Multiple linear regression and random forest to predict and map soil properties using data from portable X-ray fluorescence spectrometer (pXRF)

Abstract

Determination of soil properties helps in the correct management of soil fertility. The portable X-ray fluorescence spectrometer (pXRF) has been recently adopted to determine total chemical element contents in soils, allowing soil property inferences. However, these studies are still scarce in Brazil and other countries. The objectives of this work were to predict soil properties using pXRF data, comparing stepwise multiple linear regression (SMLR) and random forest (RF) methods, as well as mapping and validating soil properties. 120 soil samples were collected at three depths and submitted to laboratory analyses. pXRF was used in the samples and total element contents were determined. From pXRF data, SMLR and RF were used to predict soil laboratory results, reflecting soil properties, and the models were validated. The best method was used to spatialize soil properties. Using SMLR, models had high values of R² (≥0.8), however the highest accuracy was obtained in RF modeling. Exchangeable Ca, Al, Mg, potential and effective cation exchange capacity, soil organic matter, pH, and base saturation had adequate adjustment and accurate predictions with RF. Eight out of the 10 soil properties predicted by RF using pXRF data had CaO as the most important variable helping predictions, followed by P 2 O 5 , Zn and Cr. Maps generated using RF from pXRF data had high accuracy for six soil properties, reaching R 2 up to 0.83. pXRF in association with RF can be used to predict soil properties with high accuracy at low cost and time, besides providing variables aiding digital soil mapping.Index terms: Soil analyses; spatial prediction; proximal sensor. RESUMOA determinação de atributos do solo auxilia no correto manejo da sua fertilidade. O equipamento portátil de fluorescência de raios-X (pXRF) foi recentemente adotado para determinar o teor total de elementos químicos em solos, permitindo inferências sobre atributos do solo. No entanto, esses estudos ainda são escassos no Brasil e em outros países. Os objetivos deste trabalho foram prever atributos do solo a partir de dados do pXRF, comparando-se os métodos de regressão linear múltipla stepwise (SMLR) e de random forest (RF), além de mapear e validar atributos do solo. 120 amostras de solo foram coletadas em três profundidades e submetidas a análises laboratoriais. Utilizou-se o pXRF para leitura das amostras e determinou-se o teor total de elementos. A partir dos dados do pXRF, foram utilizadas SMLR e RF para predizer resultados laboratoriais, que refletem atributos do solo, e os modelos foram validados. O melhor método foi utilizado para espacializar os atributos do solo. Utilizando SMLR, os modelos apresentaram valores elevados de R² (≥0,8), porém maior acurácia foi obtida na modelagem com RF. A capacidade de troca de cátions potencial e efetiva, matéria orgânica do solo, pH, saturação por bases e teores trocáveis de Ca, Al e Mg apresentaram ajustes adequados e predições acuradas com RF. Dos dez atributos do solo preditos por RF a partir de dados do pXRF, sete apr...

Tropical Soil Toposequence Characterization via pXRF Spectrometry

“…This fact is also corroborated by the lack of Ca‐bearing minerals in quartzite‐derived soils (Araujo et al, 2014). Similar results were found in Inceptisols of the Brazilian Cerrado developed from phyllite (Teixeira et al, 2018; Silva et al, 2017).…”

“…This fact is also corroborated by the lack of Ca-bearing minerals in quartzite-derived soils (Araujo et al, 2014). Similar results were found in Inceptisols of the Brazilian Cerrado developed from phyllite (Teixeira et al, 2018;Silva et al, 2017). Since Y accumulates during weathering processes, a positive correlation with clay was observed in soils developed from parent materials other than quartzite (Marques et al, 2004b).…”

“…The correlations between elemental contents obtained by pXRF and soil properties provide new insights of relationships (positive or negative correlation) previously less explored, mainly in quartzite‐derived soils, such as: Rb, Zn, K 2 O with silt content; Sr and Ti with soil fertility properties; Zn with fine sand content; and Zr and clay content. Others already reported in Brazilian soils (Silva et al, 2017, 2018b; Teixeira et al, 2018) include: CaO and exchangeable Ca 2+ , Mg 2+ , Al 3+ , pH, and other soil properties; SiO 2 and sand (positive) and clay (negative) content; and Al 2 O 3 with clay content. The relationships between the elemental contents obtained by pXRF may contribute to studies focusing on the understanding of the geochemical behavior of some elements, such as Ca with Sr, and K and Zn with Rb.…”

“…Thus, this equipment allows for fast, low cost, environmentally friendly, and in‐field identification and quantification of several elements from the periodic table (Weindorf et al, 2014; Stockmann et al, 2016; Ribeiro et al, 2017). It has proven useful for investigations related to soil genesis and horizon variability (Stockmann et al, 2016; Weindorf et al, 2012a, 2012b; Silva et al, 2018a, Sharma et al, 2014, 2015; Zhu et al, 2011), parent material spatial variability (Mancini et al, 2019), soil property prediction and mapping (Silva et al, 2017; Duda et al, 2017; Pelegrino et al, 2018), and for other pedological and environmental studies (Rouillon and Taylor, 2016; Silva et al, 2018b; Suh et al, 2016).…”

Land-use influence on the soil hydrology: An approach in upper Grande River basin, Southeast Brazil

“…Random Forest (RF) model is a non-parametric technique that has been successfully applied to soil properties prediction (Wiesmeier et al, 2011;Castro Franco et al, 2015;Hengl et al, 2015;Chagas et al, 2016;Yang et al, 2016;Dharumarajan;Hedge;Singh, 2017;Silva et al, 2017;Blanco et al, 2018;Wang et al, 2018a). The model combines a set of decision trees to improve the accuracy of prediction of a given environmental variable, where each tree is generated by bootstrap samples (random sampling with substitution), leaving one-third of training samples, called Out-of-Bag (OOB) data, for using in the model's performance evaluation (Breiman, 2001;Liaw;Wiener, 2002).…”