توزیع مکانی کربن آلی خاک سطحی در تیپ های گیاهی مراتع آبخیزچهلگزی شهرستان سنندج، استان کردستان

Extended Abstract
Introduction and Goal
Soil organic carbon is one of the most important soil quality indicators that affects almost all physical, chemical and biological properties of the soil and causes soil fertility and is important in dry ecosystems from various aspects such as reducing greenhouse gas effects, increasing soil yield and fertility, reducing soil erodibility, increasing water and nutrient retention capacity, etc. Therefore, today it is classified as one of the determinants of soil health. The purpose of this study is to evaluate the power of spectral and non-spectral soil factors to predict the spatial distribution of topsoil organic carbon in the semi-arid Chehelgazi rangelands of Sanandaj using factor analysis and multiple linear regression.

Materials and Methods
Landsat 8 satellite data dated 06/18/2019 with path and row numbers 167 and 35, respectively, were used. Homogeneous working units for sampling were prepared using slope, direction, and elevation class maps obtained from the base and digital elevation maps of the region. Soil sampling was carried out using a stratified random method. For this purpose, 135 training sites were selected in homogeneous working units.
Of these, data from 105 sites were used to calibrate the model and data from 30 sites were used to validate it. In each of the training sites, a composite soil sample consisting of nine observations was randomly collected from a depth of 0 to 20 cm of topsoil. In each sample, the soil organic carbon content was determined using the Walkley-Block titration method. Considering the area of the study area and by reviewing the research conducted by other researchers, the independent variables selected for predicting topsoil organic carbon were classified into two groups including spectral variables (derived from Landsat 8 images) and non-spectral variables (derived from elevation and relief factors).
In order to determine a suitable model for predicting topsoil organic carbon, 22 variables were used. These variables included 15 spectral variables and seven standard shape variables. Among the 22 variables considered, only 15 variables were identified as having a normal distribution and suitable for conducting an initial factor analysis. By repeating the factor analysis and examining the adequacy of the data through the KMO index and Bartlett’s Test of Sphericity and identifying and eliminating variables with the smallest common features with a sampling precision of less than 0.5 and repeating the above test, finally 8 variables including; surface albedo, NDVI index, brightness, greenery and moisture indices of the Tesselcap transformation, clay index, slope and altitude above sea level were selected as suitable and qualified variables to enter the factor analysis model.
Of these, 5 are spectral variables and 3 are non-spectral variables. This indicates that spectral variables are more important than non-spectral variables in explaining changes in the response variable (soil organic carbon). Using multiple linear regression, a suitable regression equation was calculated to predict topsoil organic carbon (R2= 0.66).

Results and Discussion
To create a soil organic carbon prediction model, first the multiple linear regression relationship between soil organic carbon and the factor information layers at 105 sampled points was tested. By examining multiple collinearities and calculating the coefficients of independent variables using the joint method, the accuracy of the aforementioned model was examined. The comparison of the average soil organic carbon reserves in the plant types in the study area using analysis of variance was significant (p < 0.01). By performing a one-way normal mean plot analysis, the plant types were classified into two groups (more and less than the average total organic carbon of the habitat), which is equal to 1.26 percent. Regression model testing using 30 observations showed that the correlation coefficient of predicted values with actual organic carbon values was 0.71, and the root mean square error (RMSE) and relative mean absolute error (MARE) were 0.191 and 0.117, respectively.

Conclusion and Suggestions
According to the results of the one-way normal mean analysis (ANOM), the above plant types were classified into 2 groups (equal to the total mean less than the total area mean and greater than the total area mean) as follows; a) Plant types 1, 2 and 5 whose average organic carbon in the surface soil horizon has a significant difference with the average of all samples (150 samples) which is equal to 1.26 at the probability level (p < 0.05) and their organic carbon content is 1.96, 1.39 and 1.44 respectively. b) Plant types 3 and 4 whose average organic carbon in the surface soil horizon has a significant difference with the average of all samples (360 samples) which is equal to 1.26 at the probability level (p < 0.05) and their organic carbon content is 0.88 and 1.16 respectively.
According to the results obtained, the average weight of organic carbon in the topsoil of the rangelands of the studied area is 1.26%. Considering the area and topography of the area and the number of samples taken, this figure may change if the number of samples is increased and the sampling points are more appropriately distributed.