1. Introduction
Fertilizer usage represents an important part of traditional agriculture and crop yield. In a world of growing food (and other agricultural products) demand—estimates indicate up to 50% increase in the 2012–2050 time frame [
1]—fertilizer (ab)use is seldomly a go-to solution for crop yield increase. Additionally, although growth rates for arable land are expected to increase within a sustainable manner, if an arable land loss scenario due to climate changing conditions is taken into account [
2], further conflicts and competition might arise between protected lands, agricultural exploitation and human expansion.
Considering these concerns—well reflected by the United Nation’s 2030 Agenda for Sustainable Development [
3]—and also motivated to provide a solution for sustainable agriculture, our group has undertaken the task to develop a technology that is able to help farmers ensure that their crops’ needs are being met, through their fertilization procedures. Knowing what is being fed to the crops and what is being taken up, it is possible to reduce water/fertilizer consumption to an optimal level, reducing the operational costs, whilst allowing crops to develop at their optimal speed, towards a bigger crop turnover.
Spectroscopy is, among others, one of the most well-established techniques for chemical identification and quantification. Several chemical determination methodologies that rely on spectroscopy exist (e.g., ICP-AES, LIBS, FTIR, GC-VUV); nevertheless, the following limitations also exist: the sample must be responsive to electromagnetic radiation (absorption/emission); linearity outside the Beer–Bouguer–Lambert Law [
4,
5,
6] can sometimes be problematic, or the simple fact that spectroscopy is a molecular-level information tool, which can add entropy to the analysis by providing a wider range of information than the one desired. If pure compounds are analysed, little or no interference exists; when more complex mixtures are targeted, interferences might play a key role on the successful outcome. In such cases, in order to obtain an accurate and reliable measurement, interferences have to be taken into account. Chemometrics presents itself as a putative solution to, by employing varying complexity mathematical calculations, together with statistics and algorithms, allow the extraction of relevant information from the superimposed and—sometimes—latent data. Linearity-based models are unable to solve the interference pattern between any constituents present; this is the case for either interferents and non-interferents (target analytes), due to the fact that light has a wave-like nature and, hence, the sample information might suffer from constructive or destructive interferences [
7]. Nevertheless, new chemometrics methodologies that encompass interferences already allow critical developments to be achieved, e.g., on health-related point-of-care analysis [
8].
In hydroponics, most of these interferences can be attributed to the fertilizers. Fertilizers are mixtures of several different nutrients, mostly in their inorganic salt form (e.g., MgSO4, CaCO3, FeCl3) whilst some might be in aqueous solutions (e.g., Mo, Ba, B). In complex mixtures, some signals might superimpose over others, causing a concentration misevaluation, or resulting in a continuous spectrum of overlapping signals.
This study aims to provide insight on the interferences within a complex matrix orthogonal design consisting of 8
3 independent concentration Hoagland solution samples. The performed assay further complements on our previous findings [
9] on the feasibility of information extraction of highly constrained samples, by using an advanced algorithm—self-learning artificial intelligence (SLAI)—in order to find the adequate co-variance modes for accurate model prediction.
2. Materials and Methods
Hoagland solutions were chosen as a matrix due to their widely accepted status among the agronomical community as being a good model for complex nutrient solutions in hydroponics. Stock Hoagland solutions are composed as described by
Table 1.
Three stock solutions for N, P and K were freshly and individually prepared in order for it to be possible to vary each target element (N, P or K) independently. Each stock solution comprised of ionic elements already present in the base matrix, as follows: N—NaNO3/NH3; P—H3PO4; K—KCl. The ratio of NaNO3/NH3 of the stock spiking solutions was the same as the ratio on the Hoagland solution (≈93:7).
Final concentrations of all samples (matrix + individual spikes) were corrected taking into account any variations derived from the preparation of fresh stock solutions each day during the execution of the assay.
The tested final concentration ranges, for the target analytes, were as described in
Table 2.
The designed orthogonal matrix was composed by 83 samples, each one with an independent N, P and K concentration level. At each corresponding level, the N, P and K corresponding spike was added to the matrix and stirred for 10 s in order to attain full homogeneity. Afterwards, the pH value was registered (Crison GLP 21, Crison Instruments, SA, Barcelona, Spain) and the sample pumped into a custom-built flow cell for spectral data acquisition. Each sample had a final total volume of 30 mL (Hoagland + [N + P + K] spikes). After data acquisition, the sample was discarded and the system flushed with deionised water.
Data acquisition was performed with an in-house LabView-based developed software (National InstrumentsTM Corp., Austin, TX, USA) for pumping control and data acquisition.
Sample irradiation was performed with a D
2 light source (FiberLight
® D
2 HighPower DTM 10/50S, Heraeus Noblelight GmbH, Hanau, Germany) whereas the detection was performed by a miniaturized spectrometer with a 190–650 nm range (STS-UV-L-50-400, Ocean Insight, Inc., Orlando, FL, USA). Individual components were assembled with custom-length 600 nm UV-Vis optical fibres (in-house customization), as depicted in
Figure 1.
3. Results
The obtained results from the execution of this matrix were compiled and are depicted on
Figure 2.
The collection of spectral data and cross correlation with the concentration information for each solution was performed. Spectroscopy signals were processed accordingly to [
7]. Nevertheless, using advanced signal processing it is possible to train the system to recognize and extract the information from the relevant features, incorporating multi-scale interference into the NPK quantification models.
The correlation of the different levels among the NPK nutrients of the matrix design, can be represented as displayed by
Figure 3a whereas
Figure 3b shows the corresponding recorded spectra in the UV-Vis region (
circa 200–650 nm) of the factorial design samples. As expected, most of the systematic spectral variation occurs at ≈250 to 450 nm, and, to a lesser extent, to 500 nm. This figure provides evidence that information about P and K is present, because, even to the naked eye, one can observe that there are more spectral patterns in the region of ≈250 to 450 nm than the expected nitrogen levels of the experimental design; that is a good indication that the interferences between all the constituents are being registered on the spectra.
The principal component analysis (PCA) (please refer to
Figure 4a) scores plot of the corresponding experimental design spectra is shown, where the different colours represent the different levels of total nitrogen. The main variance present in the spectra corresponds to the nitrogen absorbance, where the first principal component is highly correlated to the nitrogen content. It is also possible to see that the K-level information is embedded inside each N-concentration level. Analysis of the second component allows to unveil that information of K-level also carries the information of the different P-levels of the sample matrix (please refer to
Figure 4b).
Using the data obtained from the executed matrix, it was possible to train the self-learning AI of the system in order to quantify N and K with 6.7% (0.997) and 3.8% (0.987), respectively, and to obtain qualitative results for P, as shown in
Figure 5.
4. Conclusions
A NPK spectroscopy-based, AI-supported by a robust self-learning artificial intelligence was developed in order to be able to cope with increasing interference complexity of fertilizer solutions in greenhouses. The obtained results allow to be inferred that the current system’s performance is adequate for Hoagland solutions, which are used in research and high-end hydroponic systems.
The assembled system aimed to keep a good balance between cost–benefit, without relinquishing reliability, robustness and accuracy; this objective has been successfully attained.
Further analysis of the results—not within the scope of this manuscript—as well as of unpublished data, allows further developments to be implemented to the system/prototype, in order to enhance its robustness and accuracy.
Author Contributions
Conceptualization, A.F.S. and R.C.M.; methodology, A.F.S. and R.C.M.; software, R.C.M.; validation, A.F.S. and R.C.M.; formal analysis, A.F.S. and R.C.M.; investigation, A.F.S.; resources, R.C.M., L.C. and P.J.; data curation, A.F.S.; writing—original draft preparation, A.F.S.; writing—review and editing, A.F.S., K.L., M.G., E.V.O., G.F., J.B., T.M.P., J.B.-C., L.C., P.J. and R.C.M.; visualization, A.F.S., R.C.M.; supervision, R.C.M. and P.J.; project administration, J.B.-C.; funding acquisition, J.B.-C. All authors have read and agreed to the published version of the manuscript.
Funding
This work was funded by the ERA-NET Cofund WaterWorks2015 Call, within the frame of the collaborative international consortium AGRINUPES. This ERA-NET is an integral part of the 2016 Joint Activities developed by the Water Challenges for a Changing World Joint Programme Initiative (Water JPI/002/2015). R.M. acknowledges Fundação para a Ciência e Tecnologia (FCT) research contract grant (CEEIND/017801/2018). A.F.S. gratefully acknowledges the financial support provided by FCT (Portugal’s Foundation for Science and Technology) within grant (DFA/BD/9136/2020).
Institutional Review Board Statement
Not applicable.
Informed Consent Statement
Not applicable.
Data Availability Statement
Not applicable.
Conflicts of Interest
The authors declare no conflict of interest.
References
- FAO. The Future of Food and Agriculture—Trends and Challenges; FAO: Rome, Italy, 2017. [Google Scholar]
- FAO. The Future of Food and Agriculture—Alternative Pathways to 2050; FAO: Rome, Italy, 2018. [Google Scholar]
- United Nations. Transforming Our World: The 2030 Agenda for Sustainable Development; United Nations: San Francisco, CA, USA, 2015. [Google Scholar]
- Lambert, J.H. Photometry, or, on the Measure and Gradations of Light Intensity, Colors, and Shade; Eberhardt Klett: Augsburg, Germany, 1760. [Google Scholar]
- Bouguer, P. Optics Essay on the Attenuation of Light; Claude Jombert: Paris, France, 1729. [Google Scholar]
- Beer, A. Determination of the absorption of red light in colored liquids. Ann. Phys. Chem. 1852, 162, 11. [Google Scholar]
- Martins, R.C. Big Data Self-Learning Artificial Intelligence Methodology for the Accurate Quantification and Classification of Spectral Information under Complex Variability and Multi-Scale Interference. WO Patent WO2018060967, 5 April 2018. [Google Scholar]
- Barroso, T.G.; Ribeiro, L.; Gregório, H.; Santos, F.; Martins, R.C. Point-of-care Vis-SWNIR spectroscopy towards reagent-less hemogram analysis. Sens. Actuators B Chem. 2021, 343, 130138. [Google Scholar] [CrossRef]
- Monteiro-Silva, F.; Jorge, P.A.S.; Martins, R.C. Optical Sensing of Nitrogen, Phosphorus and Potassium: A Spectrophotometrical Approach Toward Smart Nutrient Deployment. Chemosensors 2019, 7, 51. [Google Scholar] [CrossRef] [Green Version]
| Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).