Amplifying the power of proximal sensing techniques to assess the cadmium concentration in agricultural soils : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Soil Science at Massey University, Palmerston North, New Zealand

Thumbnail Image
Open Access Location
Journal Title
Journal ISSN
Volume Title
Massey University
The Author
Cadmium (Cd) accumulation in agricultural soils due to long-term phosphate fertiliser applications has raised concerns in New Zealand and globally due to the potential toxicity of Cd in food products. Elevated soil Cd concentration can enhance Cd availability for plant uptake, increasing the risk of food chain transfer. Cadmium management is generally achieved through reference laboratory methods to estimate Cd concentration in soil and plant samples. Reference laboratory methods of Cd analysis are precise; however, sample preparation and associated resource cost make them expensive. As a complementary method, proximal sensing techniques including visible-near-infrared (vis-NIR: 350–2500 nm), mid-infrared (MIR: 4000–400 cm-1) reflectance and portable X-ray fluorescence (pXRF: 0–40 keV) spectroscopy have been successfully used to monitor elevated Cd levels in mining areas and in plants showing stress or toxicity symptoms due to Cd. However, application of such technologies in agricultural soils with low Cd concentration are relatively understudied. Hence, this study was conducted to amplify the power of three proximal sensing techniques to quantify Cd in soil samples from diverse soil orders, climatic conditions, land uses, and vegetations and plant samples for cost-effective Cd monitoring at regional to farm scale. In this doctoral study, soil and plant samples were scanned using vis-NIR, MIR, and pXRF sensors. Topsoil samples were obtained from (1) the Otago-Southland regional survey (n=622), (2) a pastoral farm survey (n=87) including dairy and sheep and beef farms with long-term phosphate fertiliser application history, and (3) two independent glasshouse experiments using Pallic and Allophanic soils amended with increasing soil Cd concentrations, and with or without a model forage herb, chicory (Cichorium intybus L.). In both experiments, chicory aboveground biomass and root samples were scanned using the three sensors, along with a periodic collection of vis-NIR spectra from soil and plant in-situ. Total Cd was determined in all samples, while the distribution of Cd among geochemical fractions was studied in the pastoral farm survey samples only. Reference laboratory results and spectral information were combined to develop models for accurate Cd predictions. For regional survey samples (n=622, 0.01–0.56 mg Cd/kg) including agricultural soils (47%), validation (v) results (n=124, 0.01–0.43 mg Cd/kg) showed Granger-Ramanathan model averaging of outputs from models using individual pXRF, vis-NIR, and MIR data as input for partial least squares (PLS) – support vector machine regression performed optimally to quantify total soil Cd with normalised root mean square error (nRMSEv) of 37% and concordance correlation coefficient (CCCv) of 0.84. For agricultural soils (n=84, 0.10–1.20 mg Cd/kg), cross-validation (cv) results of models using individual vis-NIR, MIR, and pXRF data as input for PLS performed with nRMSEcv of 26%, 30%, and 31% and CCCcv of 0.85, 0.77, and 0.75 respectively. For acid soluble (0.01–0.27 mg Cd/kg) and organic matter bound (0.02–0.27 mg Cd/kg) Cd, models using vis-NIR data performed with nRMSEcv of 11% and 33% and CCCcv of 0.97 and 0.84, respectively. For exchangeable (0.003–0.25 mg Cd/kg) Cd, a model using MIR data as input performed with nRMSEcv of 40% and CCCcv of 0.57. Using the Otago and Southland regional survey soil samples spectra as a soil spectral library (SSL), Cd concentration in the local set (agricultural soil samples) were quantified. A model using MIR data from the regional SSL pastoral soil subset (n=283, 0.01–1.31 mg Cd/kg) spiked with selected local set samples (n=12) with weights (×4) as input for LOCAL algorithm quantified local soil Cd with nRMSE of 38% and CCC of 0.78. In the glasshouse experiments, Cd translocation factor (TF) values for chicory were calculated using proximal sensor data and the results showed a significant relationship (R2=0.74, p<0.001) between measured and predicted TF values. A model using in-situ leaf clip vis-NIR spectra showed optimal performance to assess Cd concentration in aboveground chicory biomass with nRMSEcv of 28% and CCCcv of 0.93. Among vegetation indices calculated ‘blue green index 2’ showed a significant (p<0.01) R2 value (0.19, 0.36) in both experiments. Models using pXRF spectra as input showed optimal performance to predict chicory root (n=28, 0.86–25.79 mg Cd/kg) and Allophanic soil (n=112, 0.41–4.81 mg Cd/kg) Cd with nRMSEcv of 16% and 9% and CCCcv of 0.95 and 0.99, respectively. A model using laboratory vis-NIR spectra showed optimal performance to quantify Pallic soil Cd (n=336; 0.17–5.45 mg Cd/kg) with nRMSEcv of 22% and CCCcv of 0.97. Optimal prediction models using proximal sensor data can potentially be used for rapid cost-effective analysis of Cd concentration in soil and plant samples. Quantitative models for soil Cd using a combination of complementary proximal sensors data and chemometrics could feasibly be deployed for long-term monitoring of soil Cd at concentrations below pXRF detection limits and with reduced matrix interference from organic matter when compared to the individual techniques alone. The use of proximal sensing techniques to determine total soil Cd concentration in New Zealand agricultural soils has the potential to improve the scale and scope of long-term repeated monitoring of soil Cd concentration required under the framework of the national Tiered Fertiliser Management System. Reflectance spectroscopy could potentially be implemented to monitor plant-available and potentially-available soil Cd fractions to minimise plant Cd uptake. The use of a large soil spectral library to assess the local Cd concentration in agricultural soils could reduce the analytical cost to the farmers and allow intensive spatial and temporal monitoring of pastoral farms based on spectral analysis only. The use of in-situ and laboratory proximal sensor data to calculate bioconcentration and translocation factors could potentially support the evaluation of Cd food chain transfer risks. The spectral library developed from this doctoral study, including soil and plant root and aboveground biomass pXRF, vis-NIR, and MIR spectra with a wide range of Cd concentration can be used as reference materials for field level and airborne remote sensing studies.
Soils, Cadmium content, Measurement, Reflectance spectroscopy, Fluorescence spectroscopy, New Zealand, X-ray spectroscopy