| Summary: | Accurate assessment of wet aggregate stability is critical in evaluating soil quality. However, a few general models are used to assess it. In this work, we use the support vector machine to evaluate wet aggregate stability and compare it with a benchmark model based on artificial neural networks. One hundred thirty-four soil samples from various land uses, such as crops, grasslands, and bare land are adopted to verify the effectiveness of the proposed method and confirm the valid input parameters. We select 107 samples for calibrating the prediction model and the rest for evaluation. Experiments show that organic carbon is the main control parameter of wet aggregate stability, although the most influential factors for different land use are various. Comparing the determination coefficient and the root mean square error, it proves that the support vector machine method is superior to the artificial neural network method. In addition, the relative importance analysis shows that contents of organic carbon, silt, and clay are the primary input parameters. Finally, the impact of land use and management types is evaluated.
|
|---|