Performance evaluation of reflectance spectroscopy for estimation of soil organic carbon content in Zrebar lake watershed, Kurdistan province

Soil organic carbon (SOC), as a great constitute of soil organic matter (SOM), has an important role in chemical, physical and biological processes of soil. SOM or SOC is a key parameter of soil quality and a soil fertility indicator. SOM has an essential role in formation of soil aggregate and its stability, water and nutrients adsorption, water holding capacity, infiltration of air and water, hydraulic conductivity, soil water repellency and carbon sequestration. Various studies have shown that the quantity and quality of SOM can be affected by anthropogenic activities such as farming practices and other economic development activities. It has also been reported a high rate of SOM loss on eroded lands. Hence, monitoring temporal and spatial variation of SOM is essential for evaluating long-term soil productivity management. However, conventional soil sampling and chemical measurement of SOC, especially in large geographic scale, is tedious, time consuming and expensive. Therefore, rapid and precise assessment of SOC content can be useful in long-term management of soil. The objective of this study was to investigate the ability of soil visible-near infrared (Vis-NIR) spectroscopy for estimating SOC in Zrebar lake watershed of Marivan, Kurdistan province, Iran.

A total of 100 soil samples were collected from the studied region, with an area about 10718 hectares. The spectral reflectance and physicochemical properties of all soil samples were measured under laboratory controlled conditions. After recording of the spectra, different pre-processing methods were applied and compared. Then, pedo-transfer functions (PTFs) and specto-transfer functions (STFs) were developed to estimate SOC content using stepwise multiple linear regression (SMLR). The accuracy and reliability of the derived PTFs and STFs were evaluated using coefficient of determination (R2), normalized root mean square error (NRMSE), mean error (ME), index of agreement (d), and ratio of performance to deviation (RPD) statistics.

Based on the results, soil organic carbon showed high and significant (significance level of 1%) correlations with spectral reflectance values at wavelengths 858 and 1916 nm. The results indicated that the derived PTFs had the higher accuracy (R2avg=0.83, NRMSEavg = 24.55%) to estimate SOC in comparison with the STFs (R2avg=0.44, NRMSEavg= 44.31%). However, SOC could be also fairly estimated by the derived specto-transfer functions (Ravg2=0.52, RPDavg= 1.44).The results also revealed that the Savitzki–Golay smoothing filter with 1st order derivative was the best spectral pre-processing method to reduce the effect of random noise and improve the calibration models.

Overall, the results indicated that although the performance of STFs was not superior to the corresponding PTFs for estimating SOC, but this approach can be used as a reasonable indirect method in case of unavailability of PTFs.