Abstract
The quality of lands may be degraded by the accumulation of salts in soils, which is typically measured as soil Electrical Conductivity (ECe). High-salinity soils developed in low elevation spots in the Harran Plain after the initiation of intensive irrigation and crop production on clayey soils under high evaporation. This study evaluated the feasibility of using hyperspectral Visible and Near Infrared Reflectance Spectroscopy (VNIRRS) as a potentially more cost-effective approach for the characterization of soil salinity. 150 locations were taken at 0–15 and 15–30 cm depths from an area of 1000 ha with salinity levels ranging from none to very high. Sieved soils were measured for ECe using saturation paste and also scanned by VNIRRS in both air dried and oven dry states. For spectral preprocessing, raw reflectance spectra were averaged over 10 nm and a continuum removal (CR) method was applied. Calibration models between spectra and ECe were based on Multiple Adaptive Regression Splines (MARS), Partial Least Square Regression (PLSR), and Classification and Regression Trees (CART, for groupings). The VNIRRS data were also combined with topographical parameters from digital elevation models to improve estimations. Results showed that the estimation quality of ECe varied depending on approaches used, with the best results using continuum removed spectra of oven dried samples using MARS after separating samples containing high amounts of gypsum (R2 = 0.86, RPD = 2.70). Topographical variables with VNIRRS data improved estimations up to 12%. CART analysis showed that soils could be categorized as saline and non-saline based on soil reflectance with 65% accuracy.
This study was in part funded by Scientific Research Administration of Harran University, Sanliurfa, Turkey (HUBAK) and the Cornell University Computational Agriculture Initiative.
Notes
SP: Saturation Percent.
SAR: Sodium Absorption Ration; ECe: soil Electrical Conductivity.
TWI: Topo Wetness Index; Flow Acc: Flow Accumulation.
B: Soluble Boron.
†Variables with sample size of 43.
**,*significant at α = 0.05, significant at α = 0.1, respectively.
†All samples (n = 150).
Footnote‡The samples in PCA group I (n = 123).
§The samples in PCA group II containing gypsum (n = 27).
MARS: Multiple Adaptive Regression Splines.
PLSR: Partial Least Square Regression.
VNIRRS: Visible Near Infrared Reflectance Spectroscopy; CART: Classification and Regression Trees.