Transferability of Retrieval Models for Estimating Soil Organic Carbon Contents Based on Regional Soil Spectral Libraries

Assessing soil organic carbon (SOC) effectively is crucial to understand the global carbon cycle and achieve sustainable management of agricultural systems. Visible/near-infrared spectroscopy has been widely used to retrieve SOC content. However, retrieval models with visible/near-infrared spectroscopy generally have regional limitations. The objective of our work was to study the transferability of models between regions using the spectra and SOC measurements of the soil samples collected from Zhongxiang and Honghu(number of samples are 100 and 96 respectively), Hubei province, Peoples Republic of China. Our results show that the regional model calibrated with Zhongxiang or Honghu datasets could not be used in another region. However, the model calibrated with the entire Zhongxiang dataset and 30 samples from Honghu performed well when estimating the SOC contents in Honghu (R²=0.88, RMSE=2.51 g·kg^-1). Although the transferability of this regional model is very limited, nevertheless, this study illustrates that by combining a small set of samples from target region with the regional soil spectral library of anther region can improve the performance when estimating SOC contents in target region, and thus can reduce the cost of sampling,measurement, and determination in a target region.