We would like to cultivate a culture of discussion

...with our agriconBLOG

agriconBLOG

22.07.2019

Precision Farming with N-Sensor or Satellite?

Since the availability of the Sentinel-2 satellite data, propositions for the use of this data have mushroomed. From the start-up to the international company, everyone is currently fighting to win the bid of the farmers. In various marketing statements, it is suggested to potential customers that they can use the inexpensive (mostly free) tools to increase yields and save fertilizer and pesticides. On top of this, these are very easy to. This raises the question: is there still a purpose of sensors mounted on farm machines? It's a fact that precision farming needs the most accurate sources of information possible as a starting point for the automation of production processes. Therefore, the suitability of an source of information should always be checked according to specific criteria. These include the following questions:

measured value,
quality of measurement and accuracy,
repeatability and reliability,
the practicability and ergonomics of the procedure as well as
the cost-benefit analysis

Whether sensor or satellite - every data source has to be assesed by the points mentioned above. We would like to show it by the example of site-specific nitrogen fertilization and will focus on the differences between the two information systems. Since the creation of application maps by using satellite data has always the same workflow, we'll go by the examples of "atfarm" by Yara and "CropView" by 365 FarmNet. Using the same field, we can illustrate how different the systems work.

The correct measurement

	N-Sensor^®	Satellite Pictures
Measureing Procedure	reflection of spectral bands	reflection of spectral bands
Derived Measured Value	Absolute indices with calibration to die absolute N-uptage in kg N/ha	Relative indices without calibration auf die absolute N-Aufnahme the absolute N-uptake ("more or less")

Both methods measure spectral information of plant stands. The N-Sensor® uses two defined wavelengths. These can then be used to calculate the exact and absolute N-uptake of a plant stand. However, the data of the satellite image providers only represent a "more" or a "less" chart. A specific reference to the current N-uptake of a plant stand is not possible.

Both methods measure spectral information of plant stands. The N-Sensor® uses two defined wavelengths. These can then be used to calculate the exact and absolute N-uptake of a plant stand.However, the data of the satellite image providers only represent a "more" or a "less" chart. A specific reference to the current N-uptake of a plant stand is not possible.

Practical Example: CropView (365 FarmNet)

[Translate to Englisch:] Klicken für Großansicht

When displaying the data, there are no numbers that define the "vegetation". The users themselves are obliged to assess the "more" and "less" and to adjust the N fertilization. The index used to determine the vegetation differences is, according to current knowledge, the dimensionless Normalized Difference Vegetation Index (NDVI). Textures are barely visible in the map. The problem is that the used NDVI at 40 kg N-uptake/ ha (cereals to EC 30/31, oilseed rape in EC 16-18) runs into a so-called saturation. This means that all measurements above this N-uptake are technically not clearly differentiable. The measured differences are therefore purely random and are visualized only by "magnifying" the scale.

Practical Example: atfarm (Yara Digital)

Also, the indices offered by the satellite service do not present any absolute numbers. The naming (as shown in the example "N-sensor display") is cleverly chosen by the marketing division since it implies a connection to the Yara sensor technology. However, compared to the sensor technology, the satellite images can not display how high the average N-uptake is, thus no direct calculation of the optimal N fertilization is possible.

If you got a N-uptake of 160 kg/ ha in oil-seed rape you don't need to fertilize, but if you got 100 kg/ha you'd need to add about 60 kg. It's a crucial information which is not provided to the users. Because with an N intake of 160 kg / ha in rape would not be fertilized, at 100 kg / ha, however, would be about 60 kg needed. The user does not receive this crucial information. It is also not known if the color gradations are 5, 10 or perhaps 20 kg N / ha. Textures or partial surfaces are barely visible in the map.

Practical Example: agriPORT (Agricon)

The data of the N-Sensor® always show the N-uptake of the standing crop at the time of the crossing in real time. So an absolute and objective assessment of the plant stand is possible. Compared to the satellite imagery, the machine-bound N-Sensor® receive a significantly higher resolution, showing structures and plots much better.

Measurement quality, accuracy and reliability

The N-Sensor® as a machine-bound system measures the reflection of the plant stand from a distance of approx. 4 to 7 m. Apart from hoarfrost in February / March or possible dust development under dry conditions (where N-fertilization makes little sense anyway), there are no disturbances which can falsify the sensor signal.

Now satellite images are taken from a distance of over 750,000 m! In between is the entire earth atmosphere. All particles contained in it (dust, water vapor, aerosols, etc.) are thus in constantly changing amounts in between the measuring device and the plants.

	N-Sensor^®	Satellite pictures
independent of weather conditions	Yes	No
Availability	Always	Only during cloudless sky while a fly over and with no shadows casted over the fields
Timing of measuring	Real time	Days, weeks, sometimes months (depending on the cloudage)

In addition to that clouds do cast shadows over earth' surface influencing the satellite imagery by significantly disrupting the interpretation of the images, or make the use of the images completely impossible.

Clouds and casted shadows

Although the cloud itself covers parts of the fields, even the non-affected plots and areas are not usable, because of the casted shadows. Thus, approx. 40% to 50% of the total section can not be used to calculate a N-application map.

Practical Example: Data availability

It's a common occurrence that there are no useful Sentinal-2 data available during the growing period. In our example you can see that due to a lot of clouds in between April 27 and May 14 2019, none of the satellite images could be used. Thus customized N-fertilization was therefore only possible in a very limited range or according to quite old data.

Practical Example: Incorrect N-application map

Despite cloudiness, users have the opportunity to use images for fertilization. If, for example, the picture of 2 May 2019 had been used to calculate an application map, the farmer would have had to expect significant flaws in the calculation of the N fertilizer quantity. Especially in the areas where clouds are over the field, there is clearly overfertilization due to the measurement index being too low.

Measuring quality and reliability

Die The quality of measurement and reliability of satellite imagery is therefore corrupted, compared to that of machine-mounted sensors, as it is exposed to significantly more disturbances.

Spatial Resolution	N-Sensor^®	Satellite Data
Measurement Values/ha	125	25
Position Accuracy	+/- 0,1 ... 0,3 m	+/- 11 m (even more in hilly terrain)

The measurements of the N-Sensor® have a five times higher spatial resolution than a satellite image. Satellite images have a average positioning inaccuracy of +/- 11 m. Each "image tile" is 20 by 20 meters and contains just one piece of information. The position of the "image tile" can defer in all directions by 11 m to the reality. Thus, small-scale differences can hardly be detected.

To the northwest side biomass is shown at the plot border, which is not present in reality. In fact, there is a dirt road located, therefore, no significant vegetation. The indicated biomass is from the neighboring plot that lies behind the dirt road. Thus, the entire map will probably have a displacement of 10 (?) meters.

The problem of positional inaccuracy and margin effect, regardless of the positional error in the field, is more noticeable in smaller fields than in large fields.

Practicability and Ergonomics

Due to the lack of absolute measurement values, no absolute agronomic control functions can be used, which is a big problem because the most important decisions are reached during this procedure. Here are the biggest and most important decisions that the user has to make:

Decision 1: According to which principle do I want to fertilize?

Robin Hood function (take from the rich and give to the poor) or
King John function (take from the poor and give the rich)

Decision 2: What is the incline of the control function?

How strongly do I want to react to differences in the map with my N-fertilization?

Almost all providers leave it to the farmers to answer this question as they please, so they call it. In reality it means farmers are left on their own due to the lack of recommendations. In addition to that there is no way for them to check if their decision was a good or a bad one.

The farmer has to specify the absolute fertilization level by himself. As a rule, he relies on his experience and "the eye of the Lord". In order to increase the accuracy, one could make several biomass cuts in rapeseed or several measurements with the N-tester in the crop. Apart from the time consuming effort, the data on the computer would have to be assigned to the corresponding tile in the satellite image. At the moment, that's hard to imagine.

Therefore, there is always the danger that

the choice of control function,
the absolute fertilization level as well as
the variation of the N amount

is fundamentally flawed.

Ergonomics

The manager or agronomist must do a relatively large amount of office work during the peak season. When using satellite imagery, the following steps must be followed for each field:

Control: are current pictures available?
If not: which images do I choose alternatively?
Where is the fertilization level?
Definition of min and max kg N / ha
Download the scam card
Manual export of each card to a USB stick
Transfer to tractor

Let's take a look at the example of a 1,000 ha farm with 70% cereal and canola in cultivation: with a average plot size of 15 ha, this is about 50 plots. So during winter the user has to count 50 application maps for the first

nitrogen dose, in March / April another 50 for the second nitrogen dose and in May / June maybe another 40 to 70 application maps for the third and fourth dose in the crop. This way, around 140 to 170 application maps come together,

which must be made in the above named steps. And all of that during the already busy spring season.

Using our N-Sensor® N fertilization process, office work is essentially limited to two things:

Create first order nitrogen application maps based on N-Sensor® autumn scans. All plots with one crop can be calculated simultaneously and almost automatically.
Order management: Send data via e-mail to the machine, where the driver can retrieve and process it immediately.

These orders can optionally be made for all doses and all fields in the run-up to the spring season and thus represent no additional burden for the farmer.

Cost, benefits and proven effects

Atfarm currently requires 8 € / ha active area. With already designated 1,000 ha of farm with 70% crop and rapeseed in cultivation, the farm pays € 5,600 a year. In ten years, that's about 56,000 €.A ready-to-use system, consisting of N-Sensor® ALS 2 and terminal, costs € 27,500 when assembled. It is amortized over five years. Every two years you should have a hardware check for 500 €. At 3% interest and two to three checks within the depreciation time, annual costs of approximately € 5,700 are incurred. In ten years this will add up to a cumulative € 32,000. But the cost side is only secondary.

The really crucial question is, what do you get for it? For the N-Sensor® the advantages are proven many times. These include N-savings of up to 15%, yield increases of 5%, improvement of the harvest and the quality of the harvested product and avoidance of N-caused storage. This results in benefits of an average of 100 € / ha and in crop protection 45 € / ha each for the application of growth regulators or fungicides.

As a result, investments in sensor technology are highly economical and a return of investment usually already exists after one year. According to current knowledge, there is still no large-scale experiments by a provider of satellite images. Thus, no evidence has yet been found that a positive effect of N fertilization after satellite images occurs.

Conclusions

The use of satellite images for N fertilization is probably better than a constant application. However, the value for money is not correct. The most important disadvantages, represented by the example of atfarm compared to the N-Sensor®, are:

Limited availability due to cloud cover and casted shadows
Location inaccuracy of the tiles
Up to 33% lower resolution
No calibration to the absolute N-uptake
Absolute control functions are not available or applicable
Poorer ergonomics mean more time spent during the season
For larger companies, higher costs than an N-Sensor®

Over the past few years, Agricon has repeatedly checked satellite imagery as a possible source of information for precision farming. However, due to the known vulnerabilities, we have deliberately decided against it. It was not possible for us to derive positive earnings effects analogous to the N-Sensor®. What use is a seemingly "cheap" digital tool if we do not get a positive effect? Then it is better not to offer a satellite solution than a worse solution compared to the sensor technology.