Mapping cation exchange capacity and exchangeable potassium using proximal soil sensing data at the multiple-field scale

Authors: FUNG, EVANGELINE - University Of New South Wales; WANG, JIE - University Of New South Wales; ZHAO, XUEYU - University Of New South Wales; FARZAMIAN, MOHAMMAD - Instituto Nacional De Investigação Agrária E Veterinária; Allred, Barry; TRIANTAFILIS, JOHN - Manaaki Whenua Landcare Research

Submitted to: Soil & Tillage Research
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/19/2023
Publication Date: 5/9/2023
Citation: Fung, E., Wang, J., Zhao, X., Farzamian, M., Allred, B.J., Triantafilis, J. 2023. Mapping cation exchange capacity and exchangeable potassium using proximal soil sensing data at the multiple-field scale. Soil & Tillage Research. 232: Article #105735. https://doi.org/10.1016/j.still.2023.105735.
DOI: https://doi.org/10.1016/j.still.2023.105735

Interpretive Summary: In four small fields near Defiance, Ohio, we found that the best LR between Veris3100 ECa for topsoil CEC was with shallow ECa (R2 = 0.38), and subsoil CEC with deep ECa (0.36). However, both were poor (0.5 > R2 > 0.3) and unsatisfactory for prediction. Nevertheless, by inverting the Veris3100 shallow and deep ECa, the best LR between s and CEC was strong (0.75) and achieved when ' was estimated using S2 inversion algorithm (EM4Soil) using a damping factor (') = 1. In terms of minimum calibration sample size, it was found that while R2 was strong (0.83) for as few as n = 5 calibration sites, prediction agreement was substantial (Lin’s > 0.8) only when n = 10 or more sites were used for calibration. We conclude that 10 calibration sample sites would be satisfactory to predict the study field at both depths with moderate uncertainties (95% CI < 5). Similarly, we found that the best LR between Veris3100 ECa and topsoil K was larger than CEC, it was still weak with shallow ECa (R2 = 0.47). Nevertheless, we developed a MLR with various sources of digital data including shallow ECa and elevation and adjusting for linear drift in predicted response of first-order trend surface components (i.e., Easting and Northing). In terms of minimum calibration sample size, it was found that the minimum number required was n = 15 because of the mostly strong R2 (i.e., > 0.7) and its equivalence (i.e., 0.68-0.75) regardless of sample size (n = 45 – 10) with no loss in prediction agreement using n = 15 (i.e., LCCC = 0.60) compared to n = 45 – 20 (i.e., 0.56 – 0.63). To improve areal prediction agreement of CEC and K and reduce CI across the four fields, we recommend the use of tighter transect spacings (< 6m), and inclusion of soil (i.e., small CEC and K) and digital (e.g., small ECa) data from adjacent fields and on nearby farms to improve calibration equations. Nevertheless, the final DSM of CEC and K enabled the enactment of Ohio and Indiana Potash (K2O) recommendations for corn at various yield potentials (Culman et.al, 2020). Future work should explore the universal potential of applying these calibrations to adjacent fields and on nearby farms.

Technical Abstract: The cation exchange capacity (CEC – cmol (+) kg-1) is the capacity of a soil to hold exchangeable cations, such as exchangeable potassium (K). Knowledge of these properties is often used to provide recommendations for fertilizers; however, they are expensive to measure in the laboratory. To create maps, proximal sensed instruments can be used. In this research, we explore the potential to develop a linear regression (LR) between apparent soil electrical conductivity (ECa - mS/m) from Veris-3100 shallow (0–0.3 m) and deep (0–0.9 m) array configurations and measured topsoil (0–0.2 m) and subsoil (0.5–0.7 m) CEC. We compare these with a LR between estimates of ' from inversion of Veris-3100 ECa using a quasi-three-dimensional algorithm (invVeris V1.1). We also determine the minimum number of calibration sample sites (i.e., 45, 40, 35, 5) required to produce a strong LR (i.e., coefficient of determination: R2 > 0.7) and substantial (Lin’s concordance correlation coefficient; LCCC > 0.8) or consistent prediction agreement. We do this by dividing the n = 60 sample sites into calibration (n = 45) and validation (n = 15) sets. A similar approach is used to develop a multiple-LR (MLR) to predict topsoil K considering digital data (i.e., ECa, elevation and trend surface parameters). The LR between topsoil CEC and shallow ECa (R2 = 0.38) and subsoil CEC with deep ECa (0.36) were poor (0.5 > R2 > 0.3). However, LR (0.75) between s and CEC was strong (using S2 algorithm and a damping factor ['] = 1). A poor LR (0.47) was also found between K and ECa, however, a MLR model (ECa, elevation and trend surface parameters) was strong (0.73). In terms of minimum calibration sample size for CEC, it was found n = 10 sites (i.e., at 2 depths) or more were required. With respect to K, the minimum calibration sample size was n = 10 (i.e., single depth). To improve areal prediction agreement of CEC and K and reduce CI across the four fields, we recommend the use of tighter transect spacings (< 6m), and inclusion of soil (i.e., small CEC and K) and digital (e.g., small ECa) data from adjacent fields and on nearby farms to improve calibration equations. The final DSM of CEC and K could be used to prescribe potash (K2O) fertilizers.