Fang Shoufan, Gertner George Z, Anderson Alan B
Department of Renewable Resources, University of Alberta, 751 General Services Building, Edmonton, Alberta T6G 2H1, Canada.
J Environ Manage. 2007 Oct;85(1):69-77. doi: 10.1016/j.jenvman.2006.08.002. Epub 2006 Oct 6.
Off-road vehicles increase soil erosion by reducing vegetation cover and other types of ground cover, and by changing the structure of soil. The investigation of the relationship between disturbance from off-road vehicles and the intensity of the activities that involve use of vehicles is essential for water and soil conservation and facility management. Models have been developed in a previous study to predict disturbance caused by off-road vehicles. However, the effect of data on model quality and model performance, and the appropriate structure of models have not been previously investigated. In order to improve the quality and performance of disturbance models, this study was designed to investigate the effects of model structure and data. The experiment considered and tested: (1) two measures of disturbance based on the Vegetation Cover Factor (C Factor) of the Revised Universal Soil Loss Equation (RUSLE) and Disturbance Intensity; (2) model structure using two modeling approaches; and (3) three subsets of data. The adjusted R-square and residuals from validation data are used to represent model quality and performance, respectively. Analysis of variance (ANOVA) is used to identify factors which have significant effects on model quality and performance. The results of the ANOVA show that subsets of data have significant effects on both model quality and performance for both measures of disturbance. The ANOVA also detected that the C Factor models have higher quality and performance than the Disturbance models. Although modeling approaches are not a significant factor based on the ANOVA tests, models containing interaction terms can increase the adjusted R-squares for nearly all tested conditions and the maximum improvement can reach 31%.
越野车通过减少植被覆盖和其他类型的地表覆盖以及改变土壤结构来加剧土壤侵蚀。研究越野车造成的干扰与车辆使用活动强度之间的关系，对于水土保持和设施管理至关重要。在之前的一项研究中已经开发出模型来预测越野车造成的干扰。然而，数据对模型质量和模型性能的影响以及模型的适当结构此前尚未得到研究。为了提高干扰模型的质量和性能，本研究旨在调查模型结构和数据的影响。该实验考虑并测试了：(1)基于修订通用土壤流失方程(RUSLE)的植被覆盖因子(C因子)和干扰强度的两种干扰测量方法；(2)使用两种建模方法的模型结构；以及(3)三个数据子集。验证数据的调整后R平方和残差分别用于表示模型质量和性能。方差分析(ANOVA)用于识别对模型质量和性能有显著影响的因素。方差分析结果表明，对于两种干扰测量方法，数据子集对模型质量和性能均有显著影响。方差分析还检测到C因子模型比干扰模型具有更高的质量和性能。尽管基于方差分析测试，建模方法不是一个显著因素，但包含交互项的模型在几乎所有测试条件下都可以提高调整后R平方，最大改进可达31%。
