环鄱阳湖区农田土壤有机碳影响因素空间分布格局分析及制图研究

利用地理加权回归(Geographically Weighted Regression,GWR)模型,对环鄱阳湖区农田土壤有机碳(Soil Organic Carbon, SOC)影响因子的空间分布格局进行系统分析,结合克里格插值方法进行土壤有机碳制图研究,同时综合比较了地理加权回归方法(GWRK)与回归克里格(Return Kriging,RK)方法在预测土壤有机碳储量结果之间存在的差异。结果表明:(1)根据Pearson相关性分析结果,所选取的各环境影响因素均与土壤有机碳具有相关性(P0.01)。(2)通过GWR模型对各影响因素空间分布格局的分析,发现SOC与坡向、年均温度和年均降水呈正相关,与坡度、植被覆盖度呈负相关。(3)GWRK法得到的土壤有机碳含量变化范围为6.35～31.93g·kg~(-1),接近采样点的实测值;RK方法得到的土壤有机碳含量变化范围在7.41～25.76 g·kg~(-1),总体空间分布特征与GWRK的结果相似,但RK方法预测的结果较为平滑,与SOC储量实际状况还存在一定差异。(4)GWR的拟合优度R~2(0.50)明显高于RK中的R~2(0.20); GWRK方法的均方根误差(RMSE)为4.58,小于RK方法的均方根误差5.35,且通过GWRK方法得出的预测结果在制图上结合采样点位置信息,使成图效果更加精细。

Spatial Distribution and Mapping of Influencing Factors of Farmland Soil Organic Carbon in the Poyang Lake Region

This study systematically analyzed the spatial distribution of the influencing factors of farmland soil organic carbon (SOC) in the Poyang Lake region by using the geographically weighted regression (GWR) model, and combined kriging interpolation algorithm to produce SOC map. The GWR Kriging (GWRK) and kriging (RK) methods were compared to identify the difference in predicting the SOC reserves. The results showed that: 1) The selected environmental influencing factor was relevant with SOC (P<0.01) according to Pearson correlation analysis; 2) There was a positive correlation between SOC and aspect, annual average temperature, and annual average precipitation, while a negative correlation was observed between SOC and gradient, vegetation coverage; 3) The values of SOC content ranged from 6.35 to 31.93 g kg-1 using the method of GWRK and approached to the measured values of sampling points. The values of SOC content ranged from 7.41 to 2.76 g·kg^-1 using the method of RK, and the overall spatial distribution characteristics were similar to the results of GWRK; however, the predicting results of RK were relatively smooth, which differed from the actual state of SOC reserves; 4) The goodness of fit of R2 in GWRK (0.50) was apparently higher than that in RK (0.20). The root mean square error of GWR is 4.58, which was less than the RK (5.35) of RMSE. Additionally, the predicting outcomes obtained using the GWRK reflected the positional information of sampling points, which could make the mapping elaborate.