Akademik Veri Yönetim Sistemi
EUROSOIL 2021, Geneve, İsviçre, 23 - 27 Ağustos 2021, ss.218
Visible-Near Infrared (Vis-NIR) spectroscopy is a good alternative to costly physical and chemical soil analysis to estimate a wide range of soil properties. Soil texture is one of the most important and fundamental soil parameters, as it affects many soil physical, chemical and biological properties. The aim of this study was to examine the potential of Vis-NIR spectroscopy for the prediction and tested of soils, having different textural distribution and contents in Çukurova Region was located in the Eastern of Mediterranean region of Turkey. Soil sampled was collected surface (0-0.3 m) and subsurface (0.3-0.6 m) depths. Sand, silt, clay and cation exchange capacity (CEC) was analyzed in the soil samples. Pre-processing of spectra such as noise removal, trimming, moving window, scatter corrections, de-trending, filtering and splice corrections was performed. Sixty soil samples collected from the study area scanned 350-2500 nm. However, the spectra 400-2450 nm was used due to noise and its overtones and were divided into calibration (75%) and validation (15%) sets. This study compared the performance of three different calibration methods, namely, principal component regression (PCR), partial least squares regression (PLSR) and artificial neural network (ANN) with Levenberg-Marquardt analyses for the accuracy of measurement of selected soil properties based on textural differences of samples. All of tested methods was excellently predict the relevant soil properties apart from negligible differences. The prediction performances of the PCR model were highest accurate (Lin’s concordance correlation coefficient (LCCC), residual prediction deviation (RPD) and ratio of performance to interquartile distance (RPIQ)) and had lowest root mean square of prediction (RMSEP) for all selected soil properties (LCCC: 0.97, RPD: 4.06, RPIQ: 5 and RMSEP: 8.74 for sand; LCCC: 0.94, RPD: 2.79, RPIQ: 3.82 and RMSEP: 5.68 for silt; LCCC: 0.95, RPD: 3.29, RPIQ: 4.69 and RMSEP: 6.72 for clay; LCCC: 0.95, RPD: 3.33, RPIQ: 5.11 and RMSEP: 2.19 for CEC). The performance of PLSR model was slightly low accurate and had high RMSEP from the other models but reliable and the results were as follows: (LCCC: 0.97, RPD: 3.94, RPIQ: 4.85 and RMSEP: 9.00 for sand; LCCC: 0.94, RPD: 2.79, RPIQ: 3.82 and RMSEP: 5.69 for silt; LCCC: 0.95, RPD: 3.29, RPIQ: 4.69 and RMSEP: 6.71 for clay; LCCC: 0.94, RPD: 2.71, RPIQ: 4.15 and RMSEP: 2.69 for CEC). Prediction of ANN models was performed, selected based on the specific wavelengths ranges: 400-700; 700-1100; 1100-2450; 700-2450 and 400-2450 nm. According to prediction results, highest R2: 0.97 at 1100-2450 nm and lowest RMSEP: 8.64 at 400-2450 nm for sand; highest R2: 0.97 at 1100-2450 and 400-2450 and lowest RMSEP: 3.71 at 400-700 nm for silt; highest R2: 0.98 at 700-2450 nm and lowest RMSEP: 0.84 at 1100-2450 nm for clay and highest R2: 0.95 at 400-2450 and lowest RMSEP: 2.12 at 700-1100 nm for CEC. In terms of the selected soil properties, soils having textural differences was successfully predicted by means of the prediction models even so was collected limited number of samples. As a result, further studies should focus on the measurement tools and calibration models so that make correctly prediction of site-specific soil properties, changing on a narrow range.