عنوان مقاله [English]
Introduction Spatial variations of soil properties is a natural event, which recognizing these changes is inevitable in order to planning and right management of both agricultural and natural resources. Soil organic carbon (SOC) is the most important factor in soil fertility and quality, climate change and reduction of greenhouse gas emissions. Furthermore, evaluating the rates and spatial distribution of the soil properties, land improvement and restoration can be traced from the carbon sequestration index. According to the above, providing quantitative and qualitative conservation of soil properties such as SOC can be considered an effective way to achieve sustainable development of natural and environmental resources. Digital soil mapping (DSM) can determine the spatial variations of soil organic carbon by exploring the relationship between soil properties and effective environmental variables. Different statistical models such as regression trees and random forest are used in order to communicate between soil characteristics and its spatial distribution. The present study was carried out to investigate the spatial distribution of SOC, as well as, to determine the most important variables affecting their prediction in Ravang watershed in Minab County.
Materials and MethodsRavang watershed with an area of 13821.6 hectares is located in Hormozgan province, Minab city. The maximum and minimum elevations are 357 and 33 meters, respectively. Digital Elevation Model of Ravang watershed was used to extract 17 environmental covariates (such as elevation, aspect, slope, valley depth,…) by SAGA software (http://www.gdem.aster.ersdac.or). Moreover, two environmental covariates related to remote sensing including Normalized Difference Salinity Index (NDSI) and Normalized Difference Vegetation Index (NDVI) were determined in the study area. In addition, the maps of land use, sand, silt, clay and pH were used as covariates in modeling. In order to determining the location of sampling points, the conditioned hyper-cube technique was used. After determining of soil sample location, field sampling was carried out at a depth of 0-30 cm. then, 100 soil samples were taken and the amount of SOC was measured. Random forest model was applied to the relationship between SOC and covariates. The model includes two user-defined parameters, including the number of variables used in the construction of each tree, which expresses the power of each independent tree and the number of trees in each forest. Considering the strength of independent trees, the predictive accuracy of the model increases, conversely, the correlation between them will decrease. The accuracy of the soil organic carbon distribution was also evaluated using root mean square error (RMSE), mean error (ME) and correlation coefficient (R2), which were determined.
Results and Discussion Based on the present study results, elevation, soil silt and sand maps, channel network base level, slope and NDVI are the most important factors on predicting the of SOC variations. The results indicated that RMSE, ME and R2 were 0.36, 0.26 and 0.38, respectively .Results also showed that affecting erosion and sediment, as well as, human effect, have the most impact on the SOC soil spatial distribution in the Ravang watershed. Moreover, result show SOC deficiency in the soil of Ravang watershed due to high salinity, low percent of vegetation cover and land use changes. In addition, drought intensifying and decrease in precipitation have reduced SOC content, which itself causes changes in the texture and chemical properties of the soil and, as a consequence, makes them more susceptible to erosion.
Conclusion The variability of SOC is very high in the study area because of intensive water erosion and land use change. Overall, the results of the present study indicated that the critical condition of soil organic carbon in the Ravang watershed, which requires a comprehensive management of the region's water and soil resources to improve soil conditions and increase the reserves of this important and influential variable in the soil structure. On the other hand, despite of the acceptable performance of the random forest model in estimating of soil properties, due to high variability of some soil properties, model prediction performance may be decreased.