Temporal evaluation of soil chemical quality using VNIR and XRF spectroscopies

Abstract

New soil sensing technologies, such as visible and near-infrared diffuse reflectance spectroscopy (VNIR) and X-ray fluorescence spectroscopy (XRF), emerged as alternatives to low-cost, rapid, and environmental-friendly soil analyses. Studies involving the prediction of integrated soil quality indices using individual and combined soil sensing techniques have not been conducted yet on tropical soils, especially regarding the temporal stability of models' performance. This study evaluated tropical soils to give insight into the potential of using individual and combined VNIR and XRF sensors to predict the Soil Chemical Quality Index (SCQI) and its sub-functions (Nutrient Availability, Acidity/Al Toxicity, and Nutrient Storage and Cycling). Furthermore, the temporal stability of these models' performance and a spiking strategy to mitigate performance loss throughout crop seasons were assessed. We conducted this study using 924 soil samples collected over three years (2018, 2019, and 2022) in a field managed with an integrated crop-livestock system (200 ha) in a Brazilian tropical area. We determined the SCQI and its sub-functions for each sample from 15 soil quality indicators. Soil samples were scanned with VNIR (350–2500 nm) and XRF (0–35 kV) spectrometers. VNIR spectroscopy predicted the SCQI, Nutrient Availability, and Nutrient Storage and Cycling with satisfactory accuracy (models classified as good, good, and excellent, respectively), outperforming the XRF technique and the fusion approach. The prediction performance of SCQI and its sub-functions does not show temporal stability. However, the performance loss can be mitigated using the spiking strategy, i.e., using new data from subsequent seasons to recalibrate the global model.