Stenon FarmLab - with software version d-1.3.0 and calibration model p-2.1.0

Table 1: Recognised parameters – results at a glance

*  Assessment explained: Meets the requirement  (✓)/ Does not meet the requirement (X)

How the measurements are taken

As a first step, the user calibrates the device using the calibration cap provided. The probe is then inserted into the soil by stepping one’s foot on it. Measuring is then triggered by tapping on the touchscreen of the control unit. After each measurement it is necessary to clean the probe. 
One measuring cycle consisting of 3 individual measurements comprises the results of all parameters listed above. Then the data are transmitted to the cloud application from where the results are returned within seconds. The measurements can also be taken while offline; in that case, the data are saved to the internal memory and synchronised automatically when an internet connection is established.

Scope of measurements

The measuring range depends on the type of soil and the measurements apply to a depth of 0 - 30 cm. The Stenon FarmLab is suitable for use in sandy, silty and loamy soils. Table 1 shows the measuring ranges for all parameters tested.

Table 2: Stenon FarmLab measuring ranges (manufacturer information)

Prediction accuracy

The testing method

The predictive accuracy on each of the three soil parameters is verified by comparing the data supplied by the sensors with data that are obtained by analysing appropriate soil samples in various labs. Then the results of these comparisons are assessed by applying the current DLG assessment scheme.
For each test parameter and each measuring range, the manufacturer must provide data on the measuring precision of the sensor. They must also state which method was applied to measure these precision levels.

The scope of the tests

The tests are carried out in 40 different fields. The fields chosen represent the widest possible range of soil and crop situations in which the product is potentially used:

• A variety of sandy / silty / loamy texture
• High / medium / low Nmin levels 
• High / medium / low humus levels
• Typical crops (e.g. asparagus, strawberries, lettuce)
• Fields under cropping to ensure the relevance of the results

The test design

An approx. 2 m by 2 m plot is marked out in each of the 40 fields. In each of these 4 m2 test plots, measurements are taken in five different sub plots following a specific design. Each measurement is taken with two soil sensors. All sensor measurements per test parameter are taken for a 0 - 30 cm depth and are then recorded and logged.

As a next step, soil samples are taken in the immediate vicinity of the individual sub plots. These samples are used for reference analyses at various labs.
These soil samples are labelled and immediately deep frozen. The frozen samples are stored until they are sent by express mail to the labs for reference analysis.

The reference labs

The reference analyses are carried out by five ­different accredited labs according to recognised scientific methods.

Evaluation

The NO3, Nmin and soil moisture levels are grouped into practice-oriented classifications (see tables 3 to 5).
To measure agreement, a procedure was developed in cooperation with the Julius Kühn Institute, the German Federal Research Centre for Cultivated Plants. This procedure evaluates the agreement between the sensorbased class predictions with the mean lab results (the class which the mean lab result is grouped in) on the one hand and the agreement between the individual lab results with the mean lab result class on the other hand. This procedure is applied to each plot and soil parameter.

The classification results per field are collected and entered into a confusion matrix which is used to calculate agreement. Cohen‘s kappa and weighted Cohen‘s kappa serve as parameters to describe the predictive accuracy of the soil sensor. Cohen‘s Kappa quantifies agreement without taking the level of misclassification into account (class distance). The values that are scattered around 1 indicate considerable agreement whereas the values scattered around 0 and those smaller than 0 indicate poor agreement.

The weighted Cohen’s Kappa quantifies agreement by taking into account the level of misclassification. To do this, the level of misclassification obtained by comparing the sensor measurements to the mean lab result is calculated. Such misclassifications are due to variations in soil quality and conditions which occur even in sub-fields, for example. 

To assess the results, the level of misclassification by the sensor is related to the level of misclassi­fication by the labs. 

Assessment

The results are assessed by applying a system that was developed by the DLG expert group together with the Julius Kühn Institute. The following diagram (Figure 5) shows the assessment workflow and required prediction accuracies. Error detection and system prompts Error and instruction prompts following an operator error, which also form a claimed feature of the device, were tested by provoking errors systematically in the field test. 

Table 3: NO3 classes

Table 4: Nmin classes

Table 5: Soil moisture classes

The Stenon FarmLab is an integrated hardware and software solution that carries out soil analyses and delivers real-time results. The test verified whether the system provides information on the soil condition which is of practical relevance for growers and their nutrient and water management decisions. In addition, it was tested whether the system detects operating errors and prompts appropriate warnings and instructions.

The field measurements were carried out in the Darmstadt-Dieburg area (Hesse) in May 2021 by using five handheld devices that used identical software versions and were calibrated to identical models. All devices were operated in typical field conditions.

Soil properties in the test fields

The fields measured and sampled with the Stenon FarmLab soil sensor were chosen and selected with the help of the Hesse State Office for Agriculture (LLH). The number of 40 different test fields reflects a relatively wide range of different soil properties and meets the aim of the test design, which is covering the widest possible spectrum of soil conditions. The individual spectrums are shown in Figures 6 to 9.

Measurement imprecisions

The manufacturer specifies measurement imprecisions per parameter and range, which are listed in the tables 6, 7, 8 below. The degree of precision is expressed by the root mean square deviation (RMSE) and the median absolute error (MedAE). The measurement imprecision value per measuring range served as a basis for extrapolating the measuring range of the device as claimed by the manufacturer.

Table 6: NO3 measurement imprecisions

Table 7: Nmin measuring imprececions

Table 8: Measuring imprececions for soil moisture

Table 9: Soil parameters and reference methods

Table 10: Stenon FarmLab prediction accuracy

Error detection and system prompts

The operator warnings listed in table 11 were provoked five times in the experiment. The Stenon FarmLab detected each error and prompted to the operator the appropriate warning on the display screen.

Table 11: Operator errors and error prompts on the discplay screen