This study aimed to investigate the feasibility of using imaging spectroscopy (IS) to predict soil organic carbon density (SOCD) either directly or indirectly through soil organic carbon (SOC). Three methods, partial least square regression (PLSR), support vector machine (SVM), and random forest (RF), were utilized to calibrate the models and map the SOCD. The results showed that direct prediction was better than indirect prediction and the best model SVM had high R 2 of 0.94 and 0.93 for calibration and validation, respectively. The measured SOCD was mainly concentrated in the surface soil layer from 0 to 40 cm, and the deeper layer tended to be gentle. The continuous depth variation trend of SOCD in the topsoil (0-40 cm) was relatively close to the measured values, while in the deeper layers (below 40 cm), it was much higher than the measured values. The study also found that the best fitting function of measured SOC stocks over time varied from linear to power and then to logarithmic with increasing depth, indicating less efficient accumulation of SOC in deep soil compared to topsoil, resulting in an overall decrease in SOC accumulation rate across soil depth. The best temporal functions for the predicted values differed from the measured values at each depth, but the changing trends of the three functions were basically consistent. It suggests that IS technology has the potential to quantitatively reveal the process of coastal soil evolution and offers a new insight for the rapid monitoring of soil property changes.

No abstract