X-ray fluorescence (XRF) spectrometry is a proximal sensing technique whereby low-power X-rays are used to make elemental determinations in soils. The technique is rapid, portable, and provides multi-elemental analysis with results generally comparable to traditional laboratory-based techniques. Nawar et al. (2019) reported low to quite high prediction accuracies for K, Ca, Mg and P (coefficient of determination [R2] values of 0.83, 0.76, 0.69, and 0.47, respectively) in soils with a Portable XRF (pXRF). Also, pXRF data were able to predict total nitrogen (R2 = 0.50), and soil organic matter (R2=0.56) using Random Forest (RF) model with moderate accuracy (Andrade et al., 2020). Different models presented adequate results to predict exchangeable Ca2+ (R2 = 0.92), pH (R2 = 0.85), and base saturation (R2 = 0.90) using pXRF (Teixeira et al., 2018). Alder et al. (2020) revealed that prediction with non-linear models were more accurate than a multivariate linear model for Cu (R2 = 0.94) and Cd (R2 = 0.80) using pXRF. Besides total concentration of nutrients, exchangeable Ca, Al, Mg, potential and effective cation exchange capacity, soil organic matter, pH, and base saturation could also be predicted with reasonable accuracy by pXRF using RF model (Silva et al., 2017). The Data Fusion PLSR model on wet soils resulted in a more accurate plant extractable K predictive ability (R2 = 0.75), compared to the individual gamma ray or XRF sensors (Nawar et al., 2022). Kandpal et al. (2022) found that compared with the results obtained from single sensor model, spectral fusion models of mid-infrared and XRF showed improvement in the prediction performance for all studied attributes. Tarvares et al. (2019) examined that XRF analysis of pressed pellets allowed a slight gain in performance over loose powder samples for the prediction of exchangeable-K and Ca. Overall, pXRF has the potential to predict important properties of diverse soils at low cost, and without chemical waste generation, but the accuracy may be improved through further research.
2. Jyotirmay Roy
(Roll No. 12829)
Soil Science and Agricultural Chemistry
ICAR-Indian Agricultural Research Institute, New Delhi
Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Advancement of X-ray fluorescence spectroscopy
for quantification of nutrients in soil
11. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Prediction of total nitrogen (TN) and soil organic matter (SOM) using
pXRF
Andrade et al. (2020)
Coefficient of determination (R2) and root mean square error (RMSE) of the total nitrogen
(TN) and soil organic matter (SOM) prediction models.
OLS - ordinary least square.
CR - cubist regression
XGB - XGBoost
RF - random forest
R
2
RMSE
Measured SOM(g kg-1) Measured TN(g kg-1)
Predicted
SOM(g
kg
-1
)
Predicted
TN(g
kg
-1
)
12. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Prediction of available P and K using spectra fusion of mid-infrared
(MIR) and XRF spectroscopy
The mid-infrared (MIR) (a) and X-ray florescence (XRF) spectra (b) of samples
MIR-TPLS = mid-infrared-traditional partial least square;
XRF-TPLS = X-ray fluorescence traditional partial least square;
SF-PLS = spectra fusion based on partial least square;
SF-SOPLS = spectra fusion sequential orthogonalized partial least squares;
SF-VIP-SOPLS = spectra fusion variable importance projection sequential
orthogonalized partial least squares
Kandpal et al. (2022)
Absorbance
(%)
Wavenumbers (cm-1)
Intensity
(counts
per
second)
Energy (KeV)
13. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Continued…
Kandpal et al. (2022)
Parameters Model Type R2 RMSEP RPD RPIQ
Available P
(mg/100g)
MIR-TPLS 0.89 0.16 3.03 3.51
XRF-TPLS 0.88 0.17 2.95 2.78
SF-PLS 0.88 0.17 2.95 2.76
SF-SOPLS 0.90 0.15 3.30 3.59
SF-VIP-
SOPLS
0.90 0.15 3.22 3.54
Available K
(mg/100g)
MIR-TPLS 0.65 14.1 1.70 1.90
XRF-TPLS 0.48 15.0 1.37 1.68
SF-PLS 0.48 14.9 1.39 1.56
SF-SOPLS 0.67 11.7 1.77 2.13
SF-VIP-
SOPLS
0.64 12.3 1.68 1.67
R2 = coefficient of determination
RMSEP = root mean square error of prediction
RPD = Residual prediction deviation
RPIQ = ratio of performance to interquartile distance
Predicted vs. measured scatter plots SF-
SOPLS
14. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Measurement of available K in soils through fusion of Gamma-rays and
pXRF spectral data
Nawar et al. (2022)
Spectral data Statistical analysis Calibration Validation
Gamma-ray R2 0.82 0.71
RMSE 47.6 31.7
RPD 2.39 1.89
RPIQ 2.62 2.36
Wet soil XRF R2 0.67 0.61
RMSE 44.6 48.8
RPD 1.74 1.64
RPIQ 2.18 1.84
Dry soil XRF R2 0.83 0.77
RMSE 37 26.5
RPD 2.44 2.14
RPIQ 2.15 3.39
Data fusion R2 0.80 0.75
RMSE 35.5 31.3
RPD 2.23 2.03
RPIQ 2.81 2.79
RMSE = root mean square error
RPD = ratio of performance deviation
RPIQ = ratio of performance to interquartile
range.
Data fusion model was developed using
gamma ray and XRF spectra measured on
wet soils.
15. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Portable XRF for predicting potassium (K), phosphorus (P), magnesium
(Mg) and calcium (Ca)
Nawar et al. (2019)
A total of 105 soil samples from a wide
range of soils collected from 10 different
countries were scanned using an Oxford
XMET8000 XRF spectrometer.
Parameters
(mg kg-1)
R2
Validation=31
RMSEP MAE RPQI
(mg/kg) (mg/kg)
Available K 0.84 2833 1569 2.33
Available P 0.5 429 342 1.49
Exchangeable Ca 0.76 7406 6024 2.15
Exchangeable Mg 0.55 3119 2362 1.92
R2 = Coefficient of determination
RMSE = Root mean square error of cross-validation
RMSEP = Root mean square error of prediction
RPIQ = Ratio of performance to interquartile range
MAE = mean absolute error
16. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Continued…
Scatter plots of measured versus predicted potassium (K), phosphorous (P),
calcium (Ca) and magnesium (Mg) Nawar et al. (2019)
Predicted
K
(mg
kg
-1
)
Predicted
Ca
(mg
kg
-1
)
Predicted
P
(mg
kg
-1
)
Predicted
Mg
(mg
kg
-1
)
Measured K (mg kg-1) Measured P (mg kg-1)
Measured Ca (mg kg-1) Measured Mg (mg kg-1)
17. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Predicting potassium (K), phosphorus (P), magnesium (Mg) and calcium
(Ca) using pXRF
Benedet et al. (2019)
Generalized Linear Model (GLM) and Random Forest (RF) models for assessing
available K (av. K), available P (av. P), exchangeable Ca2+ (ex. Ca), exchangeable Mg2+
(ex. Mg), exchangeable Al3+ (ex. Al), and remaining P (P-rem)
R2 = Coefficient of determination
RMSE = Root mean square error of cross-validation
18. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Estimation of exchangeable Ca and pH by pXRF
Teixeira et al. (2018)
Predicted valiable PXRF
data
Equation R2
Exchangeable Ca2+
(cmolc dm-3)
CaO y = 1.4999ln(*CaOpXRF) - 9.627 0.92
pH CaO y = -1.669ln(CaOpXRF) + 14.315 0.85
Equations to predict values of soil chemical analyses from pXRF and coefficients
of determination (R²) in Brazil.
19. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Available Mn and Cu analysis using pXRF spectrometry
Tatiane et al. (2019)
Parameters Equation p r R2
Available Mn Mn = −5.96 + 0.08(Mn-pXRF) <0.01 0.72 0.51
Available Cu
Cu = 0.31 + 0.06(Cu-pXRF) − 0.003(Mn-pXRF) +
0.06(Zn-pXRF) <0.01 0.83 0.69
Multiple and simple regression models obtained to predict properties of Brazilian
Cerrado soils based on pXRF data.
Predicted
available
Mn
(mg
kg
-1
)
Predicted
available
Cu
(mg
kg
-1
)
Observed available Mn (mg kg-1) Observed available Cu (mg kg-1)
20. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Prediction of available Cu and Zn concentration in soil by PXRF using
different models
Adler et al. (2020)
Lab analysed (mg kg-1)
Concentrations of copper (Cu) and zinc (Zn)
predicted from PXRF using multiple linear
regression (MLR), random forest regression (RF),
and multivariate adaptive regression splines
(MARS).
Model R2 MAE
Cu-MLR 0.90 4.40
Cu-RF 0.84 4.51
Cu-MARS 0.94 3.21
Zn-MLR 0.96 4.40
Zn-RF 0.94 5.40
Zn-MARS 0.97 4.00
R2 = coefficient of determination;
MAE = mean absolute error (mg kg−1 )
21. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Determination of base saturation percentage using pXRF
Rawal et al. (2019)
Model Validation R2 RMSE RPD RPIQ
GAM 0.58 9 1.6 1.77
MLR 0.45 10.4 1.4 1.54
RT 0.68 7.9 1.8 2.01
RF 0.62 8.7 1.6 1.85
Root mean square error (RMSE),
residual prediction deviation (RPD),
and ratio of performance to
interquartile (RPIQ) range.
GAM = generalized additive model
MLR = multiple linear regression
RT = regression tree
RF = random forest
Predicted
BSP
(%)
Predicted
BSP
(%)
Predicted
BSP
(%)
Predicted
BSP
(%)
Measured BSP (%) Measured BSP (%)
22. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Prediction of exchangeable Ca and K using PXRF as affected by different
sample preparing techniques
Soil samples without the addition of binder and with different grinding
times (A); pellets resulting from tests with different grinding times,
cellulose concentrations and brands (B)
Tavares et al. (2019)
23. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Continued…
Tavares et al. (2019)
Scatter plots of measured vs predicted ex-K, for the pellets and loose soil (A and B),
and of ex-Ca (C and D).
ex-Ca measured (mmolc dm-3)
ex-Ca measured (mmolc dm-3)
ex-Ca
predicted
(mmol
c
dm
-3
)
ex-Ca
predicted
(mmol
c
dm
-3
)
ex-K
predicted
(mmol
c
dm
-3
)
ex-K
predicted
(mmol
c
dm
-3
)
ex-K measured (mmolc dm-3) ex-K measured (mmolc dm-3)
24. Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Assessment of different metals in soil using XRF
Hu et al. (2014)
Certified PXRF
Accuracy
RPD (%)
Precision
RSD (%)
Mean
(mg/kg) SD
Mean
(mg/kg) SD
As 34 4 34.5 2.1 1.5 6.1
Pb 98 6 89.8 2.2 -8.4 2.5
Cu 21 2 21.5 3.6 2.4 16.7
Zn 680 25 580.3 15.1 -14.7 2.6
SD standard deviation
RPD relative percent difference
RSD relative standard deviation
Correlation between heavy metal
certified values and the values measured
by portable XRF.
Total concentrations of As, Pb, Cu
and Zn were measured in 47
agricultural soils.
25. Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI,
New
Delhi Challenges of applying XRF in soil analysis
XRF in soil analysis
Complex soil matrix
Spectrally inactive
parameters give
errorous results
Overfitting of data in
machine learning
NOT a substitute of
traditional
estimation methods
Redundancy and
collinearity among
predictors
Higher limit of
detection for
low atomic
number
molecules
Zhang and Wang (2024)
26. Conclusion
Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
XRF can moderate to almost accurately predict different total as well as
available nutrients present in soil.
Due to low detection limit, excellent selectivity, repeatability and stability of XRF
may be a great substitute for traditional systems.
Prediction accuracies in XRF considerably increased with data processing
techniques viz machine learning and non-linear regression models.
Machine learning and non linear regression model outperformed linear
regression equations.
27. Future scope
Division
of
Soil
Science
and
Agricultural
Chemistry,
ICAR-IARI
Expanding the X-ray fluorescence spectral libraries to include diverse soil
samples.
Testing advanced modeling strategies to address matrix effects and the local
context of the ratio between the total and plant-available contents (T/A ratio).
Evaluating the potential fusion with other sensing techniques that can serve as
auxiliary data to improve predictive performances and extend the monitored
attributes.
Assessing the potential of in situ applications and approaches to mitigate
external effects (e.g., soil moisture and roughness).