Evaluating spectro-transfer functions to estimate soil organic carbon stock at large spatial scales

Soil organic carbon (SOC) stock is critical in mitigating global warming by sequestering carbon and enhancing soil fertility. This study focuses on Campania, a region of southern Italy covering about 13,700 km², and addresses the challenging task of estimating SOC stock at relatively large spatial scales in a sustainable manner. A practical outcome is to provide public bodies and stakeholders with as many reliable SOC stock maps as possible, allowing for the uncertainties associated with the techniques employed. The assessment of SOC stock requires the knowledge of SOC content, oven-dry bulk density, soil depth, and rock fragment.

To accomplish this task, the following soil physical and chemical properties were directly measured by collecting 3,316 soil samples: particle-size distribution, soil textural classes (i.e., the sand, silt, and clay contents), oven-dry soil bulk density, soil organic content, pH, and calcium carbonate. However, direct measurements of SOC content and especially oven-dry soil bulk density are labor-demanding, time-consuming, and expensive. Therefore, we explored the use of soil spectroscopy in the visible, near-infrared, and shortwave infrared (vis-NIR-SWIR in the range 400-2500 nm) range to estimate these input properties. The spectral reflectance in the vis-NIR-SWIR range was measured on co-located 3,316 air-dried soil samples, sieved at 2 mm, whereas Spectro-Transfer Functions (STFs) have been developed to predict the SOC stock using advanced statistical methods, including neural networks, partial least square regression, and linear/nonlinear regression models.

Our findings demonstrate the superior performance of neural networks and partial least square regression in accurately estimating SOC stocks. However, we also emphasize the value of simpler linear/nonlinear regression models for their reproducibility and ease of implementation. These results highlight the potential of spectral-based approaches to estimate SOC stocks at large scales efficiently and cost-effectively, thereby improving the implementation of carbon management strategies and enhancing the assessment of agroecosystem resilience to global warming.