基于GWR模型的东江水质空间分异与水生态功能分区验证

流域水生态功能分区研究是我国正在开展的一项重要工作,如何验证分区结果的合理性,是当前亟待解决的问题。采用地理加权回归(GWR)模型评估流域特征对东江水质的影响,验证水质及流域影响空间差异是否与一二级水生态功能分区结果吻合,并对比了GWR模型与普通最小二乘(OLS)模型性能,讨论了GWR在分区验证方面的应用价值及不足。结果显示:1)水质指标以及GWR模型局部解释率(Local R~2)均在一二级水生态功能分区间存在显著差异;2)相比OLS模型,GWR模型校正R~2更高,残差空间自相关指数Moran′s I更低。研究表明东江水生态功能分区结果能合理反映水陆耦合关系,有效解释水质空间差异。此外建议选择总氮(TN)而非溶解氧(DO)和总磷(TP)作为分区验证指标。GWR模型在分区结果验证中具有广泛应用前景。降低数据空间自相关影响及改善距离测度方法是未来GWR模型研究的难点问题。

Spatial variations in water quality and their correspondence with ecoregions of the Dongjiang River basin using geographically weighted regression (GWR)

Water ecological function zoning in a watershed is an important task to perform in our country. How to validate the rationality of the zoning results has become an urgent problem to solve. In the study, we used the geographically weighted regression model (GWR) to evaluate the impact of watershed characteristics on water quality in the Dongjiang River. Our aims were to verify whether the spatial differences in impacts of watershed characteristics on water quality accorded with the results of primary and secondary water ecological function zoning, and to explore the possibility and limitation of the GWR model to validate the rationality of the zoning results according to a comparison of the results of the GWR model with the ordinary least squares (OLS) model. The results showed that: (1) Water quality indexes and Local R² showed significant differences among the primary and secondary water ecological function zones, respectively. (2) Compared with the OLS model, there was a higher adjusted R² and lower Moran''s I in the GWR model. These findings indicated that the results of water ecological function zoning could reflect the relationship between land and streams and the spatial differences in water quality in the Dongjiang River. In addition, the total nitrogen (TN) was found to better verify the zoning results than the dissolved oxygen (DO) and the total phosphorus (TP). The GWR model has wide application prospects in verifying zoning results, but a difficult problem faced by research into GWR models is reducing the spatial autocorrelation and improve the distance measure in the future.