Prediction of Soil Organic Carbon in Arid Regions Using Hyperspectral Spectroscopy: UAE Case Study

This study adopted the spectral analysis method that has become more popular with undeniable advantages over traditional methods to estimate the Soil Organic Carbon (SOC). A total of 59 soil samples and 18 fertilized irrigation water samples were collected from the agricultural fields in the United Arab Emirates (UAE) and analyzed in a laboratory for testing the Dissolved Organic Carbon (DOC) and SOC using Walkley and Black and Sievers InnovOx analyzers, respectively. Spectroscopic measurements for these soil and water samples were collected under stable laboratory conditions using two instruments to cover the full range of wavelength: Ocean Insight HR2000+ (400 to 950nm) and ARCoptix (900 to 2600nm). The objective of this study is to use these spectral reflectance measurements to develop models that can predict both SOC and DOC. Therefore, the raw spectra of 2902 bands were preprocessed to ensure an optimal correlation and noise removal and the Partial Least Squares Regression (PLSR) method was applied afterwards with leave-one-out cross-validation approach. The resultant best fit PLSR model has shown a coefficient of determination (R ² ) of 0.99 for training and 0.94-0.98 for prediction of SOC and DOC, respectively, for the SOC range of 0.46-3.15% (mean: 1.5%) and DOC range of 0.18-$10,333 \mu \mathrm{g}$ (mean: 1,620 $\mu \mathrm{g}$). The spectral signal of SOC and DOC were found to be significant across different wavelengths of the visible (401.97 – 417.23,450.89 – 452.27,448.59 and 461.93 nm) and near infrared (931.68 – 1054.41 and 2527.81 – 2601.46 nm) regions. This research is the first step toward developing a spectral library of materials in the UAE.