基于多元自适应回归样条的青藏高原温泉区域的冻土分布制图

Modeling Permafrost Distribution in Wenquan Area over Qinghai-Tibet Plateau by Using Multivariate Adaptive Regression Splines

以探地雷达、电磁测深、钻探等技术方法获得野外数据及数字高程(DEM)遥感数据为基础,通过聚类分析和相关性分析对高程、坡度、坡向等因素对多年冻土分布的影响进行了定量化研究.利用非线性的多元自适应回归样条(MARS)方法建立了基于高程、太阳辐射的多年冻土分布模型,通过自身的交叉验证及对比年平均地温模型和逻辑回归模型的总体分类精度,说明MARS模型具有较好的分类精度.运用MARS模型模拟了整个温泉区域冻土的空间分布特征.结果表明:MARS模型分类精度较高,验证了此模型模拟温泉区域冻土分布的可行性;此模型除了考虑高程对对多年冻土分布的控制作用外,还体现了太阳辐射这一局地综合因素对多年冻土分布的调整作用,较好地模拟了高程相对较低的低山区多年冻土的存在.

In order to understand the distribution patterns of permafrost in the Wenquan area on the Qinghai-Tibet Plateau,the effects of altitude,slope and aspect and other topo-climatic factors on the distribution of permafrost were studied,using the correlation analysis with digital elevation（DEM） data,borehole observations and measures from ground penetrating radar（GPR） and the electromagnetic sounding method.A permafrost distribution model based on the nonlinear multiple adaptive regression splines（MARS） method was developed,taking elevation and direct solar radiation as variables.Five-fold cross validation shows that this model has a good simulation capability in describing the permafrost distribution spatial pattern in the study area.Applying the model to the study area indicates that in the Wenquan area there is 1 881 km2 of permafrost area,accounting for 76% of the total Wenquan area.The MARS model is better than the mean annual ground temperature model and the logistic model,because the MARS model takes into account not only elevation,the predominantly controlling factors in a mountainous area,but also solar radiation,which has adjusting effect to the permafrost distribution.