Next Article in Journal
Cultivation Substrate Composition Influences Morphology, Volatilome and Essential Oil of Lavandula Angustifolia Mill.
Next Article in Special Issue
Unveiling the Efficiency of Vermicompost Derived from Different Biowastes on Wheat (Triticum aestivum L.) Plant Growth and Soil Health
Previous Article in Journal
Contrasting Effects of NaCl and NaHCO3 Stresses on Seed Germination, Seedling Growth, Photosynthesis, and Osmoregulators of the Common Bean (Phaseolus vulgaris L.)
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Agronomy
Volume 9
Issue 8
10.3390/agronomy9080410
agronomy-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Multi-Factor Diagnostic and Recommendation System for Boron in Neutral and Acidic Soils

by
Richardly Lenz Clove Dupré
,
Lotfi Khiari
*,
Jacques Gallichand
and
Claude Alla Joseph
Département des Sols et Génie Agroalimentaire, Université Laval, Québec, QC G1K7P4, Canada
*
Author to whom correspondence should be addressed.
Agronomy 2019, 9(8), 410; https://doi.org/10.3390/agronomy9080410
Submission received: 19 June 2019 / Revised: 18 July 2019 / Accepted: 22 July 2019 / Published: 25 July 2019
(This article belongs to the Special Issue Soil Fertility Management for Better Crop Production)
Browse Figures
Versions Notes

Abstract

:
Despite its inconveniences, the most recognized method to extract boron from soils is that of hot water extraction (BHW), which is used for diagnostics and recommendations. However, the Mehlich-3 (M3) method is widely used to extract and diagnose several elements at once (P, K, Ca, Mg, Al, B, Cu, Zn, Fe, and Mn) and is well adapted to routine analyses. The objective of our study was to develop a soil diagnostic and recommendation system for boron as a function of measured BM3 (and other interacting elements), crop type, and spreading methods. This system is based on three databases from either the international literature or the chemical characterization of acidic-to-neutral soils typical from Québec (Canada). The first database came from the characterization of 365 samples typical of Québec soils; it has been used to predict, by the AutoML (Automatic Machine Learnig) supervised learning algorithm, BM3 as a function of a set of parameters from the following: BHW, pHW, organic carbon (OC), CaM3, KM3, and MgM3. Depending on the parameters used, the R2 between the measured and observed BM3 varied from 0.36 to 0.99. This database allowed us to define two classifications for soil boron diagnostics and fertility evaluation. The Cate–Nelson analysis for these two models allowed us to define three boron fertility classes: Low, medium and high; that is 0.00–0.23, 0.23–0.58, and 0.58–3.70 mg B kg−1, respectively, for BHW, and 0.00–0.65, 0.65–1.03, and 1.03–12.70 mg B kg−1, respectively, for BM3. The third database was extracted from 130 yield responses to increasing levels of boron; it was used to define a recommendation model for boron, based on AutoML, as a function of BM3, pHW, the crop boron requirement (medium, high), and the type of spreading (broadcast, sidedress, foliar spraying). This model resulted in an R2 of 0.63.

Graphical Abstract

1. Introduction

Fertile soils allow for root development and provide for the plant extraction of water and nutrients [1]. It is therefore important to evaluate agricultural soil potential using analytical tests for soil diagnostics. Despite the fact that the extraction of trace elements by plants is low, these elements are limiting for crop yields [2]. The characterizing and dosing of trace elements depend on analytical artefacts, since a change in process or a low contamination may greatly affect results [3]. In Québec, fertilization grids were mainly developed for major nutrients [4]. Soil chemical fertility diagnostics for most nutrients (P, K, Ca, Mg, Cu, Zn and Mn) are done after a Mehlich-3 solution extraction. However, some trace elements, such as boron, are not on this list since there are no calibrations with yield. For boron, most calibrations with crop yield are based on hot water extraction (BHW). Until recently, the BHW method was widely used, since it is efficient in detecting and analyzing low soil boron concentration by colorimetry [5]. For alfalfa, Gestring and Soltanpour [6] compared three boron extraction methods (saturation, ammonium bicarbonate-DTPA, and hot water), and found that hot water extraction was the most related to yield [6]. However, cooling time and additional manipulations required to compensate water loss by evaporation during extraction make this method slow and imprecise [7,8]. Contrarily to Mehlich-3 extraction, hot water extraction is slower and can be done for one element at a time; it is therefore less interesting economically than Mehlich-3 [9] and for soil routine laboratory analyses [10]. The inductively coupled plasma (ICP) spectrometer can analyze multiple soil elements [11] with a same Mehlich-3 solution; this technology allows for the considerable reduction of the costs of soil analyses and an increased efficiency [12]. Shuman et al. [10] proposed to substitute BHW by BM3 for soil boron diagnostics in six eastern US States. Zbíral and Němec [9] found a significant difference between BHW and BM3, but the correlation was not high enough to make the substitution. Additionally, Tran et al. [13] proposed a conversion equation (BM3 = 1.31 × BHW) for acidic sandy soils and loamy soils in Québec. However, for clay soils with a high exchangeable Ca content and higher pH, the correlation between BHW and BM3 was almost non-existent. Many diagnostic and recommendation grids based on BHW and crop type have been reported in the literature [14,15,16,17,18,19,20,21,22]. However, to our knowledge, there exists no boron diagnostic and recommendation grid based on BM3. The objectives of this study were: (i) To evaluate BM3 from BHW for different representative soil types in Québec; (ii) to build, with data from the literature, two agronomical relative yield models, relative yield (RY) = function (BHW) and RY = function (BM3); and (iii) to develop, with yield-boron response curves from the literature, a recommendation relationship based on soil BM3, soil pH, boron spreading method, and crop type.

2. Materials and Methods

To obtain boron fertilizer recommendations from BM3 and other soil chemical properties, the method has been divided into three phases (Figure 1); that is, to develop: (i) Algorithms predicting BM3 (database 1), (ii) agronomical response models to both crop fertilizer and soil boron content (database 2), and (iii) algorithms to predict recommended boron rates in case of deficiency (database 3).

2.1. Algorithms for Prediction of Boron Mehlich-3 (BM3)

To develop these algorithms, we used a database of 365 soil samples (database 1). These samples were taken from about 80 Québec farms dedicated to various productions: Forage crops, field crops, vegetables, and forestry (Table 1).
All samples come from the arable layer (0–20 cm) and have very different soil horizons (Dystric Brunisol, Eutric Brunisol, Melanic Brunisol, Gleysol, Humic Gleysol, Luvic Gleysol, Gray Luvisol, Humic Podzol, Humo-Ferric Podzol, Ferro-Humic Podzol, Regosol and Humic Regosol) [23]. Soil samples were air-dried and sieved with 2 mm sieves. Soil pH was determined with a 1:1 soil/solution (vol/vol). Organic carbon (OC) was determined by the Walkley–Black [24] method, and soil texture was determined by the method of Day [25].
Exchangeable bases (KM3, CaM3, MgM3) and trace elements (AlM3, FeM3, CuM3, ZnM3, BM3) were extracted according to the Mehlich-3 procedure [26] and analyzed with ICP-OES (Thermo Jarrell–Ash model 61-E, Thermo Instrument, Franklin, MA). Boron was extracted from soil using the hot water method of Berger and Truog [5], and it was quantified by colorimetry [27].
The estimation of BM3 was performed by supervised learning using the h2o’s AutoML [28] algorithm within the R software (Version 3.0.2) [29]. This algorithm allowed us to predict BM3 from BHW and a combination of the other predicting variables: pHW, OC, KM3, CaM3, and MgM3. The partial dependence of predicted BM3 on each predicting variable was determined with h2o’s varimp function, and the correlation was determined between predicted and laboratory measured BM3. Additional statistical analyses were done with Excel to determine the means, ranges, and standard deviations of the main soil sample parameters.

2.2. Boron Agronomical Models

Database 2 contains a total of 672 values of BHW and yield (Y0: Control yield without boron fertilization; Ymax: Maximal yield of the boron fertilization treatment). These data come from 32 published studies (Table 2) and involve about 40 crop types. To define an agronomic model based on BM3, we used the following predictive variables: pHW, OC, Kexchangeable, Caexchangeable and Mgexchangeable, which were all, or partially, available from the literature. This data came under two forms: Tables and graphs. Data from tables were taken directly, whereas graphical data were converted to numeric values using the Data Thief software [30].
From these 32 studies, we only kept those with BHW and those with a soil pHW from neutral to acidic. From the data of these studies, the crop nominal yield for the control was expressed as a relative yield with respect to the maximal yield.
R Y   ( % )   =   Y 0 kg · ha − 1   Y m a x kg · ha − 1   × 100
The critical agronomic thresholds of BHW or BM3 are those for which a significant yield increase will not occur with an additional amount of boron fertilizer [61]. This threshold was determined iteratively according to the Cate–Nelson method [62] using relative yields from all experiments with BHW or BM3. The Cate–Nelson procedure used identifies critical thresholds with an iterative method that maximizes the sum of squares on the boron axis [63] (BHW or BM3) while minimizing, on the yield axis, the number of points in the error quadrants [64]. Two models (RY = function of BHW and RY = function of BM3) were developed using the Cate–Nelson procedure of the rcompanion package [65] with the R software. For the model RY = function of BM3, an additional step was required because all the predictive variables (BHW, pHW, OC, Kexchangeable, Caexchangeable and Mgexchangeable) were not available for all the studies.
We therefore developed four AutoML algorithms: A1, A2, A3 and A4 (Figure 1), corresponding to the following sets of predictive variables, respectively: {BHW}, {BHW, pHW}, {BHW, pHW, OC}, {BHW, pHW, OC, Kexchangeable, Caexchangeable and Mgexchangeable}. These models were build using results from 365 samples (Database 1) representative of Québec agricultural soils (Table 1).

2.3. Boron Recommandation Algorithms

Database 3 contains 130 curves of yield response to different application rates of boron. This data come from 25 published studies (Table 3). Each of these curves were examined to determine the boron application rate corresponding to the maximum yield, RateMaximum yield. To this RateMaximum yield we associated (i) the crop type corresponding to one of the two categories of boron requirements (medium, high) as recommended by CRAAQ [4] (Centre de Référence en Agriculture et Agroalimentaire du Québec); (ii) the application type of the boron fertilizer (broadcast, sidedress, field application); and (iii) the soil variables BHW, pHW, OC, K, Ca, Mg. Prediction of BM3 was done using AutoML algorithms A2, A3 and A4 (Figure 1). The prediction of RateMaximum yield was performed by algorithm AutoML A5 using BM3, estimated by A2, A3 or A4, crop type, application mode, and soil pHW.

3. Results and Discussion

3.1. Algorithms for Prediction of Boron Mehlich-3 (BM3)

The results used by our study show a wide range of soil properties (Table 1). Soil texture varies from coarse to fine with a pH from 2.3 to 7.4. Soil boron content ranges from 0.02 to 3.68 mg B kg−1 for BHW and from 0.08 to 12.7 mg B kg−1 for BM3. This shows the Mehlich-3 method extracts more boron than hot water, which was also a conclusion of Zbíral and Němec [9] and Shuman et al. [10]. For Québec acidic sandy soils and loamy soils, Tran et al. [13] found that, on average, BM3 is 31% more than BHW. Figure 2 shows simple linear correlations between BM3 and BHW, as well as between predicted and measured BM3.
The model of Figure 2a shows a very low correlation between BHW and BM3, with only 36% of the BM3 variation being explained by the model. Low coefficients of determination (R2 = 0.29; R2 = 0.45) between BM3 and BHW were also observed by Walworth et al. [75]. With such low correlations, such a model cannot be used to transpose BHW values from the literature to BM3 values in order to develop recommendation models or to diagnose soil boron status. Figure 2b shows the prediction quality criteria of BM3 using the linear regression from Figure 2a. Compared to a perfect prediction of BM3 from BHW, i.e., a slope (m) of 1.0 and an intercept (a) of 0.0, we found that m = 0.63 and that a = 0.36 mg B kg−1, a value so high that it is close to the average of all 365 samples analyzed (0.40 mg B kg−1). Such a weak model does not allow for a reliable conversion from BHW to BM3. Therefore, instead of simple linear regression, we used the supervised learning algorithm AutoML (Table 4).
The A1 AutoML algorithm predicted BM3 from BHW with an R2 of 0.63 compared to 0.36 for the linear regression, but intercepts and slopes were similar. Adding predictive variables related to the BM3 extraction improved the predictive capacity (Table 4). The algorithm A4 performed better with an R2 and a slope close to one, as well as an intercept close to 0 (Figure 3a).
To improve the conversion from BHW to BM3, we included in AutoML A4 five soil parameters that are routinely analyzed by soil laboratories. With all these parameters, a multiple linear regression gives R2 = 0.65, a = 0.23 and m = 0.65 (Figure 3i), which results in a deviation of 35% (1-m) between predicted and measured BM3. The improvement in BM3 prediction when going from simple regression (Figure 2b) to multiple regression (Figure 3i) and to AutoML shows the potential of using supervised learning algorithm A4. The A4 AutoML algorithm shows only a slight average deviation of 3%, which is much less than the value of 20% for inter-laboratory error allowed by the CEAEQ [76] (Centre d’expertise en analyse environnementale du Québec) and an intercept close to the detection limit of 0.01 mg kg−1 [77]. It is to be noted that all slopes, either from regression or supervised learning, are less than 1.0, which implies mainly an under-estimation of BM3.
Algorithms A3 and A4 predicted BM3 with deviations (1-m) less than the 20% error allowed by CEAEQ [76]. These two algorithms allow for a better prediction of BM3 by using several soil variables from which depend the soil boron status. Recent research used machine learning to obtain reliable predictions [78,79,80]. Since in Québec routine soil determination includes elements extractible by Mehlich-3, the pH, and soil organic matter [79], algorithm A4 is the most appropriate to convert BHW to BM3. The scaled (0–1) importance of the variables used to predict BM3 by A4 is shown in Figure 3b; they are, in order of importance: CaM3 > pHW > BHW > MgM3 > OC > KM3. The influence intervals of each of these predictors are shown in Figure 3c–h. Generally, the value of predicted BM3 increased, up to a limit, with the values of the predictors. Contrarily to what was expected, BHW is not the most important variable in predicting BM3. The two most important variables are CaM3 and pHW, which is explained by the composition of the Mehlich-3 solution.
In comparison with the official method of ammonium acetate pH 7, the Mehlich-3 pH 3 method extracts approximately the same levels of K and Mg. However, Mehlich-3 extracts more calcium than ammonium acetate, i.e., 17% [80] or 28% [26]. This additional extraction could influence the amount of boron moving from the soil to the Mehlich-3 solution. Besides, this is the most plausible reason to explain that BM3 is higher than BHW [13]. Mehlich [26] included, in his extraction solution, ammonium fluoride (NH4F) which induces an interaction between soil pH and that of the solution. Mehlich [26] concluded that soils with a high pH are at risk of calcium precipitation in the form of CaF2, which makes the Mehlich-3 solution not relevant for alkaline soils (pHW ≥ 7.5). This is seen in Figure 3c, where the BM3 is dependent of the exchangeable calcium up to a concentration of about 13,000 mg CaM3 kg−1. A higher pHW results in less exchangeable calcium and in an increase of anionic boron in the form of B(OH)4− by the hydrolytic effect [81]. Cancela et al. [82] found that the Mehlich-3 extraction is more reliable for soils with a low calcium content and which are more acidic. In high pH soils, Tran et al. [13] measured BM3 values 2–5.2 more than BHW; in addition, the correlation between these two variables was almost zero. Moreover, in Figure 3d, we can observe that BM3 is more affected by pH values superior to 6. For organic carbon, KM3 and MgM3 (Figure 3f–g) their predictive influences on BM3 are within the ranges usually found in Québec mineral soils, i.e., 0%–10% for OC, 0–500 mg B kg−1 for KM3, and 0–1800 mg B kg−1 for MgM3.

3.2. Boron Agronomic Models

Two models were developed: One based on BHW (Figure 4) and another based on predicted BM3 (Figure 5). The Cate–Nelson procedure applied to these two models allowed for the definition of two critical thresholds: 0.23 and 0.58 mg BHW kg−1 (Figure 4c) and 0.65 and 1.03 mg BM3 kg−1 (Figure 5c). These thresholds correspond to high points of the curve between the sum of squares and the variable concerned [62]. These critical values allowed us to partition soil boron into three fertility classes for diagnostics: Low, medium and high. For BHW, these three classes correspond, respectively, to 0–0.23, 0.23–0.58, and 0.58–3.70 mg BHW kg−1, whereas for BM3 they are 0–0.65, 0.65–1.03, and 1.03–12.70 mg BM3 kg−1. The high probabilities of economical response for crops to boron fertilizers are in the true positive quadrants (TP) of low boron contents, i.e., less than 0.23 mg BHW kg−1 or less than 0.65 mg BM3 kg−1, which correspond to relative yield from 20 to 85% and from 20 to 95%, respectively. The high limits of 85% and 95% for critical relative yield were obtained by minimizing the number of points in the error quadrants: false negative(FN) + false positive (FP) of Figure 4b and Figure 5b for BHW and BM3, respectively. Contrarily, stable yields are observed in the true negative quadrants (TN) corresponding to low probabilities that yields respond to boron fertilization. In these TN quadrants, relative yield values reach a stability [64] ranging from 85% to 100% for BHW and from 95% to 100% for BM3 (Figure 4a and Figure 5a). Error quadrants (FP) correspond to high relative yields despite pertaining to a low boron fertility class, whereas FN quadrants correspond to low relative yield while being in a high boron fertility class. For Figure 4 and Figure 5, we obtained a robustness R2 of 82.6 and 69.1%, respectively, corresponding to the probability to make a correct diagnosis for soil boron. For both cases, model robustness was high. These R2 values are defined as the ratio the number of points in quadrants TP and TN to the total number of points. The specificity TN/(TN + FP) represents the probability to make the correct decision (do not fertilize) with respect to all observations with a yield stability, i.e., a relative yield >85% or >95%, respectively, for the models of Figure 4 and Figure 5. These specificities were 83.8% and 65.1% for the BHW model (Figure 4) and the BM3 model (Figure 5), respectively.
These values of 83.8% and 65.1% represent the probability that high relative yield be associated to soil with a boron content higher to the agronomical critical threshold of 0.23 mg kg−1 for BHW and 0.65 mg kg−1 for BM3. The sensitivity [TP/(TP + FN)] represents the probability to make a good decision to fertilize with boron with respect to all observations with less relative yields, i.e., a relative yield <85% for the models in Figure 4 or <95% for the models in Figure 5. These sensitivities were 78.7% for the BHW model and 73.5% for the BM3 model. In these cases, the sensitivity is the probability that lower yields occur for soils with a boron content less than the critical agronomical threshold of 0.23 mg kg−1 for BHW and 0.65 mg kg−1 for BM3. Positive predictive values (PPV) [TP/(TP+FP)] are the probability of a positive response of yield to boron fertilization when soil boron content is less than the critical agronomic threshold of 0.23 mg kg−1 for BHW and 0.65 mg kg−1 for BM3. These PPVs are 59.2% for the BHW model (Figure 4) and 65.7% for the BM3 model (Figure 5).
The negative predictive values (NPV) [TN/(TN + FN)] are the probability that crops do not respond to boron fertilization when the soil boron content is more than the critical agronomical threshold. These NPVs are 92.9% for the BHW model (Figure 4) and 73.0% for the BM3 model (Figure 5). As can be seen from Figure 4 and Figure 5, the distribution patterns of the points are similar, with small statistical differences from the Cate–Nelson partitioning. The BHW agronomical model shows better statistics (robustness, specificity, sensitivity, PPV, NPV) than the BM3 agronomical model. The better performance of the BHW model is explained by the fact that it is developed from measured BHW values, whereas the BM3 model is based on the results of four algorithms with a predictive power ranging from 65% to 99%. Moreover, the critical relative yield value of 95% for the BM3 model is more limiting than the value of 85% for the BHW model; this causes the BM3 model having more points in the error quadrants.
Since the BM3 method extracts more boron from the soil, the critical agronomical thresholds are higher for BM3 than for BHW. Besides, Datta et al. [83] found, for a same culture and for identical experimental conditions, different critical thresholds with four boron extraction methods. The two critical thresholds that we found (0.65 and 1.03 mg BM3 kg−1) are located in the median part of the interval [0.1 to 2.0 mg BHW kg−1] [84], below which soil boron fertility is classified as low to medium [85].

3.3. Boron Recommendation Model

Usually, a recommendation model for a given element is based solely on the soil content of that element, which implies that yield depends on that element only and that any other fertility indicators or fertilization practices do not have any influence. However, Simard et al. [86] proposed considering both pHW and BM3 for boron recommendations. Additionally, Gupta and Cutcliffe [74] showed the effect of boron spreading type on crop yield. It is also known that boron requirements vary from crop to crop [4]. Therefore, the AutoML A5 algorithm, combining all these factors to predict the boron dose, gave good results, as shown in Figure 6.
This figure shows a correlation between observed and predicted RateMaximum yield, giving an R2 of 0.63, an intercept of 0.6 kg B ha−1, and a slope of 0.7. Therefore, 63% of the RateMaximum yield variance is explained by the model. Usually, recommendation models presented in the literature have very low R2 values. For a same soil analysis, these models present so much variation that researchers use averages or mathematical expectations to propose recommended rates [87]. The A5 algorithm underestimates RateMaximum yield, as it shows a mean deviation of 30% between observed and predicted RateMaximum yield. In addition to the main variable, BM3, the inclusion of three other predicting variables (pHW, application type and crop boron requirements) in the A5 algorithm significantly improves the boron rate prediction. This improvement is explained by the interaction between BM3 and the three other variables. As shown in Figure 6b, pHW is the second most important predicting variable after BM3. In Figure 6c, the least influencing interval of BM3 is located between 0.65 and 1.03 mg BM3 kg−1, which coincides with the critical agronomical values of the model of Figure 5a. This interval is between the region of boron deficiency (BM3 < 0.65 mg kg−1) and that of boron sufficiency (BM3 > 1.03 mg kg−1). It is within this intermediary interval that we find the most estimation error on the boron recommended dose. Besides, in the proposed agronomical model (Figure 5), the points in the error quadrant FN are mostly concentrated in this interval. Therefore, the boron recommendation is relatively less reliable within that interval than outside of it. Figure 6d shows that for a pHW above 6.0, the required boron amount to reach maximum yield is increased.
This figure also shows that for a range of 3 pHW units (from 4.5 to 7.5), the interval of recommended application rate is from 1 to 2.5 kg ha−1. This is in concordance with findings of Peterson and Newman [88], who recommend a boron rate of 2.5 higher for a soil of pHW of 7.4 compared to another with a pHW of 4.7. These observations show the fixation of boron added to the soil increase considerably when the soil acidity is neutralized, which implies the application of higher boron rates. While predicting variables BM3 and pHW are numerical, the two others (spreading type and crop boron requirement) are categorical; however, they also have an effect on boron recommended rates. The 130 crop response curves analyzed (Database 3) show that RateMaximum yield decreases for the following order of application type: Broadcast > sidedress > foliar application. Such a difference due to the application method is explained by the fact that fertilizer losses are greater for broadcasting than for sidedress and for foliar application [89]. With the proposed A5 multi-variable model, boron fertilizer applications can be done with more precision. The precision of this A5 model could be further increased as more data become available.

4. Summary and Conclusions

In this study, we used three databases to develop a multi-variable system for soil boron diagnostics and recommendation. Database 1 is composed of 365 soil samples representative of the various Québec agricultural regions. This database was used to develop an AutoML algorithm (A4) which successfully converts BHW to BM3 using five covariates (pHW, OC, CaM3, MgM3 and KM3). This A4 algorithm allows for a robust prediction of 99% (with only a slight 3% average deviation) and an intercept so small that it is in the same order of magnitude as the detection limit of 0.01 mg kg−1 of BM3. Database 2 is composed of 672 observations coming completely from published literature and was used to develop two agronomical models, one for BHW and one for BM3. These two models showed critical thresholds defining the intervals of medium deficiency, that is 0.23 and 0.58 mg kg−1 for BHW and 0.65 and 1.03 mg kg−1 for BM3. These two intervals discriminate soils with a risk of boron deficiency from those with high potential of boron sufficiency. Database 3 include 130 crop yield response to various boron rates; all data were extracted from the literature. These response curves allowed us to define 130 boron rates corresponding to maximal yields (RateMaximum yield). An AutoML supervised learning algorithm, A5, allowed us to relate RateMaximum yield to BM3, pHW, fertilizer application type, and boron crop requirement category. With an R2 of 63%, the performance of algorithm A5 is much greater than those of other recommendation models based solely on soil reserves. Results from this study show the superiority of supervised learning compared to traditional prediction methods, despite the heterogeneity of sources and data. It therefore becomes possible to include boron in the common extractant of Mehlich-3 to diagnose and recommend boron fertilization, especially because this method is well suited to the neutral to acidic soils in Québec.

Author Contributions

D.R., K.L. and G.J. planned and designed the research; D.R. carried out the experiments; D.R. and K.L. wrote the manuscript; J.C. was involved in structuring the research.

Funding

This research did not receive any external funding.

Acknowledgments

We are grateful to Ministry of Agriculture Fisheries and Food of Quebec (MAPAQ) for the access to their soil database.

Conflicts of Interest

The authors declare that there is no conflict of interest.

References

  1. Du, C.; Zhou, J. Evaluation of soil fertility using infrared spectroscopy: A review. Environ. Chem. Lett. 2009, 7, 97–113. [Google Scholar] [CrossRef]
  2. Gupta, U.C.; Wu, K.; Liang, S. Micronutrients in soils, crops, and livestock. Earth Sci. Front. 2009, 15, 110–125. [Google Scholar] [CrossRef]
  3. Lindsay, W.L.; Cox, F.R. Micronutrient soil testing for the tropics. Fertil. Res. 1985, 7, 169–199. [Google Scholar] [CrossRef]
  4. Parent, L.E.; Gagné, G. CRAAQ Guide de Référence en Fertilisation, 2nd ed.; Centre de Référence en Agriculture et Agroalimentaire du Québec: Québec, QC, Canada, 2010; p. 473. [Google Scholar]
  5. Berger, K.C.; Truog, E. Boron Determination in Soils and Plants Using the Quinalizarin Reaction. Ind. Eng. Chem. Anal. Ed. 1939, 11, 540–545. [Google Scholar] [CrossRef]
  6. Soltanpour, W.D.; Gestring, P.N. Evaluation of the ammonium bicarbonate-DTPA soil test for assessing boron availability to alfalfa. Soil Sci. Soc. Am. J. 1984, 48, 96–100. [Google Scholar]
  7. Offiah, O.; Axley, J.H. Improvement of boron soil test. Commun. Soil Sci. Plant Anal. 1988, 19, 1527–1542. [Google Scholar] [CrossRef]
  8. Wei, Y.; Zarcinas, B.A. An improved procedure for the extraction of plant available soil boron. In Boron in Soils and Plants; Springer: Dordrecht, The Netherlands, 1997; pp. 77–81. [Google Scholar]
  9. Zbíral, J.; Němec, P. Comparison of some soil extractants for determination of boron. Commun. Soil Sci. Plant Anal. 2009, 40, 96–105. [Google Scholar] [CrossRef]
  10. Shuman, L.M.; Bandel, V.A.; Donohue, S.J.; Isaac, R.A.; Lippert, R.M.; Sims, J.T.; Tucker, M.R. Comparison of mehlich-1 and mehlich-3 extractable soil boron with hot-water extractable boron. Commun. Soil Sci. Plant Anal. 1992, 23, 1–14. [Google Scholar] [CrossRef]
  11. Şimşek, A.; Korkmaz, D.; Velioğlu, Y.S.; Ataman, O.Y. Determination of boron in hazelnut (Corylus avellana L.) varieties by inductively coupled plasma optical emission spectrometry and spectrophotometry. Food Chem. 2003, 83, 293–296. [Google Scholar] [CrossRef]
  12. Redd, S.A.; Shiffler, A.K.; Jolley, V.D.; Webb, B.L.; Haby, V.A.; Shiffler, A.K.; Jolley, V.D.; Webb, B.L.; Haby, V.A. Mehlich 3 Extraction of Boron in Boron Treated Soils as Compared to Other Extractants. Commun. Soil Sci. Plant Anal. 2008, 39, 1245–1259. [Google Scholar] [CrossRef]
  13. Tran, T.S.; Giroux, M.; Audesse, P.; Guilbault, J. Importance des elements mineurs en agriculture. Agrosol 1995, 8, 12–22. [Google Scholar]
  14. Gupta, U.C.; Cutcliffe, J.A. Boron nutrition of broccoli, brussel sprouts, and cauliflower grown on Prince Edward Island soils. Can. J. Soil Sci. 1973, 53, 275–279. [Google Scholar] [CrossRef]
  15. Gupta, U.C.; Cutcliffe, J.A. Boron deficiency in cole crops under field and greenhouse conditions. Commun. Soil Sci. Plant Anal. 1975, 6, 181–188. [Google Scholar] [CrossRef]
  16. Rashid, A.; Rafique, E.; Bughio, N. Diagnosing boron deficiency in rapeseed and mustard by plant analysis and soil testing. Commun. Soil Sci. Plant Anal. 1994, 25, 2883–2897. [Google Scholar] [CrossRef]
  17. Mahler, R.L.; Shafii, B. Relationship between soil-test boron and lentil yields in the inland Pacific Northwest. Commun. Soil Sci. Plant Anal. 2007, 38, 1193–1202. [Google Scholar] [CrossRef]
  18. Mahler, R.L.; Shafii, B. Relationship between soil test boron and pea yields in the inland Pacific Northwest. Commun. Soil Sci. Plant Anal. 2009, 40, 2603–2615. [Google Scholar] [CrossRef]
  19. Loué, A. Micronutrients in Agriculture [Oligo-Éléments en Agriculture]; SCPA-NATHAN: Antibes, France, 1993. [Google Scholar]
  20. Mohib, R.; Naz, M.; Muhammad, S.; Hamid, A.; Bibi, F. Effect of boron on the flowering and fruiting of tomato. Sarhad J. Agric. 2012, 28, 37–40. [Google Scholar]
  21. Moniruzzaman, M.; Rahman, S.M.L.; Kibria, M.G.; Rahman, M.A.; Hossain, M. Effect of boron and nitrogen on yield and hollowstem of broccoli. J. Soil Nat. 2007, 1, 24–29. [Google Scholar]
  22. Nyomora, A.M.S.; Brown, P.H.; Krueger, B. Rate and time of boron application increase almond productivity and tissue boron concentration. HortScience 1999, 34, 242–245. [Google Scholar] [CrossRef]
  23. Agriculture Canada Expert Committee on Soil Survey. The Canadian System of Soil Classification, 2nd ed.; Agriculture Canada Publishing: Ottawa, ON, Canada, 1987. [Google Scholar]
  24. Nelson, D.W.; Sommers, L.E. Total carbon, organic carbon, and organic matter. In Methods of Soil Analysis, 2nd ed.; Page, A.L., Ed.; American Society of Agronomy: Madison, WI, USA, 1982. [Google Scholar]
  25. Day, P.R. Particle fractionation and particle size analysis. In Methods of Soil Analysis; Black, C.A., Ed.; American Society of Agronomy: Madison, WI, USA, 1965; Volume 1. [Google Scholar]
  26. Mehlich, A. Mehlich 3 soil test extractant: A modification of Mehlich 2 extractant. Commun. Soil Sci. Plant Anal. 1984, 15, 1409–1416. [Google Scholar] [CrossRef]
  27. John, M.K.; Chuah, H.H.; Neufeid, H.J. Application of improved azomethine-H-method to the determination of boron in soil and plants. Anal. Lett. 1975, 8, 559–568. [Google Scholar] [CrossRef]
  28. LeDell, E.; Gill, N.; Aiello, S.; Fu, A.; Candel, A.; Click, C.; Kraljevic, T.; Nykodym, T.; Aboyoun, P.; Kurka, M.; et al. Package ‘h2o’: R Interface for ‘H2O’. 2019. Available online: https://github.com/h2oai/h2o-3 (accessed on 20 July 2019).
  29. R Development Core Team. R: A Language and Environment for Statistical Computing; Foundation for Statistical Computing: Vienna, Austria, 2016. [Google Scholar]
  30. Tummers, B. Data Thief III Manual V 1.1. 2006, p. 52. Available online: https://datathief.org/ (accessed on 20 July 2019).
  31. Bagchi, D.K.; Banerjee, A.; Bhattacharya, B. Effect of nitrogen, boron and molybdenum on yield of root and leaf protein of radish (Raphanus sativus). Indian J. Agric. Sci. 1988, 58, 224–226. [Google Scholar]
  32. Cifu, M.; Zhihong, C.; Peikuen, J.; Guomo, Z.; XianGui, L.; Qiufang, X. Effects of application of boron on growth, yields, and quality of red bayberry. J. Plant Nutr. 2007, 30, 1047–1058. [Google Scholar] [CrossRef]
  33. Cirak, C.; Odabas, M.S.; Kevseroglu, K.; Karaca, E.; Gulumser, A. Response of soybean (Glycine max) to soil and foliar applied boron at different rates. Indian J. Agric. Sci. 2006, 76, 603–606. [Google Scholar]
  34. Malhi, S.S.; Raza, M.; Schoenau, J.J.; Mermut, A.R.; Kutcher, R.; Johnston, A.M.; Gill, K.S. Feasibility of boron fertilization for yield, seed quality and B uptake of canola in northeastern Saskatchewan. Can. J. Soil Sci. 2002, 83, 99–108. [Google Scholar] [CrossRef]
  35. Cutcliffe, J.A. Effects of boron, molybdenum and lime on yield and leaf tissue nutrient concentration of green peas. Can. J. Plant Sci. 1986, 66, 971–976. [Google Scholar] [CrossRef]
  36. Ganie, M.A.; Farida Akhter, M.A.B.; Najar, G.R. Growth, yield and quality of french bean (PHASEOLUS VULGARIS L.) as influenced by sulphur and boron application on inceptisols of Kashmir. Bioscan 2014, 9, 513–518. [Google Scholar]
  37. Dunn, D.; Stevens, G.; Kending, A. Boron fertilization of rice with soil and foliar applications. Crop Manag. 2005, 1, 1–10. [Google Scholar]
  38. Oyewole, O.I.; Aduayi, E.A. Evaluation of the growth and quality of the “ife plum” tomato as affected by boron and calcium fertilization. J. Plant Nutr. 1992, 15, 199–209. [Google Scholar] [CrossRef]
  39. Debnath, M.R.; Jahiruddin, M.; Rahman, M.M.; Haque, M.A. Determining optimum rate of boron application for higher yield of wheat in Old Brahmaputra Floodplain soil. J. Bangladesh Agric. Univ. 2011, 9, 205–210. [Google Scholar] [CrossRef]
  40. Rerkasem, B.; Netsangtip, R.; Bell, R.W.; Loneragan, J.F.; Hiranburana, N. Comparative species responses to boron on a Typic Tropaqualf in Northern Thailand. Plant Soil 1988, 106, 15–21. [Google Scholar] [CrossRef]
  41. Devi, K.N.; Singh, L.N.K.; Sumarjit Singh, M.; Basanta Singh, S.; Khamba Singh, K. Influence of Sulphur and Boron Fertilization on Yield, Quality, Nutrient Uptake and Economics of Soybean (Glycine max) under Upland Conditions. J. Agric. Sci. 2012, 4, 1–10. [Google Scholar] [CrossRef]
  42. Sharma, K.R.; Srivastava, P.C.; Ghosh, D.; Gangwar, M.S. Effect of boron and farmyard manure application on growth, yields, and boron nutrition of sunflower. J. Plant Nutr. 1999, 22, 633–640. [Google Scholar] [CrossRef]
  43. Dursun, A.; Turan, M.; Ekinci, M.; Gunes, A.; Esringü, A.; Yildirim, E. Effects of boron fertilizer on tomato, pepper, and cucumber yields and chemical composition. Commun. Soil Sci. Plant Anal. 2010, 41, 1576–1593. [Google Scholar] [CrossRef]
  44. Sherrell, C.G. Boron deficiency and response in white and red clovers and lucerne. New Zeal. J. Agric. Res. 1983, 26, 197–203. [Google Scholar] [CrossRef]
  45. Gupta, U.C. Effects of boron and limestone on cereal yields and on B and N concentrations of plant tissue. Pant Soil 1977, 47, 283–287. [Google Scholar] [CrossRef]
  46. Sherrell, C.G.; Toxopeus, M.R.J. Effects of boron applications on yield and boron concentration of lucerne (medicago sativa l.) grown on yellow-brown pumice soils. New Zeal. J. Exp. Agric. 1978, 6, 145–150. [Google Scholar] [CrossRef]
  47. Gupta, U.C. Boron nutrition of alfalfa, red clover, and timothy grown on podzol soils of eastern Canada. Soil Sci. 1984, 137, 16–22. [Google Scholar] [CrossRef]
  48. Singh, A.K.; Singh, S. Response of soybean to phosphorus and boron fertilization in acidic upland soil of Nagaland. J. Indian Soc. Soil Sci. 2012, 60, 3–6. [Google Scholar]
  49. Turan, M.; Ketterings, Q.M.; Gunes, A.; Ataoglu, N.; Esringü, A.; Bilgili, A.V.; Huang, M.Y. Boron fertilization of Mediteranean aridisols improves lucerne (Medicago Sativa L.) yields and quality. Soil Plant Sci. 2009, 60, 427–436. [Google Scholar]
  50. Gupta, U.C.; Macleod, J.A. Boron nutrition and growth of timothy as affected by soil pH. Commun. Soil Sci. Plant Anal. 1973, 4, 389–395. [Google Scholar] [CrossRef]
  51. Valenciano, J.B.; Boto, J.A.; Marcelo, V. Response of chickpea (Cicer arietinum L.) yield to zinc, boron and molybdenum application under pot conditions. Span. J. Agric. Res. 2010, 8, 797–807. [Google Scholar] [CrossRef]
  52. Hussain, M.J.; Sarker, M.M.R.; Sarker, M.H.; Ali, M.; Salim, M.M.R. Effect of different levels of boron on the yield and yield attributes of mustard in surma-kushiara flood plain soil (AEZ 20). J. Soil Nat. 2008, 2, 6–9. [Google Scholar]
  53. Wojcik, P.; Wojcik, M. Effects of boron fertilization on ‘Conference’ pear tree vigor, nutrition, and fruit yield and storability. Plant Soil 2003, 256, 413–421. [Google Scholar] [CrossRef]
  54. Davis, J.M.; Sanders, D.C.; Nelson, P.V.; Lengnick, L.; Sperry, W.J. Boron improves growth, yield, quality, and nutrient content of tomato. J. Am. Soc. Hortic. Sci. 2003, 128, 441–446. [Google Scholar] [CrossRef]
  55. Wojcik, P. Response of black currant to boron fertilization. J. Plant Nutr. 2005, 28, 63–72. [Google Scholar] [CrossRef]
  56. Wojcik, P. Response of ‘bluecrop’ highbush blueberry to boron fertilization. J. Plant Nutr. 2005, 28, 1897–1906. [Google Scholar] [CrossRef]
  57. Razmjoo, K.; Henderlong, P.R. Effect of potassium, sulfur, boron, and molybdenum fertilization on alfalfa production and herbage macronutrient contents. J. Plant Nutr. 1997, 20, 1681–1696. [Google Scholar] [CrossRef]
  58. Yu, X.; Bell, P.F. Boron and lime effects on yield and deficiency symptoms of rice grown in greenhouse on acid typic glossaqualf. J. Plant Nutr. 2002, 25, 2591–2602. [Google Scholar] [CrossRef]
  59. Lombin, G. Evaluation of the Micronutrient Fertility of Nigeria’s Semi-Arid Savannah Soils: Boron and Molybdenum. Soil Sci. Plant Nutr. 1985, 31, 13–25. [Google Scholar] [CrossRef]
  60. Zhang, Y.; Hu, C.; Tan, Q.; Nie, Z.; Zheng, C.; Gui, H.; Sun, X.; Zhao, X. Soil application of boron and zinc influence fruit yield and quality of Satsuma mandarin in acidic soils. Agron. J. 2015, 107, 1–8. [Google Scholar] [CrossRef]
  61. Zamuner, E.; Lloveras, J.; Echeverría, H. Can Agronomic Phosphorus Recommendations for Potato be Environmentally Sustainable? Better Crop. Plant Food 2017, 101, 20–22. [Google Scholar]
  62. Nelson, L.A.; Anderson, R.L. Soil testing: Correlating and interpreting the analytical results. In Partitioning of Soil Test-Crop Response Probability; Stelly, M., Ed.; ASA Spec. Publ.: Madison, WI, USA, 1984; pp. 19–38. [Google Scholar]
  63. Cate, R.B.; Nelson, L.A. A Simple Statistical Procedure for Partitioning Soil Test Correlation Data Into Two Classes. Soil Sci. Soc. Am. J. 1971, 35, 658–660. [Google Scholar] [CrossRef]
  64. Beegle, D. Interpretation of soil testing results. In Recommended Soil Testing Procedures for the Northeastern United States; Sims, J.T., Wolf, A.M., Eds.; Agricultural Experiment Stations: Newark, DE, USA, 1995; pp. 111–118. [Google Scholar]
  65. Mangiafico, S. Package R ‘Rcompanion’ Version 2.2.1. 2019. Available online: http://rcompanion.org/ (accessed on 20 July 2019).
  66. Gupta, U.C.; Cutcliffe, J.A. Boron nutrition of carrots and table beets grown in a boron deficient soil. Commun. Soil Sci. Plant Anal. 1985, 16, 509–516. [Google Scholar] [CrossRef]
  67. Haque, M.E.; Paul, A.K.; Sarker, J.R. Effect of Nitrogen and Boron on the Growth and Yield of Tomato (Lycopersicon esculentum M.). Int. J. Bio. Ressour. Stress Manag. 2011, 2, 277–282. [Google Scholar]
  68. Maurya, A.N.; Rai, K.N.; Lal, S. Effects of boron and nitrogen fertilizers on radish (RAPHANUS SATIVUS). Exp. Agric. 1977, 13, 301–303. [Google Scholar] [CrossRef]
  69. Freeborn, J.R.; Holshouser, D.L.; Alley, M.M.; Powell, N.L.; Orcutt, D.M. Soybean yield response to reproductive stage soil-applied nitrogen and foliar-applied boron. Agron. J. 2001, 93, 1200–1209. [Google Scholar] [CrossRef]
  70. Quddus, M.A.; Nasser, H.M.; Hossain, M.A.; Abul Hossain, M. Effect of zinc and boron on yield and yield contributing characters of lentil in low ganges river floodplain soil at madaripur, Bangladesh. Bangladesh J. Agric. Res. 2014, 39, 591–603. [Google Scholar] [CrossRef]
  71. Gupta, U.C.; Cutcliffe, J.A. Effects of applied and residual boron on the nutrition of cabbage and fields beans. Can. J. Soil Sci. 1984, 64, 571–576. [Google Scholar] [CrossRef]
  72. Gupta, U.C.; Cutcliffe, J.A. Determination of optimum levels of boron in rutabaga leaf tissue and soil. Soil Sci. 1971, 111, 382–385. [Google Scholar] [CrossRef]
  73. Srivastava, S.P.; Bhandari, T.M.S.; Yadav, C.R.; Joshi, M.; Erskine, W. Boron deficiency in lentil: Yield loss and geographic distribution in a germplasm collection. Plant Soil 2000, 219, 147–151. [Google Scholar] [CrossRef]
  74. Gupta, U.C.; Cutcliffe, J.A. Effects of methods of boron application on leaf tissue concentration of boron and control of brown-heart in rutabaga. Can. J. Plant Sci. 1978, 58, 63–68. [Google Scholar] [CrossRef]
  75. Walworth, J.L.; Panciera, M.T.; Gavlak, R.G. Mehlich 3 extractant for determination of available B, Cu, Fe, Mn, and Zn in cryic Alaskan soils. Can. J. Soil Sci. 1992, 72, 517–526. [Google Scholar] [CrossRef] [Green Version]
  76. Centre d’Expertise en Analyse Environnementale du Québec. Critères de Variation Relatifs; Ministère du Développement durable, de l’Environnement et des Parcs du Québec: Québec, QC, Canada, 2015; Volume 24. [Google Scholar]
  77. Centre d’Expertise en Analyse Environnementale du Québec. Détermination des métaux, des oligoéléments, du phosphore et du soufre par les méthodes d’extraction Mehlich-III, extraction du bore à l’eau chaude et dosage par spectrométrie d’émission au plasma d’argon; Ministère de l’Environnement du Québec: Québec, QC, Canada, 2003. [Google Scholar]
  78. Brungard, C.W.; Boettinger, J.L.; Duniway, M.C.; Wills, S.A.; Edwards, T.C. Machine learning for predicting soil classes in three semi-arid landscapes. Geoderma 2015, 239, 68–83. [Google Scholar] [CrossRef]
  79. Morellos, A.; Pantazi, X.E.; Moshou, D.; Alexandridis, T.; Whetton, R.; Tziotzios, G.; Wiebensohn, J.; Bill, R.; Mouazen, A.M. Machine learning based prediction of soil total nitrogen, organic carbon and moisture content by using VIS-NIR spectroscopy. Biosyst. Eng. 2016, 152, 104–116. [Google Scholar] [CrossRef] [Green Version]
  80. Pantazi, X.E.; Moshou, D.; Alexandridis, T.; Whetton, R.L.; Mouazen, A.M. Wheat yield prediction using machine learning and advanced sensing techniques. Comput. Electron. Agric. 2016, 121, 57–65. [Google Scholar] [CrossRef]
  81. Kaplan, D.I.; Adriano, D.C.; Mills, G.L.; Sajwan, K.S.; Burkman, W.G. Determination of boron in soils containing inorganic and organic boron sources. Soil Sci. Soc. Am. J. 1990, 54, 708–714. [Google Scholar] [CrossRef]
  82. Cancela, R.C.; De Abreu, C.A.; Paz-González, A. DTPA and Mehlich-3 micronutrient extractability in natural soils. Commun. Soil Sci. Plant Anal. 2002, 33, 2879–2893. [Google Scholar] [CrossRef]
  83. Datta, S.P.; Bhadoria, P.B.S.; Kar, S. Availability of extractable boron in some acid soils, West Bengal, India. Commun. Soil Sci. Plant Anal. 1998, 29, 2285–2306. [Google Scholar] [CrossRef]
  84. Sims, J.T.; Johnson, G.V. Micronutrients soil tests. In Micronutrients in Agriculture; Mortvedt, J.J., Ed.; Soil Science Society of America: Madison, WI, USA, 1991; pp. 427–476. [Google Scholar]
  85. Korndörfer, G.H.; Snyder, G.H.; Ulloa, M.; Powell, G.; Datnoff, L.E. Calibration of soil and plant silicon analysis for rice production. J. Plant Nutr. 2001, 24, 1071–1084. [Google Scholar] [CrossRef]
  86. Simard, R.R.; Charron, G.; Pageau, D. Field calibration of boron soil tests for barley. Commun. Soil Sci. Plant Anal. 1996, 27, 1631–1646. [Google Scholar] [CrossRef]
  87. Khiari, L.; Parent, L.E.; Pellerin, A.; Alimi, A.R.A.; Tremblay, C.; Simard, R.R.; Fortin, J. An agri-environmental phosphorus saturation index for acid coarse-textured soils. J. Environ. Qual. 2000, 29, 1561–1567. [Google Scholar] [CrossRef]
  88. Peterson, L.A.; Newman, R.C. Influence of soil pH on the availability of added boron1. Soil Sci. Soc. Am. J. 1976, 40, 280–282. [Google Scholar] [CrossRef]
  89. Gupta, U.C. Essential elements-micronutrients. In Hanbook of Plant Nutrition; Barker, A.V., Pilbeam, D.J., Eds.; CRC Press: Boca Raton, FL, USA, 2007; pp. 241–277. [Google Scholar]
Agronomy 09 00410 g001 550
Figure 1. Summary of the different steps in the development of the models for the diagnosis and recommendation of boron in acidic to neutral soils. BM3: Boron Mehlich-3; BHW: Boron hot water; pHW: pH water; OC: Organic carbon; KM3, MgM3, and CaM3 are, respectively, Mehlich-3 extracted potassium, magnesium and calcium; MAPAQ: Ministry of Agriculture Fisheries and Food of Quebec.
Figure 1. Summary of the different steps in the development of the models for the diagnosis and recommendation of boron in acidic to neutral soils. BM3: Boron Mehlich-3; BHW: Boron hot water; pHW: pH water; OC: Organic carbon; KM3, MgM3, and CaM3 are, respectively, Mehlich-3 extracted potassium, magnesium and calcium; MAPAQ: Ministry of Agriculture Fisheries and Food of Quebec.
Agronomy 09 00410 g001
Agronomy 09 00410 g002 550
Figure 2. (a) Linear regression between measured boron extracted by hot water (BHW) and measured boron extracted by the Mehlich-3 solution (BM3); (b) Linear regression between measured BM3 and predicted BM3 (by the linear model of Figure 2a).
Figure 2. (a) Linear regression between measured boron extracted by hot water (BHW) and measured boron extracted by the Mehlich-3 solution (BM3); (b) Linear regression between measured BM3 and predicted BM3 (by the linear model of Figure 2a).
Agronomy 09 00410 g002
Agronomy 09 00410 g003 550
Figure 3. Boron Mehlich-3 (BM3) prediction and statistical analysis for algorithm AutoML A4 (a–h) and for multiple regression (i). (a) Correlation between measured and predicted BM3; (b) scaled importance of predictors (CaM3, pHW, MgM3, KM3, BHW, OC); (c–h) partial dependence on CaM3, pHW MgM3, KM3, BHW, OC, respectively; (i) correlation between measured and predicted BM3 by multiple regression.
Figure 3. Boron Mehlich-3 (BM3) prediction and statistical analysis for algorithm AutoML A4 (a–h) and for multiple regression (i). (a) Correlation between measured and predicted BM3; (b) scaled importance of predictors (CaM3, pHW, MgM3, KM3, BHW, OC); (c–h) partial dependence on CaM3, pHW MgM3, KM3, BHW, OC, respectively; (i) correlation between measured and predicted BM3 by multiple regression.
Agronomy 09 00410 g003
Agronomy 09 00410 g004 550
Figure 4. Construction of the diagnostic and statistical model of the Cate–Nelson classification for hot water boron showing the (a) Cate–Nelson graph for the critical thresholds identified; (b) number of points outside the model for determination of the critical relative yield (RY); (c) sum of squares for determination al of the critical BHW; and summary table (lower right)n: Number of points in the different quadrants; Performance indicators of the partition model. R2, accuracy; NPV, negative predictive value; PPV, positive predictive value.
Figure 4. Construction of the diagnostic and statistical model of the Cate–Nelson classification for hot water boron showing the (a) Cate–Nelson graph for the critical thresholds identified; (b) number of points outside the model for determination of the critical relative yield (RY); (c) sum of squares for determination al of the critical BHW; and summary table (lower right)n: Number of points in the different quadrants; Performance indicators of the partition model. R2, accuracy; NPV, negative predictive value; PPV, positive predictive value.
Agronomy 09 00410 g004
Agronomy 09 00410 g005 550
Figure 5. Construction of the diagnostic and statistical model of Cate–Nelson classification for boron Mehlich-3 showing the (a) Cate–Nelson graph for the critical thresholds identified; (b) number of points outside the model for the determination of the critical relative yield (RY); (c) sum of squares for determination of the critical BM3; and and summary table (lower right) n: Number of points in the different quadrants; Performance indicators of the partition model. R2, accuracy; NPV, negative predictive value; PPV, positive predictive value.
Figure 5. Construction of the diagnostic and statistical model of Cate–Nelson classification for boron Mehlich-3 showing the (a) Cate–Nelson graph for the critical thresholds identified; (b) number of points outside the model for the determination of the critical relative yield (RY); (c) sum of squares for determination of the critical BM3; and and summary table (lower right) n: Number of points in the different quadrants; Performance indicators of the partition model. R2, accuracy; NPV, negative predictive value; PPV, positive predictive value.
Agronomy 09 00410 g005
Agronomy 09 00410 g006 550
Figure 6. Prediction of rate of born corresponding to maximum yield (RateMaximum yield) and statistical analysis. (a) Correlation between measured and predicted RateMaximum yield; (b) scaled importance of predictors (BM3, pHW, application type, boron requirement); (c) partial dependence on BM3; (d) partial dependence on pHW.
Figure 6. Prediction of rate of born corresponding to maximum yield (RateMaximum yield) and statistical analysis. (a) Correlation between measured and predicted RateMaximum yield; (b) scaled importance of predictors (BM3, pHW, application type, boron requirement); (c) partial dependence on BM3; (d) partial dependence on pHW.
Agronomy 09 00410 g006
Table
Table 1. Selected properties of the 365 soil samples in the plough layer (0–20 cm) (Data base 1).
Table 1. Selected properties of the 365 soil samples in the plough layer (0–20 cm) (Data base 1).
Soil PropertyRangeMeanStandard Deviation
pHW2.30–7.406.200.60
g·kg−1
Organic carbon8–4084568
Sand 0–890417234
Silt50–600341145
Clay60–850242159
mg·kg−1
BHW0.02–3.680.400.70
BM30.08–12,71.002.30
AlM353–1679966369
CaM389–1708438563964
MgM321–2940370513
FeM3101–1205263150
KM329–806235507
CuM30.50–30.603.003.30
ZnM30.40–81.405.308.60
pHW: pH water; BHW: Boron hot water; BM3: Boron Mehlich-3; AlM3: Aluminium Mehlich-3; CaM3: Cacium Mehlich-3; MgM3: Magnesium Mehlich-3; FeM3: Iron Mehlich-3; KM3: Potassium Mehlich-3; CuM3: Copper Mehlich-3; ZnM3: Zinc Mehlich-3.
Table
Table 2. Literature data for the boron agronomic model (Data base 2).
Table 2. Literature data for the boron agronomic model (Data base 2).
Referencen *CulturesReferencen *Cultures
Bagchi et al. [31]3RMahler and Shafii [17] 224L
Cifu et al. [32]1CsMahler and Shafii [18]268Be
Cirak et al. [33]1SMalhi et al. [34]14Can
Cutcliffe [35]10GpGanie et al. [36]3Fb
Dunn et al. [37]6RiOyewole and Aduayi [38]1T
Debnath et al. [39]2WRerkasem et al. [40]13Pe, Ri, S, Su, W, Bb
Devi et al. [41]2SSharma et al. [42]4Su
Dursun et al. [43]6T, P, CuSherell [44]7Rc, Wc, Ac, A
Gupta [45]6W, BSherell and Toxopeus [46]8A
Gupta [47]20A, Rc, TiSingh and Singh [48]3S
Gupta and Cutcliffe [15]3Br, Bs, CauTuran et al. [49]3A
Gupta and MacLeod [50]4TiValenciano et al. [51]3Ch
Hussain et al. [52]6MWojcik and Wojcik [53]2Pt
Davis et al. [54]4TWojcik [55,56]4Bc, Hb
Razmjoo and Henderlong [57]7AYu and Bell [58]1Ri
Lombin [59]30PeZhang et al. [60]3Ma
Total = 672
* number of points; A = Alfalfa; Ac = Alsike clover; B = Barley; Bb = Black beans; Bc = Black currant; Be = Beans; Bl = Blackgram; Br = Broccoli; Bs = Brussel sprouts; Cab = Cabbage; Can = Canola; Car = Carrot; Cau = Cauliflower; Ch = Chickpea; Cs = Chinese strawberry; Cu = Cucumber; Fb = French beans; Gp = Green peas; Hb = Highbush blueberry; L = Lentils; M = Mustard; Ma = Mandarin; P = Pepper; Pe = Peanuts; Pt = Pear tree; R = Radish; Rb = Red bayberry; Rc = Red clover; Ri = Rice; Ru: Rutabaga; S = Soybean; Su = Sunflower; T = Tomato; Tb = Table beets; Ti = Timothy; W = Wheat; Wc = White clover.
Table
Table 3. Literature data for the boron recommendation algorithms (Data base 3).
Table 3. Literature data for the boron recommendation algorithms (Data base 3).
Referencen *CulturesReferencen *Cultures
Bagchi et al. [31]2RGupta and Cutcliffe [66]2Tb, Car
Cifu et al. [32]1RbHaque [67]2T
Debnath et al. [39]2WHussain et al. [52]4M
Devi et al. [41]2SMahli et al. [34]13Can
Dursun et al. [43]6T, P, CuMaurya et al. [68]1R
Freeborn et al. [69]5SQuddus et al. [70]3L
Ganie et al. [36]3FbRerkasem et al. [40]7S, Pe, Bl, W
Gupta [45]6W, BSingh and Singh [48]3S
Gupta and Cutcliffe [71]20A, Rc, TiC. G Sherell [44]7A, Ac
Gupta and Cutcliffe [72]3RuS.P. Srivastava et al. [73]1L
Gupta and Cutcliffe [74]16RuTuran et al. [49]3A
Gupta and Cutcliffe [71]16Be, Cab
Total = 130
* number of points; A: Alfalfa; Ac: Alsike clover; B: Barley; Be: Beans; Bl: Blackgram; Cab: Cabbage; Can: Canola; Car: Carrot; Cu: Cucumber; Fb: French beans; L: Lentils; M: Mustard; P: Pepper; Pe: Peanuts; R: Radish; Rb: Red bayberry; Rc: Red clover; Ru: Rutabaga; S: Soybean; T: Tomato; Tb: Table beets; Ti: Timothy; W: Wheat.
Table
Table 4. BM3 prediction results by four AutoML (Automatic Machine Learnig) algorithms.
Table 4. BM3 prediction results by four AutoML (Automatic Machine Learnig) algorithms.
BM3 predicted vs BM3 measured
AlgorithmsPredictorsm1−maR2
A1BHW0.350.650.380.63
A2BHW, pHW0.630.370.230.85
A3BHW, pHW, OC0.880.120.080.98
A4BHW, pHW, OC, KM3, CaM3, MgM30.970.030.020.99
m: Slope; 1-m: Deviation; a: Intercept; R2: Robustness; BHW: Boron hot water; pHW: pH water; OC: Organic carbon; KM3: Potassium Mehlich-3; CaM3: Calcium Mehlich-3; MgM3: Magnesium Mehlich-3; BM3: Boron Mehlich-3.

Share and Cite

MDPI and ACS Style

Dupré, R.L.C.; Khiari, L.; Gallichand, J.; Joseph, C.A. Multi-Factor Diagnostic and Recommendation System for Boron in Neutral and Acidic Soils. Agronomy 2019, 9, 410. https://doi.org/10.3390/agronomy9080410

AMA Style

Dupré RLC, Khiari L, Gallichand J, Joseph CA. Multi-Factor Diagnostic and Recommendation System for Boron in Neutral and Acidic Soils. Agronomy. 2019; 9(8):410. https://doi.org/10.3390/agronomy9080410

Chicago/Turabian Style

Dupré, Richardly Lenz Clove, Lotfi Khiari, Jacques Gallichand, and Claude Alla Joseph. 2019. "Multi-Factor Diagnostic and Recommendation System for Boron in Neutral and Acidic Soils" Agronomy 9, no. 8: 410. https://doi.org/10.3390/agronomy9080410

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop