Epsilon Archive for Student Projects

Abstract

Visible-Near Infrared (VNIR) spectroscopy provides rapid, low cost, non-destructive and simultaneous measurements of several constituents from a single scan. The method has recently gained attention as an alternative to conventional soil analytical techniques in soil quality assessment. However, most of the research has focused on spectral measurement after sample preprocessing in the laboratory. This study was conducted in Ethiopia in collaboration with Ethiopian Agricultural Transformation Agency (ATA). In this study, 112 soil samples were scanned in the field and the result was compared with scans made in the laboratory on pre-processed samples. All samples were collected from a geographical area of 10km x 10Km block. Partial Least Square (PLS) regression was used to determine the best re-gression model between laboratory reference data and spectral data for both field and laboratory measurements. The result demonstrated that VNIR spectroscopy can predict soil chemical and physical properties both in the field and under laboratory conditions but that the predictions are slightly better for soil particle size distribution (texture) and few chemical properties in the field than in laboratory. This could be either due to the number of scans per sample performed or co-variation of soil physiochemi-cal properties with soil moisture. Fraction of organic carbon (FOC), CEC, Ca, pH, Co and %clay particle size were all predicted with better accuracy (Residual Prediction Deviation (RPD) ≥2 and R2 >0.7) for both field and laboratory measurements. Log transformation of P reference data seemed as if it improved the model performance since R2 increased (R2>0.7) but the back transformation showed failure in prediction since RMSECV increased. Performance of N, Mg, Cu, Si, Fe, Al, %silt and %sand were acceptable with 1.4≤RPD≤2 and 0.54≤R2≤0.72. Prediction of K, S, Mn, Zn, Na, B, EC, with raw and various transformation of the spectra and reference data failed to produce a useful result with RPD <1.4 and R2<0.5. The predictive ability achieved makes VNIR spectroscopy suitable for rapid assessment of soil quality under field conditions.