基于地理加权回归的漫湾库区景观破碎化及影响因子分析

应用地理加权回归模型分析漫湾库区景观破碎化指数——有效筛网大小与相关因子之间的空间关系。选取的解释变量分别是距道路的距离、距乡村的距离、距河流的距离、坡度。结果表明：大坝修建后4种解释变量与有效筛网大小呈现较显著的正相关性。与线性回归模型相比,地理加权回归模型的拟合效果显著提高。1974~1988年,有效筛网大小对各影响因子最敏感的区域面积呈现显著的时空变化这为确定水电站建设及其他因素对景观破碎化影响的大小,并进一步改善库区景观破碎化的现状提供了依据。

Landscape Fragmentation and Affecting Factors of Manwan Reservoir Based on Geographically Weighted Regression

Applying Geographic Weighted Regression model, this article analyzed the spatial relationships between landscape fragmentation index-effective mesh size and related factors of Manwan Reservoir, Lancing River, Yunnan Provence. The selected explanatory variables covers factors of nature and human activity, including distance to main road, distance to county, distance to river and slop, aiming to determining the contribution of different factors to fragmentation. The results showed that: all the related factors exhibited significant positive correlations with effective mesh size after dam construction, which indicates these three variables can be used as factors for the spatial analysis of effective mesh size. We compared GWR model with Ordinary Least Square（OLS）model which presented that GWR model gave a much better fitting result with lower AICc value and higher adjusted R² value. Besides, the spatial distribution of residuals can examine the validity of the results. Apparent gathering characteristic indicated that the results of the model are invalid because the key explanatory factor is lost. Therefore global Moran′s I statistics on the residuals from OLS and GWR models were tested. For all the GWR models global Moran′s I ranges from 0.042 7-0.344 2 (p<0.01), while for all the OLS models it ranges from 0.478 6 to 0.545 8 (p<0.01) which indicated that the GWR models produced smaller global Moran′s I than OLS models with the same explanatory variables and reduced the spatial autocorrelation residuals of the models. Hence a GWR model improved the reliability of the relationships and was the optimization of OLS models. Coefficients of regression models reflect the sensitivity of the effective mesh size to each factor. The big coefficient represented the strong impact explanatory variable had on the effective mesh size. So we got the maximum value of coefficients as the most sensitive factor of effective mesh size. The most sensitive area to distance to road gradually reduced in 1974-1988. Before the dam construction in 1974, road was the most influential factor to the landscape fragmentation of the study area and its affected area occupied by almost 50% of study area, while after the construction and operation of Manwan hydropower station, its affected area reduced to 25% of the study area. However, the most sensitive area to distance to river and slop gradually increased. The most sensitive area to distance to county exhibited a trend of increasing firstly and then reducing. Although the impact of slope on effective mesh size in the three periods was the smallest, it showed a significant change. In terms of spatial distribution, the most sensitive area to distance to river located within 2 km to Manwan dam, the tail of the reservoir and the narrow-shaped part in the middle of the reservoir, which expanded to the entire reservoir area in 1974-2004.