Detecting urban land-use configuration effects on NO2 and NO variations using geographically weighted land use regression

• Published in: Atmospheric environment (1994) 2019-01, Vol.197, p.166-176
• Main Authors: Song, Weize, Jia, Haifeng, Li, Zhilin, Tang, Deliang, Wang, Can
• Format: Article
• Language: English
• Subjects: Geographically weighted regression; Land use regression; Landscape metrics; NO2
• ISSN: 1352-2310; 1873-2844
• DOI: 10.1016/j.atmosenv.2018.10.031

Land use regression (LUR) has been used to predict NO2 and NO distribution. However, previous studies overlooked the possibility that the effect of land-use configuration on NO2 and NO may not always be constant across the study domain. The objective of this study was to depict the spatially varying effect so as to better predict NO2 and NO. First, a LUR model was adopted to screen the land-use factors for NO2 and NO predictions. Then, a geographically weighted regression (GWR) model was developed to delineate the spatial non-stationarity in the relationship. The results show that the GWR model improved NO2 and NO prediction accuracy, with increases of 29.3% and 6.9%, respectively. The road ERSD (i.e., shortest distance to express road) factor had a negative effect on NO2. The impervious AWMSI (i.e., area-weighted mean shape index) factor had a larger effect on NO in the northwest of Foshan, due to more uneven and dense distribution of impervious patches. NO had a steeper distribution gradient than NO2, which implies that NO is more localized. The relationships between land-use configuration, NO2 and NO concentrations are not constant in space. This means that the predictive abilities of land-use factors for NO2 and NO are different across Foshan. Overall, our approach can obtain a higher estimation accuracy than the LUR at city scale. It could also be applied easily for other air pollutants and in cities worldwide. •Landscape metrics depict the spatial disparity of urban land-use configuration.•Discrepant influences of land-use configuration on NO2 and NO were revealed.•GWR increased NO2 and NO prediction accuracy by 29.3% and 6.9%, respectively.•NO had a steeper spatial distribution gradient than NO2.•The findings benefit optimal regulation of the urban land-use configuration.