Soil is the largest carbon pool in terrestrial ecosystems. Soil organic carbon (SOC) and total nitrogen (TN) play major roles in soil quality and the global carbon budget.  

SOC and TN could be measured via visible and near-infrared reflectance (VNIR) spectroscopy, but its reliability is still challenged by the influence of heterogeneity. 

JIANG Qinghu, supervised by Prof. LIU Feng from Wuhan Botanical Garden of the Chinese Academy of Sciences, conducted a study in Badagongshan in central China to estimate SOC and TN concentrations in highly heterogeneous soils from forest ecosystems using VNIR spectroscopy, and explored the possibility of using spiking to improve spectroscopic model applicability. 

The results of initial models (global and layered models) showed successful prediction ability for SOC [R²_P= 0.79 - 0.90, ration of performance to inter-quartile range (RPIQ) = 3.07 - 3.97, the root mean square error (RMSE_P) = 0.54% - 0.88% ] and TN (R²_P= 0.66 - 0.86, RPIQ = 2.12 - 3.78, RMSE_P = 0.05% - 0.08%).

Besides, spiking improved the applicability of the initial models (RMSE_P and absolute prediction bias were obviously reduced) and the accuracy was further improved when the spiking subset was extra-weighted. 

This paper highlighted the potential application of VNIR spectroscopy as a reliable tool to quantify SOC and TN concentrations in mid-subtropical forest soils.  

Spiking alone and spiking with extra-weighing were effective approaches to improve model applicability in the VNIR estimation of SOC and TN between different soil layers. 

This approach is potentially useful in rapidly quantifying and monitoring soil carbon and nitrogen pools in heterogeneous landscapes. 

Results were published in Geoderma entitled “Estimation of soil organic carbon and total nitrogen in different soil layers using VNIR spectroscopy: Effects of spiking on model applicability”.  

This study was funded by the Chinese National Key Development Program for Basic Research and the Natural Science Foundation of China. 

Values of RMSE and bias for (a) SOC and (b) TN estimation using Partial least squares regression (PLSR)with the initial (no-spiked) calibrations and spiked calibrations (without and with extra-weight) models (Image by JIANG Qinghu)