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

LIU Shi-liang,LIU Qi,WANG Cong,ZHAO Qing-he,DENG Li,DONG Shi-kui

DOI: https://doi.org/10.13249/j.cnki.sgs.2014.07.003

2014-01-01

Abstract:Applying Geographic Weighted Regression model, this article analyzed the spatial relationships be-tween 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, in-cluding distance to main road, distance to county, distance to river and slop, aiming to determining the contri-bution 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 ex-planatory 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 small-er global Moran′s I than OLS models with the same explanatory variables and reduced the spatial autocorrela-tion residuals of the models. Hence a GWR model improved the reliability of the relationships and was the op-timization 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 af-fected 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.