Introduction
Soil organic carbon (SOC), as one of the most important components of the global carbon cycle, plays a vital role in maintaining soil quality, enhancing fertility, and moderating climate change. The spatio-temporal variations of SOC are influenced by various factors, including land use, climatic conditions, topography, and human activities. Additionally, SOC contributes to diverse functions in natural and agricultural ecosystems, such as increasing soil fertility, controlling erosion, enhancing water permeability in soil, and reducing the effects of greenhouse gases.
Materials and Methods
Given the critical role of SOC in enhancing soil quality, this study aims to investigate the spatio-temporal variability of SOC using a reverse modeling approach based on a spatial model developed in 2024. The model is extended to analyze data from the years prior to 2015, 2010, and 2000, incorporating environmental variables derived from remote sensing (RS), topographic attributes, climatic data, land use, and geological information within the Zayandeh Rud watershed. For the environmental covariates, RS data, land use, and climatic information were obtained from Google Earth Engine's open-source spatial database for the relevant years from 2000 to 2024. In total, 76 auxiliary variables were prepared, including vegetation indices derived from band ratios of RS data, as well as the digital elevation model (DEM), geology, land use, and climatic factors, which were used as representatives of soil-forming factors. A relative importance feature selection method was employed to finalize the dataset of environmental covariates. Furthermore, three machine learning models—Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) tree—were utilized to explore the relationships between environmental factors and SOC. Four common statistical indices, including the coefficient of determination (R²), concordance correlation coefficient (CCC), root mean square error (RMSE), and percentage of normalized root mean square error (nRMSE), were used to evaluate the performance of the machine learning models Two uncertainty quantification approaches for prediction, namely bootstrap and k-fold cross-validation, were also applied.
Results and Discussion
The evaluation of machine learning models for predicting SOC revealed a notable decline in the performance of all models from the year 2024 back to 2000, as indicated by the R² statistic. Among the models assessed, the RF model exhibited superior performance, achieving the highest R² values for the years 2024 and 2015, thereby indicating its effectiveness in capturing the complexities of SOC dynamics. The SVR model demonstrated intermediate performance, while the XGBoost model showed relatively weaker results compared to the other two model. Despite these variances in performance, all three machine learning models effectively established a robust connection between SOC and the predictive environmental variables, affirming their suitability for this analysis. Furthermore, the uncertainty quantification of SOC predictions highlighted that the bootstrap method outperformed the k-fold method, yielding lower values for both standard deviation (SD) and mean uncertainty, which suggests that the bootstrap approach provides a more reliable prediction of SOC variability. In terms of the relative importance of environmental variables in predicting SOC, the analysis across all time periods indicated that climatic factors played the most significant role, closely followed by topographical attributes. Other environmental variables, including land use, geology, and RS data, had a lesser impact on explaining spatial variations in SOC. The spatial analysis indicated alarming increases in areas with very low SOC content, suggesting soil degradation risks. Furthermore, higher rainfall and lower temperatures were associated with highest SOC levels, emphasizing the need for effective soil management strategies.
Conclusion
This study emphasizes the necessity of ongoing monitoring and management of Soil Organic Carbon (SOC) to combat soil degradation and promote sustainable agriculture. The findings also provide a framework for creating soil property maps in data-scarce regions, enhancing decision-making for effective soil management strategies.
Overall, the findings from this study highlight the need for continuous monitoring and management of SOC to mitigate risks related to soil degradation and to promote sustainable agricultural practices. Also, the methodology and framework that applied in this research can be used as useful guideline for land manager, young pedometrisians, expert in digital soil mapping, and decision makers for preparing the high detailed maps of other key soil properties such as (soil texture component: Sand, Silt, Clay, Gypsum, calcium carbonate equivalent, soil EC, available phosphorus and potassium that directly affecting on soil quality index in the similar zones with study area. |
