Years ago, thanks to some colleagues from Itagra, I knew of the existence of a handheld meter of nitrogen / chlorophyll content in crops called N-tester. This was distributed by Yara, a Norwegian multinational, although I think the device itself was manufactured by the Japanese company Minolta and now are other manufacturers and distributors. This device is intended to determine the nitrogen requirements of plants (cereals, mainly) directly in the field, in order to adjust the nitrogen fertilizer needed for the stem elongation and head emergence. The N-tester measures the chlorophyll content of the leaf, and how this relates to the status of its nitrogen nutrition (if there is no other obstacle to the crop development, of course) you can calculate the nutritional status of the crop , at least with respect to nitrogen.  

Yara N-tester doing a measurement. Source: Yara

This calculation is very important because for a proper development of the crop, the application of enough quantities of this macro-nutrient (basic for the crop together with phosphorus and potassium and several micro elements) is required. Furthermore, it is necessary to apply at the right time and properly distributed, because otherwise some zones could be "supercharged", leaving other crop areas other insufficiently nourished. On the other hand, excessive application of nitrogen increases the risk of leaching and / or run-off, which can affect the quality of groundwater, or surface water may be contaminated. N-tester operation is quite simple. The leaf you want to analyze is placed in a kind of clip in which on one side are two diodes emitting at 650 nm (red) and 960 nm (near infrared) respectively. A receiver (photodiode) collects the light passing through the leaf and a microprocessor calculates the result that is convenient presented on the LCD screen. The number presented is dimensionless and may be related to the amount of nitrogen through tables provided by the manufacturer for different crops. There are many studies available about this. Try Googling, if you are interested in exploring this topic. So, device operation was interesting, and this, together with our interest in things related to remote sensing and the improvement of agricultural production and the development based on open hardware and software, led me to try to create a device with similar functions, but with a fraction of the cost of the N-tester. In this way, and with our own resources, Nduino was developed.

Nduino final prototype.

Nduino is a device to measure the leaf chlorophyll / nitrogen content, based in the Arduino microcontroller, and specifically, in this first prototype, in Arduino Uno. It consists, like the N-tester, in a clamp where the leaf sample is placed. A LED emits a white light which is reflected on the surface of the leaf, and the reflected light is collected by a RGB sensor. The information captured by the sensor is transmitted to the Arduino, where a software developed for that purpose performs a series of calculations and provides a greenness index. Additionally the location of the measurement is recorded by a GPS receiver and the information is recorded in a SD memory card to export to our favorite GIS. Best of all, the cost of prototype materials (excluding development and assembly) be around € 120-150.

Nduino first prototype.

Here you can have detected some of the differences that have Nduino and N-tester. First, the measurement is made completely in the visible spectrum rather than in the red and near infrared bands. Second, Nduino measures the reflected light and not the transmitted light. Third, the collected data is georeferenced, so it is possible to map the greenness plot in a GIS. And fourth, the measurements are stored in a memory card, in CSV format, so you can dump the information to any statistical analysis software or, as I said, to a GIS. There are several improvements, planned, of course, but there is a major factor in measuring in visible bands of spectrum, which can be a problem. The choice of an RGB sensor has its foundation in a lower cost, but usually greenness indexes work with near-infrared and red bands. This is because the vegetation appears relatively dark in the photosynthetically active region and is relatively bright in the near infrared band, allowing relatively easily identify the growing level of a crop using only those regions of the spectrum. For example, the well-known NDVI uses only these two bands (which are the same that N-tester uses). In the Nduino case, by providing only the red, green and blue bands, the solution came through vegetation indices based on visible bands (we talk about them in a future article) such as the rate of excess green (ExG) and others as the rate of excess red (EXR) and the subtraction ExG-ExR. The ExG was described by Woebbecke et al (1995) and the other two by George E. Meyer et al (2008). Thus, the EXG, for example is the result of the next operation, 2 · GRB, and ExR is the result of applying 1.4 · RB, where R, G and B are the normalized channels of red, green and blue. As can be seen, the index is dimensionless, like the NDVI. There are several studies that have compared these visible greenness indices with other indices using near infrared, finding generally a strong correlation between both. In the pictures below you can see a representation of the ExG and EXG-ExR indexes generated from an RGB image using GRASS GIS.

RGB image of an abandoned lot.


ExG index of the RGB image.. In red, the greatest photosynthetic activity.


ExG-ExR index of the RGB image. In red color, the greatest photosynthetic activity.

In this case I tried to test if the Nduino can make some measurements that could be related to those carried out with the N-tester. To simplify the tests I used leaves that I had on hand, such as Ficus benjamina, Robinia pseudoacacia and Solanum jasminoides. I used leaves in various stages of senescence. 18 samples were used in the first case, 10 in the second and 12 in the third. I understand it's not a very generous sampling, but for a first approximation can serve. With the Nduino I took the mean of 3 samples per leaf, and for N-tester the average of 30 samples per leaf were taken as it is required by the device to give a measurement. Simple linear regression between the measurements of several indices RGB vegetation and N-tester was conducted. I will not go into depth on statistical results. As a simply summary, in the case of ficus, the best index was EXG-ExR (-0.84 correlation coefficient and R-square of 70.4%, with a P value of 0.0000). For Robinia, the better index was EXG-ExR too, with a correlation coefficient of 0.79, an R-square of 63.0% and a P value of 0.0062. Finally, in the case of Solanum jasminoides the index with better behavior was ExG, with a correlation coefficient of -0.92 and an R-square of 84.4%. As in the first case the P value was 0.0000. The ExG-ExR index in this last case gave a correlation coefficient of 0.71 and an R-squared of 50.63% with a P 0,0095. What can we conclude from these first tests? First, it is necessary to take more samples and measurements in crops with agricultural value (corn, beet, ...). Second, the tests data has to be compared with NDVI measurements obtained from aerial images (from a drone, for example) and do a deeper statistical analysis to develop models of interpretation of the indices for these crops, equivalent to those offered with the N-tester, but preferably associated with a GIS. In any case, it seems clear that the relationship between measurements exist, which may be of interest in those places where there is a need to adjust fertilization and is not possible, for economic reasons, using measurements with N-tester, drones , satellites, etc. For example, in developing countries. We can think in a device that was calibrated for a specific crop, and can give a simple orientation of fertilization per hectare in each area of ​​a plot, using a device like this and a GIS open as QGis or Grass or an online GIS designed for this purpose, for example. In the same way, it could be applied to the colorimetric measurement of fruits, grapes, olives, ... so that if we find the right mathematical relationship, we could organize the harvest by qualities, e.g. The equipment can be miniaturized, as there are quite smaller designs than Uno. And its low cost can make it accessible through NGOs to farmers in developing countries, for example. So the idea is simple but applications are promising. We try to make them happen.