Hydro N-Sensor: A system for on-line variable N application

Hydro N-senzor: Technologie pro přímé diferencované hnojení dusíkem

Olfs Hans-Werner

Abstrakt

Vývoj metod umožňujících optimalizaci dávky dusíku pro přihnojování porostu a pro diferencované hnojení N je vyhodnocen v předloženém příspěvku. Je popsán princip funkce N senzoru a jeho schopnost eliminovat změny povětrnostích podmínek, především osvětlení. Praktické využití senzoru při přihnojování dusíkem je porovnáno s klasickým rovnoměrným hnojením. Vyšší homogenita zeleného zbarvení porostu po diferencovaném hnojení se potvrdila i lepší vyrovnaností výnosové mapy a zvýšeným výnosem (0,24 t/ha) v porovnání s pozemkem hnojeným klasicky.

Precision farming aims to improve the management of agricultural inputs by using them according to the spatially variable requirements of a crop, in order to obtain the economic optimum yield and/or quality at each spot within a field (Cassman & Plant, 1992). Although nitrogen (N) is the most interesting nutrient for spatially variable application with the highest economic and environmental potential, other nutrients (P, K and lime) are in the forefront of precision farming (Ferguson et al., 1996). The reason for this is that appropriate methods to predict the N fertiliser requirement at a high spatial resolution have been missing, e.g. soil sampling for mineral N in the 0 - 90 cm layer on a grid basis is a time consuming and quite costly procedure. Furthermore the information is not directly available for the farmer and due to nitrate leaching the soil N status might even have changed until N application based on that soil analysis can take place. The challenge for precision farming concepts is therefore to determine and correctly interpret the variability of soil and/or plant features as precisely, quickly and cost efficient as possible in order to vary the fertilizer rate according to the crop requirements.

Determination of the N fertiliser rate

For the determination of the optimum N fertiliser rate, it is important to consider the N supply from the soil (Fig. 1). The N supply from the soil consists of the inorganic N content in soil in early spring (Nmin) and the net mineralisation during the crop growth. Soil analysis is used to measure the Nmin content in the rooting zone at the beginning of the plant growth in spring. However, after soil analysis, N is subsequently exposed to leaching, immobilisation and denitrification processes which vary both in space and time (Simmelsgaard & Djuhuus, 1997). Therefore, methods based on soil analysis to correctly determine the available N for the crop have limited reliability. However, due to the lack of alternatives, soil Nmin analysis has been successfully introduced in Europe as the basis for the recommendation of the first N application in spring.

1. Determination of N fertilizer requirement

The potential of the soil to mineralise N during the growing season varies widely depending on soil and weather conditions and can not be correctly predicted by soil analysis (Olfs & Werner, 1994). Therefore, farmers and scientists have been seeking alternative methods, based on direct plant analysis in the field for determining the N rate for subsequent application. In the past the most commonly-used method has been visible judgement based upon crop colour (e.g. Früchtenicht, 1965). In the sixties, this qualitative approach was turned into a more quantitative one by using colour cards to compare with the actual crop colour. As this approach was not calibrated to give any N-recommendation, it failed to find its way into practical use in agriculture.

The next generation of tests were based on the rapid analysis of N in the plant (Binford et al. 1990). The nitrate sap test (e.g. Wollring & Wehrmann 1990) determines the nitrate content in the stem base of cereals by using a chemical (diphenylamine/sulphuric acid) which reacts with nitrate to form a colour complex. This was the first instantaneous method to give a quantitative N recommendation for the second and third N dressing for cereals. This test was introduced into German agriculture in the mid-eighties after several years' field calibration trials. However, these methods have not been widely accepted by farmers within Germany and other European countries because of the complexity, which involve for example cutting the stem into pieces and use of aggressive chemical reagents.

Starting in the late eighties a small handy lightweight chlorophyll meter (Minolta Spad 502, Hydro N-Tester), which measures the leaf chlorophyll concentration non-destructively and directly in the field, through optical transmittance was tested in field trials (Jemison & Fox, 1988). Close correlations between the N-Tester value and the chlorophyll content as well as the leaf N concentration (Fig. 2) made it possible to establish a recommendation system for the 2nd and 3rd N dressing to cereals. Compared to other methods the N-Tester has a clear advantage in that it is simple to use and gives an immediate N recommendation, without costly and time-consuming laboratory analysis. Since it was first introduced on a large scale in 1995, it has been widely accepted by both farmers and agricultural advisors in several European countries (e.g. Germany, France, Czech Republic, Poland, Sweden, Switzerland and the United Kingdom). Recommendations are updated annually to take into account latest trial results, and all newly registered cultivars.

2. Relationship between N-Tester value and leaf N content of winter barley at EC 31

Simple to use methods for quick plant analysis (nitrate sap test, N-Tester) were breakthroughs in the determination of the optimal N rate directly in the field, but recommendations using these methods are made for the whole field as if it were homogenous, without considering spatial variability of N supply and/or yield potential (Fig. 3). Applying a constant amount of N on the whole field will result in loss of yield on those parts of the field where the N rate is not sufficient to cover the N need of the crop for a high yield. On other parts of the field the crop is overfertilized because the yield is limited by other growth factors, e.g. available water, other nutrients than N.

Furthermore field trial results suggest, that the cost for the required grid sampling can not be covered by the relative small yield increase: field trials carried out to compare variable and uniform N application based on grid sampled Nmin content in spring for the 1st dressing and N-Tester values for the 2nd and 3rd N dressing have resulted in an average yield increase of only 3 % by variable N application. Therefore, remote sensing is seen as the future method to determine the crop's N demand on a spatial basis because it allows cost effective on-line measurements directly in the field at the required high resolution

3. A constant N application rate results in over- and underfertilization for certain parts within a field

Spatially variable N application with the Hydro N-Sensor

System description

At Hydro Agri´s Centre for Plant Nutrition and Environmental Research Hanninghof in Dülmen, Germany, a tractor mounted sensor system has been developed to measure canopy reflectance. Tractor-based sensors overcome limitations of satellite-based remote systems (limitation of data availability during cloudy conditions and the delay in delivery of this data to the farmer) by taking on-line images of the crop, with simultaneous data processing during the operation allowing simultaneous variable N application (Reusch 1997).

4. Hydro N-Sensor: System components

The Hydro N-Sensor´s main unit contains 2 diode array spectrometers (400 - 1000 nm), control electronics and a computer in a rugged and waterproof case mounted on the top of the tractor's cabin (Fig. 4). Each of the spectrometers is connected to the optical inputs through fibre optics. One spectrometer collects radiation from the left and right side of the tramline, whereas the second spectrometer is used to measure the actual irradiance conditions compensating different light situations (e.g. bright sunlight vs. cloudy sky). The system is controlled by the driver through a terminal, which displays current system information and logs the data on a chip card. From the reflectance spectra so-called "spectral indices" or "sensor values" are derived by mathematically combining reflectance at several wavelengths which show high correlation to the nutrient status of the crop. Several year's results derived from N rate trials were used to establish the relationships between these values and the optimum N application rates. This algorithm is now used for transferring sensor values into application rates.

N recommendations derived from N-Sensor values show a high reproducibility even when crop and weather conditions change dramatically. 12 repeated measurements of a 250 m tramline during one day starting in the morning with a wet crop and windy, grey weather, while during noon the crop had partly been dried with windy weather and in the afternoon the crop was totally dry and the sun was shining the average of all standard deviations was just 3 kg N/ha (Fig. 5).

5. Reproducibility of N recommendations based on N-Sensor values derived from repeated measurements of one tramline

Agronomic calibration

Since 1997, field trials with winter wheat were carried out to test the sensor's performance regarding variable application of N compared to the uniform application of the same amount of N according to conventional practice. The typical field size was between 5 and 40 ha. The first N application in spring was applied uniformly as usual over the whole field. One or two strips perpendicular to the tramlines received additional N at the first N dressing in spring, to test if the N-Sensor is able to identify this artificially-induced N variability at the second dressing.

In each field for the 2nd and 3rd application two treatments were imposed: one sensor-controlled or "variable" treatment and a "uniform" treatment, in which the fertiliser was spread at a uniform rate (Fig. 6). Each treatment included a minimum of four tramlines. To be able to compare the results of the treatments, the average application rate at each N dressing was exactly the same in both treatments. This was achieved by first fertilising the variable treatments, then calculating the average N rate from the area and the total fertiliser amount used. This average N rate was then applied uniformly on the uniform treatment.

6. Schematic trial set-up for N-Sensor trials

Results

N-Sensor maps show typical inhomogenities across the field: for example areas with low chlorophyll content in the middle of the north-eastern part of the field (Fig. 7). On the other hand, high chlorophyll densities can be found especially in the south-west corner. In addition, the strip with the extra 40 kg N/ha applied at first dressing can clearly be found in the southern third of the field, reaching from west to northat a width of 20 to 30 m, matching the spreading width of the spreader. This proves that the sensor system is working correctly by differentiating areas with different N supply.

7. N-Sensor values across a field at second application

Based on the agronomic calibration a N application map can be calculated by transforming sensor values into application rates (Fig. 8). For the sensor-controlled tramlines, this map shows how the N rate was actually varied within the field.For the treatments with a uniform N rate, this is only an "apparent application map", that would have been given if the fertiliser had been applied variably. In reality, these tramlines have received a uniform application according to the trial layout.

As expected, this map and the sensor value map show similar patterns: areas with high application rates correspond to low sensor values and vice versa. Typical application rates in this field range from nearly 0 to 90 kg N/ha for the second N dressing. Similar results (in some cases even ranges between 0 and 120 kg N/ha) have been found on other fields where sensor trials have been conducted.

8. Apparent N application rates calculated from sensor values

To assess effectiveness of the sensor on crop homogeneity both variable and uniformly applied treatments were sensed again 3-4 weeks after the second N rate was applied. For the field mapping, sensor values were separated into two data sets, one for the uniform and one for the variable treatment. By this approach, two sensor maps from the same field were generated, reflecting the variable and uniform treatment in the field (Fig. 9). Visible comparison already indicate the effect of variable rate application on reducing the variability of crops' chlorophyll content. Statistical analysis showed that the standard deviation of the variable treatment was only 58% of that of the uniform treatment. As, at the third N dressing, the only difference between both treatments is the different fertiliser distribution at the second N application, this decrease of within-field variability can clearly be assigned to the sensor-controlled variable application of the second dressing. This effect of a more homogenous crops after variable rate N application is found in more than 90% of the trials. However, it can be questioned if a more homogeneous crop in terms of chlorophyll content during the growth period as such is advantageous for the farmer.

Based on the calibration trials of the sensor reading in field experiments, it is found that the N rate decreases with increasing chlorophyll content and vice versa. This principle is not valid in extreme situations e.g. very low sensor readings caused by sub-optimal plant densities. In such situations, the sensor analyses spectral indices that contain information on the total crop biomass and as a consequence of low biomass the system modifies the N recommendation to adopt the N rate to the lower N demand of the crop.

The effect of variable N application on final yield data is of most interest to the farmer. Harvesting was done with a combine harvester equipped with a yield meter and a DGPS-positioning device. In addition, every tramline's yield was measured manually by accurate weighing to double-check the data. Yield maps were produced from each treatment separately to visualise the same field treated either uniformly or variably with the sensor (Fig. 10).

From the yield map it can be seen that on this field the variable treatment results in a yield increase of 2.4 dt/ha compared to the uniform treatment. Though this is not dramatic, it should be considered that this increase was obtained only by different distribution of the same amount of fertiliser. Moreover, there was no additional information (such as soil maps or yield maps) or work (such as soil sampling and testing) necessary. In 1999 totally 96 field trials following this experimental protocol were conducted in several countries (CZ, D, Dk, F, GB, H, I, S, and USA). For these wheat trials the average yield increased by 1.6 dt/ha (= 2.12 %), leading to a profit advantage of 10.3% (Fig. 11). In 7 US maize trials the yield was improved by 1.6% for the N-Sensor treatment compared to uniform application (data not shown). The overall increase in N efficiency was 4.4 kg N/ha, which leads to reduced Nmin content after harvest.

More uniform crops as a result of variable N rate according to the crops' spatially-variable requirement will have other positive effects. With the sensor system regional overfertilisation can be avoided, which will reduce lodging and as a result will significantly increase the yield in these areas. Besides lodging, the residual N content in the soil after harvest will be reduced especially in parts of the field where excess N inputs are avoided. As fungal infestations increase when N is applied in excess, variable rate N supply will contribute to reduce such disease problems. Other positive effects are a more uniform ripening of the crop which will reduce harvesting costs, as well as losses of the grain. Futhermore results show a slight increase (+ 0.1 %) in grain protein due to variable N application and less variability in protein content over the field, which will result in more uniform quality for the grower.

Conclusions

A number of different methods are advocated to improve site-specific N management through variable rate N application. However, the accuracy and the cost of these methods to predict the spatially variable optimum N rate will finally determine the acceptability by farmers. Most results show that yield and environmental benefits of site-specific N management compared to conventional uniform application are relatively small. Therefore only methods which can inexpensively provide the spatially variable N requirement will find wide acceptance. Variable N rate application, based on soil sampling, would require a even higher grid sampling density than that usually considered for P and K. Alternatively, the growing crop can be used as a direct indicator of the crops' N fertiliser requirement, as it reflects soil N availability. Compared to methods based upon soil/crop sampling and analysis, sensing systems certainly have more potential to cost-efficiently monitor crop N status at the required high resolution. The tractor-mounted crop sensing system which has been developed to remotely monitor the N requirement in real time at high resolution, while the spreader varies the N rate accordingly is regarded as a first practical step to improve site-specific N management.

References mentioned in the paper are available from authors.

9.

10. Sensor values across a field at third dressing (data has been separated into two data sets, one for uniform and one for variable treatments)

11.

Yield maps from one field (data has been separated into two data sets, one for uniform and one for variable treatment)

12. Results from 96 N-Sensor trials in 1999