关键词: 巴丹吉林沙漠, 湖泊群, 地下水补给, 聚类分析

Abstract: This article introduced 13 basic variables including the lakes' longitude and latitude, elevation, water temperature, water pH and lake Na⁺, K⁺, Mg²⁺, Ca²⁺, Cl^-, SO₄^2-, CO₃^2-, HCO₃^- concentration and took these variables as the clustering indices for nonlinear clustering of 105 perennial water lakes in the sorth of Badain Jaran Desert by building the self-organizing feature map (SOFM) network. By comparing with linear clustering results by using the group average method, we found that when 105 lakes were divided into two groups through SOFM network, the spatial boundary was generally consistent with Yabulai Mountain Northeastern-Southwestern Strike Fault Zone. The lakes appeared two clustering groups obviously, lakes which were distributed at the north of the Fault Zone gathered in one group while the rest of lakes that were close to or at the south of the fault zone got together as the other group. Meanwhile the field investigation confirmed the clustering results. After analyzing the clustering results of two methods we found that clustering results of SOFM network are more accurate and reliable, and have more advantage in identifying subtle differences of geographic phenomena. Different types of lakes distribute in different sides of the fault zone, so we infer the spatial non-heterology of the underground water recharge of lakes in the north of Badain Jaran Desert and spatial differences of underground rock structure.

Key words: Badain Jaran Desert, group of lakes, cluster analysis, groundwater recharge