Reflectance spectroscopy and remote sensing for prediction of soil organic carbon and total nitrogen across different agricultural contexts

Abstract

Techniques for monitoring soil attributes have rapidly evolved, driven by the demand for more efficient, sustainable, and large-scale applicable methods. Soil organic carbon (SOC) and total nitrogen (TN) are essential components for ecosystem functioning. Although traditional laboratory methods are precise, they require considerable time, resources, and labor. Reflectance spectroscopy combined with statistical modeling emerges as a promising tool, allowing estimation of soil attributes from spectral data. This study evaluated the predictive performance of reflectance spectroscopy through three approaches: i) using the hyperspectral satellite EnMAP; ii) with the proximal ASD FieldSpec® sensor directly in the soil surface; and iii) in the laboratory with dried soil samples. The research was conducted in two regions: the Lower Acaraú Irrigation District (R1) in northeastern Brazil, and California’s Central Valley (R2), USA. Surface soil samples were collected from the first centimeter in R1 and from 0–10 cm in R2. Laboratory spectral analysis was performed on both sets in darkroom after the oven drying. SOC and TN quantification was done by titration in R1 and by elemental analyzer in R2. Predictive models were developed for each regional set and for a combined global set (R1&R2), using raw reflectance, first derivative, and continuum-removed spectra. The predictive performance of EnMAP and in situ FieldSpec sensor was assessed exclusively in R1, using images obtained by EnMAP. Near the imaging dates, spectral analysis with FieldSpec was conducted at the point corresponding to the central coordinate of the pixel representing exposed soil in the EnMAP image prior to sampling, enabling direct comparison between spectral responses from the orbital and proximal sensors. Results indicate that the EnMAP satellite showed good correlations with spectral data collected by multispectral sensors, with R² between 0.50 and 0.86 for individual bands and NDVI. SOC and TN prediction by EnMAP was comparable to that of the FieldSpec sensor in the field, with adj. R² = 0.75 for SOC and 0.61 for TN. Accuracy improved for SOC > 11.6 g/kg and TN > 0.55 g/kg, reaching adj. R² = 0.91 and 0.82, respectively. Statistical modeling confirmed the efficiency of reflectance spectroscopy to predict SOC and TN in both local contexts and the global model, regardless of the laboratory methods used. The first derivative transformation was most effective in producing reliable models, especially for R2 using PLSR, with adj. R² = 0.91 and 0.92 for SOC and RPD of 3.34 and 3.49 for TN. The global R1&R2 set also showed good predictive performance with PLSR (adj. R² = 0.82 and 0.90; RPD = 2.09 and 2.86 for SOC and TN, respectively). These results reinforce the potential of reflectance spectroscopy and hyperspectral remote sensing for precise and efficient monitoring of organic soil attributes, offering viable alternatives for diverse agricultural realities and promoting advances in sustainable soil management practices. Keywords: hyperspectral imaging; organic attribute monitoring; multivariate modeling