Search

Abstract Land use regression (LUR) has been used to predict NO 2 and NO distribution. However, previous studies overlooked the possibility that the effect of land-use configuration on NO 2 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 NO 2 and NO. First, a LUR model was adopted to screen the land-use factors for NO 2 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 NO 2 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 NO 2. 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 NO 2 , which implies that NO is more localized. The relationships between land-use configuration, NO 2 and NO concentrations are not constant in space. This means that the predictive abilities of land-use factors for NO 2 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. Highlights • Landscape metrics depict the spatial disparity of urban land-use configuration. • Discrepant influences of land-use configuration on NO 2 and NO were revealed. • GWR increased NO 2 and NO prediction accuracy by 29.3% and 6.9%, respectively. • NO had a steeper spatial distribution gradient than NO 2. • The findings benefit optimal regulation of the urban land-use configuration. [ABSTRACT FROM AUTHOR]