TY - JOUR
T1 - Influence of auxiliary soil variables to improve PXRF-based soil fertility evaluation in India
AU - Dasgupta, Shubhadip
AU - Chakraborty, Somsubhra
AU - Weindorf, David C.
AU - Li, Bin
AU - Silva, Sérgio Henrique Godinho
AU - Bhattacharyya, Kallol
N1 - Publisher Copyright:
© 2022 Elsevier B.V.
PY - 2022/9
Y1 - 2022/9
N2 - Portable X-ray fluorescence (PXRF)spectrometry has already been established as a rapid and cost-effective tool for predicting various soil physicochemical properties. This study used PXRF in combination with physiographic, agro-climatic, soil parent-material, and physicochemical attributes (pH, electrical conductivity (EC), loss on ignition organic matter, and organic carbon) as auxiliary properties to predict multiple soil fertility indicators [available K, Ca, Mg, Fe, Cu, Zn, Mn, B, K/Mg ratio, total exchangeable bases (TEB), and sulfur availability index (SAI)] via four machine-learning algorithms (random forest, support vector regression, stepwise multiple linear regression, and an averaged model). Principal component analysis (PCA) indicated the links between PXRF-reported elements, agro-climatic zones, and soil parent materials. Although no universal prediction model can be selected to suit all 11 soil fertility parameters, three parameters (available Ca, Fe, and TEB) produced reasonable model performance with an R2 > 0.50 for most prediction model-dataset combinations. Concatenation of auxiliary soil parameters with PXRF data showed relative improvement in model accuracy compared to PXRF in isolation. Notably, the agro-climatic zone appeared influential for predicting available K, Mg, Zn, Fe, Mn, B, K/Mg ratio, and TEB. For potential fertilizer recommendation, six parameters (available K, Ca, Mg, Cu, Mn, and B) produced reasonable classification performance via the averaged model using all auxiliary predictors (κ > 0.30). The same categorical model was used, as an instance, for delineating a conceptualized framework for (PXRF+ auxiliary properties)-based fertilizer recommendation facilitating site-specific nutrient management. More research is needed to enhance model prediction/classification accuracy by including a well-balanced dataset and other relevant auxiliary variables with PXRF. Nevertheless, the importance of adding auxiliary soil properties with PXRF elemental data for cost-effective and accessible nutrient management in resource-poor countries seems promising.
AB - Portable X-ray fluorescence (PXRF)spectrometry has already been established as a rapid and cost-effective tool for predicting various soil physicochemical properties. This study used PXRF in combination with physiographic, agro-climatic, soil parent-material, and physicochemical attributes (pH, electrical conductivity (EC), loss on ignition organic matter, and organic carbon) as auxiliary properties to predict multiple soil fertility indicators [available K, Ca, Mg, Fe, Cu, Zn, Mn, B, K/Mg ratio, total exchangeable bases (TEB), and sulfur availability index (SAI)] via four machine-learning algorithms (random forest, support vector regression, stepwise multiple linear regression, and an averaged model). Principal component analysis (PCA) indicated the links between PXRF-reported elements, agro-climatic zones, and soil parent materials. Although no universal prediction model can be selected to suit all 11 soil fertility parameters, three parameters (available Ca, Fe, and TEB) produced reasonable model performance with an R2 > 0.50 for most prediction model-dataset combinations. Concatenation of auxiliary soil parameters with PXRF data showed relative improvement in model accuracy compared to PXRF in isolation. Notably, the agro-climatic zone appeared influential for predicting available K, Mg, Zn, Fe, Mn, B, K/Mg ratio, and TEB. For potential fertilizer recommendation, six parameters (available K, Ca, Mg, Cu, Mn, and B) produced reasonable classification performance via the averaged model using all auxiliary predictors (κ > 0.30). The same categorical model was used, as an instance, for delineating a conceptualized framework for (PXRF+ auxiliary properties)-based fertilizer recommendation facilitating site-specific nutrient management. More research is needed to enhance model prediction/classification accuracy by including a well-balanced dataset and other relevant auxiliary variables with PXRF. Nevertheless, the importance of adding auxiliary soil properties with PXRF elemental data for cost-effective and accessible nutrient management in resource-poor countries seems promising.
KW - Classification
KW - Entisols
KW - Fertilizer recommendation
KW - Inceptisols
KW - PXRF
KW - Random forest
KW - Soil fertility
KW - STCR
KW - Support vector regression
UR - http://www.scopus.com/inward/record.url?scp=85134490541&partnerID=8YFLogxK
U2 - 10.1016/j.geodrs.2022.e00557
DO - 10.1016/j.geodrs.2022.e00557
M3 - Article
AN - SCOPUS:85134490541
SN - 2352-0094
VL - 30
JO - Geoderma Regional
JF - Geoderma Regional
M1 - e00557
ER -