs. plant sensor

The creation of fertiliser maps for our service “N fertilisation according to the measurement of N uptake by satellite in autumn” for winter oilseed rape is in the final phase.

Many of our customers have a YARA N sensor, but do not perform an autumn scan. The reasons for this are the limited operating times of “old” passive sensors in autumn, peak workloads, cost savings for an extra trip or the poor accessibility of the fields. This means, for example, that only a few fields could be scanned, but planning is required for the entire rapeseed area.

An opportunity for us to compare the satellite and plant sensor in practice when recording the parameter N uptake in oilseed rape!

Figure 2: Left: N uptake map derived from satellite images according to EXAgT algorithms, right: interpolated N uptake map from ALS-2 measurements from the Agricon Agriport.

The comparison shows, on the one hand, a comparable level of the measured N uptake values, but also the greater level of detail of the map from satellite data.

The differences lie in the type of measurement. The Sentinel-2 satellite works similarly to a digital camera with a defined resolution such as the 10x10m spectral channels used here.

With the YARA sensor, an average value is calculated from the measurements of each of the two sensor heads (2x 3m measuring width per head). For each track, this one value is used as the default value for the entire working width (36 metres in this case).

Modern centrifugal spreaders have the ability to spread different quantities at least to the left and right, while pneumatic spreaders and sprayers have many individually controllable sections. This means that a higher resolution of the application maps can also be realised within the track.

We create application maps from these values using the tried and tested Rapool algorithm; agricon has its own calculation methods.

Ask us, we look forward to your tasks! We specialise in company-specific solutions, we value you and your challenges =;-).

Our contact details are:
+49 (0) 176 72588814, +49 (0) 34324 269737
+49 (0) 173 352 8960, +49 (0) 34324 269739

This entry was posted in Agronomy, Technology. Bookmark the permalink.