برآورد کربنات‌کلسیم خاک با استفاده از تصاویر ماهواره‌ای و طیف‌سنجی فروسرخ نزدیک

Introduction: One of the advantages of remote sensing and visible-near infrared (Vis–NIR) spectroscopy is the speed, simplicity, and cost-effectiveness of analysis compared to traditional methods. Remote sensing is a scientific discipline that involves collecting data while minimizing direct physical contact with the objects being studied. To fully leverage the rapid analysis capabilities of Vis–NIR spectroscopy, it is essential to exploit its advantages over conventional analytical techniques. The aim of this research is to utilize Landsat 8 satellite sensors and the near-infrared spectrum for agricultural and forestry applications in the Gyan Nahavand plain of Hamadan Province to estimate soil calcium carbonate.
Material and methods: Forty-eight soil samples were collected from the surface layer (0-30 cm), followed by air drying and sieving to a 2-mm particle size. Several physicochemical characteristics of the soils were analyzed. A Landsat 8 image from September 2019 was utilized for remote sensing studies. The calculated values for each sample unit were generated using ERDAS Imagine 9.1 software. The values for each band at the 48 sampling points were entered into Excel, and the variables were statistically described. In the remote sensing method, the spectral reflectance of the samples was extracted and processed across ten primary bands. In addition to the primary bands, combinations of bands and calcite indices were also employed. Correlations between the values of the primary bands, band compositions, calcite indices, and the amount of soil calcite were analyzed. The best model was selected by fitting various multivariate regressions without excluding outlier data. Spectral analysis of the targeted soils was conducted using a FieldSpec 3 spectroradiometer, with a wavelength range of 350-2500 nm. After recording the spectra, various preprocessing methods were evaluated. The Pearson correlation test and linear regression analyses were performed using SPSS 24.1 software.
Results and discussion: Laboratory results indicated that the average soil calcium carbonate content in agricultural and forested areas was 30% and 22.22%, respectively. The findings revealed that bands 10 and 11 exhibited a significant correlation with soil calcite in forested areas (p < 0.05). Twelve band compositions at the 5% significance level and six band compositions at the 1% significance level demonstrated a significant correlation with the amount of soil calcite. Additionally, the R1 index ((Band5/Band4)/(Band5/Band2)) showed a significant correlation with soil calcite (p < 0.05). The correlation between the measured calcite in the laboratory and the equation derived from satellite imagery was found to be moderate (r² = 0.45) for agricultural use. In the spectroscopic analysis, the highest correlation was observed at a wavelength of 612 nm (r² = 0.85**). Based on modeling using Partial Least Squares Regression (PLSR), the determination coefficient for the calibration group for calcite was 0.8, with a Root Mean Square Error (RMSE) of 4.8%. In the validation group, the determination coefficient was 0.5, and the RMSE was 7.8%. Among the models fitted using multivariate regression with satellite images, the Stepwise Multivariate Linear Regression (SMLR) model is recommended as a suitable approach for estimating calcite. The Partial Least Squares Regression (PLSR) model has also proven to be nearly suitable for estimating calcite using the spectroscopic method.
Conclusion: The overall results indicate that the model developed using regression statistical methods in remote sensing has effectively estimated the amount of soil calcite in agricultural lands. The quantities obtained through remote sensing and laboratory analyses show minimal differences. Therefore, it can be concluded that the remote sensing method is successful in estimating soil calcite levels. Additionally, the results from spectrometry demonstrate that the PLSR model is suitable for estimating soil calcite, provided that a larger number of samples is utilized. In general, we can conclude that the Vis-NIR spectroscopy method offers greater accuracy compared to remote sensing and titration methods, although it requires a more extensive sample set. It is recommended to increase the number of spectroscopic samples to enhance the accuracy of the findings and to ensure that the same number of samples is used for a more effective comparison between the two land uses.