| Abstrakt: |
Analyzing the spatial variability of soil organic carbon (SOC) content is crucial for evaluating soil quality and associated factors, including structural stability, nutrient cycling, biological activity, and soil aeration. The precise mapping of SOC distribution is vital for gaining scientific insights and promoting sustainable land management. This study, carried out in the semi-arid ecosystem of Corvera, Murcia, Spain, introduces an innovative modeling approach that combines drone-based multispectral sensor data with laboratory measurements to estimate SOC content. The hybrid model incorporates various index variables, including the Differential Vegetation Index (DVI), Enhanced Vegetation Index (EVI), Soil Adjusted Optimized Vegetation Index (OSAVI), Normalized Difference Vegetation Index (NDVI), and Soil Adjusted Vegetation Index (SAVI). To validate the model, 76 soil samples were collected at a depth of 30 cm, and SOC content has been quantified using the Walkley–Black method, where 80% of the samples are reserved for model training, and 21 auxiliary predictors are integrated. The primary objectives of this study involve assessing the predictive performance of machine learning algorithms, including artificial neural networks (ANN), support vector machines (SVM), random forests (RF), and multiple linear regression (MLR). The focus is on evaluating the effectiveness of these algorithms in predicting Soil Organic Carbon (SOC) content. The results indicate that the random forests (RF) algorithm outperforms others, demonstrating high efficacy (R2= 0.97, RMSE = 1.41, RPIQ = 2.48). In addition to algorithm evaluation, SOC mapping results for the semi-arid Corvera region reveal distinctive spatial patterns. Central areas exhibit higher SOC content, while lower levels are found along the periphery. This spatial variation provides a nuanced understanding of SOC distribution, which is critical for effective soil management and environmental planning in the studied region. |
Full text from SpringerLink