黄土丘陵沟壑区地形定量因子的关联性分析

Correlation of Quantitative Terrain Factors in Gully Hill Areas of China Loess Plateau

不同地形因子虽然在语义概念、计算方法等方面均有明显的差异,但各地形因子之间并不是绝对孤立的,它们之间相互关联、相互影响。这种关联的强弱与趋势,都从不同角度揭示着地形起伏变化与地貌发育的本质及内在规律,同时,还在一定程度上映射着地表形态的发育过程。文章以黄土高原丘陵沟壑区的15个样本地区为实验样区,以高分辨率、高精度的1∶1万比例尺DEM为基础数据,应用BP神经网络模型,探讨地形定量因子与地面坡度之间的关联性特征,并将神经网络的方法与传统的多元回归方法进行比较。结果表明,相对于传统的多元回归方法,带隐含层的BP神经网络分析方法能更为有效地反映地形因子间隐含的关联特征。该研究方法为进行地貌多定量指标的的选择和多因子之间关联性的量化提供了一种新的方法。

Terrain factors, although different in the definition and calculation method, relate each other at different extent. Such relationship can be represented by a correlation index, which reveals the process and stage of terrain development as well. This paper focuses mainly on the correlation between different terrain factors and the mean-slope by means of the Back Propagation model of Neural Network with a latent layer. Furthermore, the regression model and the NN model without a latent layer are compared with the NN model with a latent layer. Fifteen loess gully-hill areas are selected as the experimental area, and the relevant 1：10 000 scale DEMs （5 m × 5 m grid） are applied as the basic data. From the results of the NN model with a latent layer, it is found that roughness and undulation are the most closely correlated with mean-slope. Compared with others, channel density and mean elevation are the least correlated with meanslope. Experiment results show the NN model with a latent layer is better than the others and it can effectively evaluate the correlation between the terrain factors extracted from DEMs. This method provides a new methodology in the selection of suitable and available terrain factors and the estimation of the relevancy between these factors.