Prediction Of Soil Properties Research Articles

Simultaneous prediction of soil properties from VNIR-SWIR spectra using a localized multi-channel 1-D convolutional neural network

The use of visible near-infrared and shortwave-infrared (VNIR-SWIR) diffuse reflectance spectroscopy for the estimation of soil properties is increasingly maturing with large-scale soil spectral libraries (SSLs) of laboratory spectra developed across the globe. Such an SSL is the publicly available LUCAS topsoil database with approximately 20,000 soil samples encompassing 23 countries of the European Union. A wide variety of machine learning tools have been applied to the LUCAS SSL to predict some of the soil samples’ physicochemical properties with different degrees of accuracy. In this paper, we developed and examined the use of a novel one-dimensional convolutional neural network (CNN) to simultaneously predict ten physicochemical properties of the LUCAS SSL. Leveraging on the use of multiple-input channels it uses as model inputs the absorbance spectra along with some pre-processed spectra developed using standard techniques. Moreover, it exploits the use of local spectral neighborhoods to perform an adaptive error-correction mechanism. This novel localized multi-channel 1-D CNN was applied to all the available physicochemical properties of the LUCAS SSL and was statistically compared with the current state-of-the-art where it was shown to statistically outperform its counterparts, as well as with other CNNs where it exhibited the best performance. In particular, for the mineral soil samples, the RMSE for the Clay content was 4.80% (R2 0.86), for soil organic carbon the RMSE was 10.96 g kg−1 (R2 0.86), while for total nitrogen the RMSE was 0.66 g kg−1 (R2 0.83).

VisNIR integrated multi-sensing penetrometer for in situ high-resolution vertical soil sensing

An in situ penetrometer system that can measure profile soil properties rapidly, cost-effectively, and at high vertical resolution would benefit the soil science and agriculture communities. A visible and near infrared (VisNIR) integrated multi-sensing penetrometer system was developed to automatically measure in situ soil VisNIR reflectance spectra, penetration resistance, and insertion depth along a soil profile. The system was tested in 11 agricultural fields in Nebraska, Illinois, Iowa, and South Dakota of the U.S. An independent soil VisNIR spectral library was used to build calibration models for soil property prediction. External Parameter Orthogonalization (EPO) was used to correct for the field intactness of in situ VisNIR spectra. The results showed that EPO was effective in correcting for the spectral disparity between in situ and dry-ground VisNIR spectra. The EPO correction showed an improvement of prediction accuracy of soil total carbon (R2 and RMSE improved from 0.29 and 3.06 % to 0.5 and 0.79 %, respectively) and total nitrogen (R2 and RMSE improved from 0.51 and 0.36 % to 0.62 and 0.06 %, respectively). The system also predicted soil bulk density with an RMSE of 0.12 g cm−3 and R2 of 0.80. It is concluded that the VisNIR multi-sensing penetrometer, along with the use of external soil spectral libraries and the spectral correction algorithm EPO, can lead to a rapid, robust and cost-effective system for in situ high resolution vertical soil sensing.

Three-dimensional digital soil mapping of multiple soil properties at a field-scale using regression kriging

Three-dimensional digital soil mapping (3D-DSM) quantifies both the horizontal and the vertical variability of soil properties. Most current studies in 3D-DSM were based on either one-dimensional profile depth functions or two-dimensional horizontal interpolation techniques, which did not allow true 3D visualization of spatial soil heterogeneity. Only a few studies have utilized the 3D variograms for mapping. Recent advances in proximal soil sensing technologies allow measurement and prediction of soil properties rapidly at multiple depths which could serve as input data for DSM. Various soil physical and chemical properties have already shown either direct or indirect relationships with the proximal soil sensing data. This study aims to test the methodology of 3D-DSM by incorporating a 3D regression kriging (RK) with multiple proximal soil sensing techniques. In this study, vis-NIR spectra were collected in-situ at 148 locations to about 1-m depth using the Veris® P4000 soil profiler at Field 26 of Macdonald Farm, McGill University. Additionally, 32 soil cores were collected out of the 148 locations to 1-m maximum depth and sectioned at 10-cm depth intervals for laboratory analysis of volumetric water content (VWC), soil organic matter (SOM), and clay content. Cubist spectral models were developed for each soil property at the 32 locations and then predicted to the 148 locations, which were then randomly split into calibration (70%, 103 locations) and validation (30%, 45 locations) datasets for mapping. The 3D-RK method included a trend prediction between calibration dataset and environmental covariates (including apparent soil electrical conductivity, gamma-ray radiation, and elevation) and a residual kriging. The generalized linear model (GLM), regression tree (RT), and random forest (RF) models were compared for trend prediction. The covariates were also simulated 100 times using sequential Gaussian simulations to fit into 3D-RK and calculate model uncertainty. As a result, complete 3D digital soil maps with uncertainty were developed. We found that the RF model outperformed GLM and RT in regard to interpreting non-linear soil-landscape relationships and resulting in marginally higher validation accuracy and smaller prediction uncertainty for VWC and clay. The GLM model resulted in slightly better validation results and smaller model uncertainty for SOM only. SOM and clay showed large horizontal and vertical variability and affected the spatial distribution of VWC. The validation accuracy was higher in the soil surface for most soil properties due to the uniform environment in the plow layer and sufficient environmental covariates collected at the soil surface. The mapping uncertainty increased with depth for VWC and clay content but decreased with depth for SOM because SOM content decreases with depth.

Tuning support vector machines regression models improves prediction accuracy of soil properties in MIR spectroscopy

Estimating soil properties in diffuse reflectance infrared Fourier transform spectroscopy in the mid-infrared region (mid-DRIFTS) uses statistical modeling (chemometrics) to predict soil properties from spectra. Modeling approaches can have major impacts on prediction accuracy. However, the impact of selecting best parameters for an algorithm (tuning), to optimize non-linear models for predicting soil properties, is relatively unexplored in the domain of soil sciences. This study aimed to evaluate the predictive performance of linear (partial least squares, PLS) and non-linear (support vector machines, SVM) multivariate regression models in estimating soil physical, chemical, and biological properties with mid-DRIFTS. We evaluated the impact of optimizing two hyperparameters (epsilon and cost) based on the noise tolerance in the ε-insensitive loss function of SVM models using two contrasting and diverse sets of soils, one from northern Tanzania (n = 533) and another one from USA Midwest (n = 400). Regression models were trained on calibration sets (75%) and tested on independent validation sets (25%) separately for each dataset. Support vector machines outperformed PLS models for all tested soil properties (clay, sand, pH, total organic carbon, and permanganate oxidizable carbon) in both datasets. Tuning hyperparameters epsilon and cost maintained or improved prediction accuracy of SVM models based on root mean squared errors of independent validation sets. Support vector machines tuned hyperparameters differed among soil properties and also for the same soil property in distinct datasets, suggesting the need for parameterizing non-linear models for specific soil properties and datasets. Optimizing SVM regression models in mid-DRIFTS improves prediction accuracy of soil properties and therefore will likely enable obtaining more robust predictive outcomes even in datasets with diverse land uses, parent materials, and/or soil orders. We recommend that tuning should be included as a routine step when using SVM for estimating soil properties.

Mid-infrared spectroscopy as an alternative to laboratory extraction for the determination of lime requirement in tillage soils

Lime is a crucial soil conditioner to bring agricultural soils to optimum pH values for nutrient availability. Lime recommendations are typically determined in laboratory extractions, the most common being the “Shoemaker-McLean and Pratt” (SMP) buffer method, that requires carcinogenic reagents soon to be abolished under the EU legislation. As an alternative to wet chemistry, mid-infrared (MIR) spectroscopy has shown to be a cost-and time effective method at predicting soil properties. The capability and feasibility of diffuse reflectance infrared spectroscopy (DRIFTS) to predict lime requirement (LR) in tillage fields is examined. Samples from 41 cereal tillage fields (n = 655) are used to build a calibration for DRIFTS using partial least squares regression (PLSR). The samples were split into calibration set (31 fields, n = 495) and validation set (10 fields, n = 160). After pre-processing with trim, smoothing and standard normal variate, a calibration model using 6 latent variables, provided R2 of 0.89 and root mean square error of cross-validation (RMSECV)of 1.56 t/ha. Prediction of all fields from the validation set resulted in R2 of 0.76 and root mean square error of prediction (RMSEP) of 1.68 t/ha. The predictions of the single fields ranged from R2 values of 0.41 to 0.72, RMSEP of 0.48 to 4.2 t/ha and ratios of performance to inter-quartile distance (RPIQ) of 0.45 to 3.56. It was shown that the signals of soil constituents having an influence on the LR were picked up in the spectra and were identified in the loading weights of the PLSR. While the error is too high to predict the variability of LR within the field, MIR prediction using field averages provided a viable alternative to current laboratory methods for blanket spreading of lime on tillage fields.

Modelling soil properties from horizon depth functions and terrain attributes: An example with cation exchange capacity

The objective of this study was to, through the distribution of some soil properties, model cation exchange capacity (CEC) in soils formed on gently undulating coastal plain sands of southeastern Nigeria using genetic horizon functions and terrain attributes. A total of 19 profile pits were prepared, described and 104 genetic horizons were identified and sampled, processed and analysed in the laboratory. Data were generated on the soil characteristics, including particle size fractions, hydraulic conductivity, bulk density, organic carbon, pH and electrical conductivity. Terrain attributes that were generated from digital elevation model include aspect, compound topographic index (CTI), Flow direction, curvatures, slope and stream power index (SPI). Data generated were analysed using descriptive statistics, correlation and regression. The terrain attributes were modified with genetic horizon depths, bulk density and clay content for the modelling process. Sand content, bulk density and cation exchange capacity possess geogenic rather than pedogenic characteristics and were normally distributed. The indication is that the two groups of terrain attributes depended on the mass per unit area of soil and clay content in their influence on these ultisol profiles. Paired comparison, root mean square error and normalized root mean square error indicated that the model was a good fit and could be useful in the prediction of soil properties and management of coastal plain sands.

Effect of land-use management systems on coupled physical and mechanical, chemical and biological soil processes: how can we maintain and predict soil properties and functions?

Effect of land-use management systems on coupled physical and mechanical, chemical and biological soil processes: how can we maintain and predict soil properties and functions?

Development of Mid-Infrared Spectroscopic Feature-Based Indices to Quantify Soil Carbon Fractions

Mid-infrared spectroscopy (MIRS) provides a rapid-throughput methodology in predicting many soil properties. The objective was to evaluate the performance of spectral feature-based indices to predict soil carbon fractions and compare it to partial least square regression (PLSR) model. We scanned 135 soil samples by MIRS and analyzed for total organic carbon (TOC), microbial biomass carbon (Cmic), hot-water extractable carbon (HWEC) and total inorganic carbon (TIC). Ninety samples were used for model calibration, while remaining 45 samples were further used for model validation. To develop spectral indices, peak area at specific absorbance feature related to labile organic carbon (PA 2930 cm–1) and carbonate (PA 2515 and 713 cm–1) was calculated and correlated to measured values of interest. PLSR resulted in successful calibrated models with R2 = 0.95, 0.99 for TOC, TIC, and moderate successful models with R2 = 0.85, 0.81 for Cmic and HWEC, respectively. The same results were also obtained for model validation. In terms of spectral indices, regression analyses between PA 2930 cm–1 and values of TOC, Cmic and HWEC was moderate successful with R2 values 0.85 to 0.88. Correlation between PA 2515 and 713 cm–1 and TIC resulted in a successful regression model with R2 = 0.98. As conclusion, precision level of spectral indices in predicting soil properties is almost equivalent to the PLSR models. However, correct large-scale application of the method requires excluding spectral mineral interference and model development by taking samples with a greater variability of soils and mineralogy.

Estimating the spatial distribution of soil total nitrogen and available potassium in coastal wetland soils in the Yellow River Delta by incorporating multi-source data

Abstract Coastal wetland soils play a vital role in providing habitats for wildlife and vegetation, and in affecting water quality and primary production in coastal wetland ecosystem. The spatial prediction of soil properties provides important data support to help sustainable wetland management. This research explored the relationships between soil total nitrogen (TN) and soil available potassium (AK) and multiple environmental variables such as remote sensing spectral indices, geographic attributes and land use types of coastal wetlands in the Yellow River Delta (YRD). Soil moisture and geographic attributes were the major factors controlling the spatial distribution of soil TN. Soil salinity, vegetation and geographic attributes were the key factors affecting the spatial distribution of soil AK. The research established spatial prediction models of soil TN and AK by random forest method, and showed heterogeneous spatial patterns of TN and AK in the study area. Soil TN model showed relatively higher prediction performance compared with soil AK model. The research framework utilized in this study can be promoted to predict and map other soil properties, and has significant implication for the site-specific soil management of coastal wetlands in the YRD.

Tropical soil pH and sorption complex prediction via portable X-ray fluorescence spectrometry

Portable X-ray fluorescence (pXRF) spectrometry delivers results rapidly, at low-cost, and without generating chemical residues. This study aimed to predict soil pH, sum of bases (SB), base saturation percentage (BSP), cation exchange capacity (CEC), and Al saturation (Alsat) of 2017 contrasting Brazilian soil samples through the association of pXRF and three different algorithms [Cubist, Random forest (RF), and stepwise multiple linear regression (SMLR)]. Soil samples were collected from the surface (SURF) and subsurface (SUB) horizons in seven Brazilian states. The prediction models were generated for the SURF and SUB horizons separately and combined (SURF + SUB dataset). Overall, the best predictions were achieved via Cubist followed by RF. For the pH predictions, the model combining SURF and SUB horizons data presented better results. Satisfactory results were achieved for the predictions of SB (validation R2 = 0.86), BSP (validation R2 = 0.81) and Alsat (R2 = 0.76). Moreover, promising results were obtained for predicting pH (R2 = 0.63). Notably, CaO appeared as the most influential variable for soil property prediction models. Overall, pXRF showed great potential for predicting soil fertility properties for diversified tropical soils with low cost, rapidity, and without chemical waste generation.

Increment-averaged kriging for 3-D modelling and mapping soil properties: Combining machine learning and geostatistical methods

Prediction and mapping of soil properties for different soil depths can provide important information for effective land management. Such predictions and maps are often built based on soil data from multiple soil surveys, and the sampled depths will rarely align with depths for which the predictions and maps are required. A recently proposed approach to deal with such datasets, termed increment-averaged kriging (IAK), fits a single model using data from all profiles and all soil depths. The approach considered several potential stumbling blocks: the effect that different widths of the sampled depth intervals has on the variances (sample support); potential non-stationarity in the covariances, depending on soil depth; the need to account for interactions between depth and the effect of covariates. In this work, we present some extensions to the original work that (i) extend the covariance model, (ii) integrate the use of a machine learning algorithm, Cubist, into the framework so that its regression parameters can be fitted while accounting for the correlation and sample support of the data, and (iii) apply a composite likelihood approximation to enable estimation and prediction with large datasets. Code to implement the method in R is also made available so that practitioners can test IAK on their own datasets.

Digital soil mapping of key GlobalSoilMap properties in Northern Karnataka Plateau

Accurate and quantitative information on soil properties of each and every location is essential for site specific sustainable management of land resources. A study was conducted to predict the different key soil properties of Northern Karnataka as per GlobalSoilMap specifications using Quantile Regression Forest (QRF) Model. Along with Sentinel-2 data, terrain attributes such as elevation, slope, aspect, topographic wetness index, topographic position index, plan and profile curvature, multi-resolution index of valley bottom flatness, multi-resolution ridge top flatness and vegetation factors like NDVI and EVI were used as covariates. Equal-area quadratic splines were fitted to soil profile datasets to estimate soil properties viz. pH, OC, CEC, clay, sand, silt, field capacity and permanent wilting point at six standard soil depths (0–5, 5–15, 15–30, 30–60, 60–100 and 100–200 cm) as per GlobalSoilMap specifications. The coefficient of determination (R2), mean error (ME) and root mean square error (RMSE) were calculated in order to assess model performance. Prediction interval coverage percentage (PICP) was calculated to evaluate the associated uncertainty predictions. The predicted soil properties are reliable with minimum errors and the QRF model captured maximum variability for most of the soil properties.

Prediction of soil properties with machine learning models based on the spectral response of soil samples in the near infrared range

Abstract One of the basic methods for soil analysis time and cost reduction is using soil sample spectral response in laboratory conditions. The problem with this method lies in determining the relationship between the shape of the soil spectral response and soil physical or chemical properties. The LUCAS soil database collected by the EU’s ESDAC research centre is good material to analyse the relationship between the soil properties and the near infrared (NIR) spectral response. The modelling described in the paper is based on these data. The analysis of the impact of soil properties configuration on absorbance levels in various NIR spectrum ranges was conducted using the stepwise regression models with the properties, properties squared and products of properties being explanatory variables. The analysis of partial correlation of soil properties values with absorbance values and absorbance derivative in the entire spectral range was conducted in order to evaluate the impact of the absorbance transformation (the first derivative of absorbance vector) on the change of significance of relationship with properties values. The Multi Layer Perceptron (MLP) models were used to estimate the absorbance relationship with single soil features. Soil property modelling based on the selection and transformation algorithm of raw values and first and second absorbance derivatives was also conducted along with the suitability evaluation of such models in building digital soil maps. The absorbance is affected by a limited number of tested soil features like pH, texture, content of carbonates, SOC, N, and CEC; P and K contents have, in case of this research, a negligible impact. The NIR methodology can be suitable in conditions of limited soil variation and particularly in development of thematic soil maps.

Prediction and Analysis of Soil properties in Guntur District, Andhra Pradesh

A significant statistical analysis is done over Eleven Thousand and Two hundred soil data samples of thirteen villages in Guntur District of Andhra Pradesh. Useful relationship among major soil parameters was found. An analytical approach is used to assess the properties of the soil using pandas, numpy and matplotlib libraries in python. Hidden patterns among the soil properties are identified. By using Random Forest Regressor soil data of Guntur district is predicted with considerably higher accuracy.

Application of Mid‐Infrared Spectroscopy in Soil Survey

Development of more efficient (rapid) and cost‐effective methodologies are needed in soil survey to meet the demand for quantitative data in digital soil mapping and updates. The objective of this study was to pilot the application of mid‐infrared (MIR)‐diffuse reflectance spectroscopy (DRS) coupled with partial least squares regression (PLSR) in a soil survey field office where soil samples are processed, MIR spectra are acquired, and predictions are obtained using calibration models developed and validated from the Kellogg Soil Survey Laboratory spectral library. Mid‐infrared models were built for total C, organic C, CaCO3 equivalent, total clay, cation exchange capacity, 1500 kPa water, and pH in water and CaCl2 for Mollisols of the central United States. Validation results (from the MIR library) using Lin's concordance correlation (rc) of measured versus predicted values showed that most properties predicted very well (rc = 0.967–0.996), whereas models for total clay in B horizons and 1500 kPa water in B horizons predicted fairly well (rc = 0.844–0.955). Models for pH predicted the least well (rc = 0.750–0.921). The MIR‐DRS coupled with PLSR was successful in predicting soil properties for completely independent samples that were collected, processed, and MIR scanned in a soil survey field office. Predicted results using rc ranged from 0.697 to 0.992, with pH in water having the lowest rc and CaCO3 having the highest rc. All properties except pH had an acceptable level of accuracy for use in soil survey and a marginal level of acceptable accuracy for total clay. Direct calibration transfer was feasible.

Visible and Near-Infrared Reflectance Spectroscopy Analysis of a Coastal Soil Chronosequence

The soil chronosequence is a useful method for investigating pedological theories. Soil chemical, physical and mineralogical properties in chronosequences change over time and exhibit systematic and time-dependent trends, which can be used to analyze the rates and directions of pedogenic changes. The potential of soil spectroscopy as an emerging, rapid and cost-effective technique for predicting soil properties has been widely accepted and has motivated the application of spectroscopic techniques to the analysis of soil chronosequence. We present a soil chronosequence derived from 1000-year-old calcareous marine sediments and examine changes in six soil properties over this period. We evaluated the utility of a soil spectroscopic method to detect soil property changes and to predict the pedogenic properties and soil ages of the chronosequence. The results show that some soil pedogenic processes, such as soil organic matter accumulation, CaCO3 leaching and clay migration, can be identified in the millennium chronosequence. Power chronofunctions are formulated for soil organic matter (SOM) and Logarithmic chronofunctions are fitted for clay, CaCO3 and pH. These pedogenic processes are identified in the reflectance intensity and absorption features of soil spectroscopy, and pedogenic properties can be calibrated via soil reflectance spectroscopy. Profile ages can also be predicted via pseudo multi-depth spectra of soil profiles, and soil spectral curves for 0–30 cm generated the best prediction results (RPD = 1.85). We conclude that soil properties, changing due to weathering and soil formation, act as a bridge linking spectroscopy and weathering levels/pedogenic processes. The results imply that applying spectroscopy techniques to chronosequence study and mapping the degree of soil development in certain areas should be possible.

Assessing heavy metal concentrations in earth-cumulic-orthic-anthrosols soils using Vis-NIR spectroscopy transform coupled with chemometrics

Soil visible and near infrared (Vis-NIR) has become an applicable and interesting technique to predict soil properties because it is a fast, cost-effective, and non-destruction technique. This study presents an application of diffuse reflectance spectroscopy (DRS) and chemometric techniques for evaluating concentrations of heavy metals in earth-cumulic-orthic-anthrosols soils. 44 soil samples of 0-30 cm were collected from three representative agriculture areas (Fufeng, Yangling, and Wugong transects with 16, 10, and 18 samples, respectively) and analyzed for Cr, Mn, Ni, Cu, Zn, As, Cd, Hg, and Pb by Vis-NIR spectroscopy (350-2500 nm). Average levels of Cr, Mn, Ni, Cu, Zn, As, Cd, Hg, and Pb were 17.95, 274, 12.77, 7.29, 15.81, 7.51, 0.40, 12.58, and 21.05 mg kg-1, respectively. Twenty-four preprocessing methods were extracted sensitive bands. Partial least squares regression (PLSR) used to obtain effective bands and predict soil heavy metals concentrations. The accuracy of the predictive models were assessed in terms of coefficient of determination (R2), the root mean squared error (RMSE), standard error (SE) and the ratio of performance to deviation (RPD). The results revealed that excellent predictions for Hg(Rv2 = 0.99, RPD = 8.59, RMSEP = 0.12, SEP = 0.13), Cr (Rv2 = 0.97, RPD = 5.96, RMSEP = 0.10, SEP = 0.10), Ni (Rv2 = 0.93, RPD = 3.74, RMSEP = 0.13, SEP = 0.13), Pb (Rv2 = 0.97, RPD = 5.57, RMSEP = 0.10, SEP = 0.01), and Cu (Rv2 = 0.92, RPD = 3.38, RMSEP = 0.08, SEP = 0.08). Models for As (Rv2 = 0.87, RPD = 2.58), Mn (Rv2 = 0.80, RPD = 2.09), and Cd (RPD = 2.77) had Rv2 < 0.9 and RPD<3.0, not excellent predictions. For the element of Zn, although Rv2 = 0.91, RPD = 3.13, the offset had too much deviation, and it cannot be considered an excellent model. Therefore, a combination of spectroscopic and chemometric techniques can be applied as a practical, rapid, low-cost and quantitative approach for evaluating soil physical and chemical properties in Shaanxi, China.

Data fusion of vis–NIR and PXRF spectra to predict soil physical and chemical properties

AbstractWe tested one front‐end data fusion method to combine visible near‐infrared (vis–NIR) and portable X‐ray fluorescence (PXRF) spectra for predicting different soil properties and investigated the contribution of different sensor data. A total of 197 soil samples were collected from 25 Alfisols and Mollisols in south‐central Wisconsin, USA. Soils were analysed in the laboratory for clay, sand, silt content, total carbon (TC), total nitrogen (TN) and pH. Air‐dried soil samples were scanned with vis–NIR and PXRF spectrometers. A principal component analysis (PCA) was applied to each of the vis–NIR and PXRF spectra to extract the first 10 principal components (PCs), which were used in the Cubist model. Five types of input data were compared for constructing the models using a front‐end data fusion approach, including: (a) 10 PCs from vis–NIR spectra, (b) 10 PCs from spectra of PXRF beam 1 (XRF40), (c) 10 PCs from spectra of PXRF beam 2 (XRF10), (d) concatenating 20 PCs from two PXRF spectra (XRF40 + 10), and (e) concatenating 30 PCs from two PXRF spectra and vis–NIR spectra (XRF40 + 10 + NIR). In addition, the performances of the preprocessing methods (four smoothing methods with or without background removal) of PXRF spectra were compared. Multiple linear regression was also used to predict soil properties directly from 11 PXRF‐estimated elements. Our results suggest that smoothing should be applied to the PXRF spectra prior to developing predictive models. PXRF spectra and elemental data can predict soil texture with validation R2 &gt; 0.85, better than using solely vis–NIR spectra, and combining vis–NIR and PXRF spectra improved the prediction. Combining PXRF and vis–NIR can also predict TC and TN moderately well. Soil pH cannot be predicted from vis–NIR or PXRF spectra in this dataset due to its weak correlations with other soil properties and elements. It is concluded that PXRF spectra can be solely used to estimate soil texture, whereas combining vis–NIR and PXRF spectra via PCA should be used to estimate TC and TN.Highlights The use of front‐end data fusion method to combine vis–NIR and PXRF spectra for predicting soil properties. Different preprocessing algorithms were compared systematically for PXRF spectra. Spectral smoothing should be applied to the PXRF spectra prior to developing predictive models. PXRF spectra can be used solely to estimate soil texture, whereas combining vis–NIR and PXRF spectra via PCA should be used to estimate TC and TN.

Prediction of soil fertility via portable X-ray fluorescence (pXRF) spectrometry and soil texture in the Brazilian Coastal Plains

Traditional methods of soil chemical analysis are time consuming, costly, and generate chemical waste. Proximal sensors, such as portable X-ray fluorescence (pXRF) spectrometry, may help to overcome these issues since they have been shown to produce accurate predictions of many soil properties. However, such processes need to be further investigated in Brazilian soils. This work aimed to assess the influence of soil management and mineralogy on elemental composition of soils and predict exchangeable Al3+, Ca2+, Mg2+, and available K+, and P contents from pXRF data alone and associated with soil texture through machine learning algorithms [stepwise generalized linear models (SGLM), and random forest (RF)] in soils of the Brazilian Coastal Plains (BCP). A total of 285 soil samples were collected from the A (n = 123) and B (n = 162) horizons and subjected to laboratory analyses and pXRF scans. Samples were randomly separated into 70% for modeling and 30% for validation. Soil mineralogy and management mainly influenced Al, and Ca and K total content, respectively. In general, the inclusion of the auxiliary input data of soil texture did not change the predictive power of the models. The best results highlight a considerable promise of pXRF technique for rapidly assessing exchangeable Ca2+ (RMSE = 176.3 mg kg−1, R2 = 0.71), Mg2+ (37.7 mg kg−1, 0.60), and available K+ (27.46 mg kg−1, 0.67). The algorithms could not generate reliable models to predict exchangeable Al3+ (30.6 mg kg−1, 0.47) and available P (19.9 mg kg−1, 0.14). In sum, pXRF can be used to reasonably predict soil fertility properties in the BCP soils. Further studies may extend predictions to other soil properties.

Assessing models for prediction of some soil chemical properties from portable X-ray fluorescence (pXRF) spectrometry data in Brazilian Coastal Plains

Portable X-ray fluorescence (pXRF) spectrometry is becoming increasingly popular for predicting soil properties worldwide. However, there are still very few works on this subject under tropical conditions. Therefore, the objectives of this study were to use pXRF data to characterize the Brazilian Coastal Plains (BCP) soils and assess four machine learning algorithms [ordinary least squares regression (OLS), cubist regression (CR), XGBoost (XGB), and random forest (RF)] for prediction of total nitrogen (TN), cation exchange capacity (CEC), and soil organic matter (SOM) using pXRF data. A total of 285 soil samples were collected from the A and B horizons representing Ultisols, Oxisols, Spodosols, and Entisols. The pXRF reported elements helped in the characterization of the BCP soils. In general, the RF model achieved the best performances for TN (R2 = 0.50), CEC (0.75), and SOM (0.56) when A and B horizons were combined, although better results have been reported in the literature for soils from other regions of the world. The results reported here for the BCP soils represent alternatives for reducing costs and time needed for assessing such data, supporting agronomic and environmental strategies.