Using Land Use Regression Models To Model Spatial And Temporal Variability Of NO₂ And PM₁₀ In Changsha,China

The lack of in-depth exploration on the influence of buffer types, the application of meteorological factors and the improving of temporal resolution on NO₂ and PM₁₀ LUR simulating research lead to that LUR model cannot explain the influence of meteorological factors on pollutants. This defect becomes a serious impediment for improving model simulation accuracy, expanding the application scope of LUR model and playing the advantage of higher spatial resolution. In such a case, it is difficult to provide comprehensive, accurate, high-precision NO₂ and PM₁₀ pollution data for decision-makers and researchers in epidemiological studies. Therefore, in this study, simulation of NO₂ and PM₁₀ seasonal average concentrations in 2010 and daily average concentration between April 2013 and April 2014 in Changsha city were chosen as example. Based on common land use characteristic variables and meteorological factors variables, we built LUR models with different buffer types, MFR models and BPNN models to analyze the spatial and temporal variation of NO₂ and PM₁₀. Results show that:The contribution rate of different characteristic variables for NO₂ and PM₁₀ pollution is different, of which the most important spatial explanatory variables included major roads, residential land and public facilities land, indicating that the spatial distributions of NO₂ and PM₁₀ are closely related to traffic conditions and human activities. The results also show that the pollution contributions of land use variables changed significantly with spatial scale, the variables with in 1200 m buffers have large influence on pollutants.The circular-buffer-based（CBB） LUR model and semicircular-buffer-based（SCBB） LUR model constructed based on the same modeling process and the same land use variables have large difference in predicting NO₂ and PM₁₀. Results showed SCBB LUR model had significantly higher R2 values than traditional CBB LUR models, supporting the usefulness of this approach incorporating wind direction in the LUR model.The MFR model and BP neural network model constructed based on the characteristics of meteorological variables. Important temporal explanatory variables included temperature, wind speed, cloud cover and percentage of haze. Pearson’s r values between predicted and measured concentrations were much higher in BPNN models than in MFR models. The results demonstrate that the BPNN models showed a better performance than the MFR models in modeling temporal varia tion of NO₂ and PM₁₀.