VNIR, XRF, and LIBS spectroscopies for soil sensing on precision agriculture

• Autor(a) principal: Tavares, Tiago Rodrigues
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/11/11152/tde-15092021-114515/
• topic: Análise da fertilidade do solo; Green chemistry; Hybrid laboratories; Laboratórios híbridos; Química verde; Sistemas sensores inteligentes; Smart sensor systems; Soil fertility testing

Visible and near infrared diffuse reflectance spectroscopy (VNIR), X-ray fluorescence spectroscopy (XRF), and laser-induced breakdown spectroscopy (LIBS), are promising techniques for rapid and environmentally-friendly soil fertility characterization. Integrating these sensing tools with the already established analytical methods can enable advances towards the modernization of the traditional analytical procedures by allowing, e.g. , in-situ and mobile laboratory analysis, which, in turn, will enable new paradigm in soil management using precision agriculture approaches. The main objective of this thesis is to develop strategies for using the output generated by VNIR, XRF, and LIBS sensors in the calibration of agronomic algorithms for predicting key soil fertility attributes [clay, organic matter (OM), cation exchange capacity (CEC), pH, base saturation (V), and extractable (ex-) nutrients (ex-P, ex-K, ex-Ca, and ex-Mg)]. A set of 102 soil samples collected from two Brazilian agricultural fields was used. In the first stages, studies were conducted to (i) simplify the soil sample preparation for XRF sensor analysis, (ii) develop a simple and transparent method for XRF data acquisition and processing, and (iii) find an accurate and efficient method for LIBS data modeling. Afterwards, the individual and combined prediction performances of VNIR, XRF, and LIBS were compared. The results showed that overall the predictive performance decreased as follows: LIBS > XRF > VNIR. VNIR confirmed to be the best technique for the prediction of clay and OM content [2.61 ≤ residual prediction deviation (RPD) ≤ 3.37], while the chemical attributes CEC, V, ex-P, ex-K, ex-Ca, and ex-Mg were better predicted (1.82 ≤ RPD ≤ 4.82) by elemental analysis techniques. Regarding multi-sensor approaches, the prediction quality decreased in the order of: VNIR + XRF + LIBS > XRF + LIBS > VNIR + LIBS > VNIR + XRF. Our results indicate that there is no unique optimal sensor combination for predicting all the key soil fertility attributes and that this is attribute specific. However, we also consider that it is too soon to establish a clear trend, which requires further research evaluating larger data sets including other soil types, with greater variability in soil mineralogy, textural classes, attributes concentration range, and submitted to different agricultural practices. Lastly, we evaluated the potential of applying XRF for in-situ analysis, focusing on the effect of both soil moisture content and scanning time on the sensors\' performance. Our results proved that it is possible to make drastic reductions in scanning time (from 90 to 2s) maintaining satisfactory performances (RPD ≥ 1.92) for predicting fertility attributes. Moreover, we showed that although XRF is less sensitive to soil moisture increment when compared to VNIR sensor, the presence of water impacts on the XRF\'s predictive performance and methods for mitigating soil moisture effect should be applied aiming at more accurate in-situ analysis. This thesis brought some advances for the application of VNIR, XRF, and LIBS sensors in Brazilian tropical soils as fast and clean analytical methods for characterizing fertility attributes. The advantages and drawbacks of these techniques were highlighted and directions were left for future research that will continue the technological maturation toward the modernization of soil fertility diagnostics.