Inter-annual variability in national exposure estimates from iand use regression models

摘要

Background: Land Use Regression (LUR) models have become a common method for estimating exposures to traffic-related air pollution. However these empirical models are usually based on short term measurement campaigns of several weeks, in some cases repeated 3-4 times at different seasons of one year. The relevance of these models to assessing chronic exposures over many years remains unclear. Aims: To examine the inter-annual variability in exposure estimates over a decade. Methods: National LUR models were developed for NO2 for all of Israel from existing air quality monitoring network (AQMN) data available between 1998-2010, with an increase from 11 to 78 in the number of sites over that time. First, a base case model was developed from the collection of the annual mean NO2 at each station and each year (N = 688). Two options were then employed for testing inter-annual variability of the LUR, using the same set of predictors: 1. calculating a separate model for each year 2. Applying the regression coefficients of the base case model to each year. Results: The base case model achieved an adjusted R2 of 0.75 (leave-one-outcross validation 0.74, p-value＜0.0005, slope 0.99). The inter-annual variability in the models across 13 years (option 1), in terms of adjusted-R2, ranged between 0.52-0.73. Applying the regression coefficients of the base case to each year (option 2) yielded adjusted-R2 in the range of 0.69-0.91, an increase of ～8-20% in R2 for individual years, compared to the first option. No trend could be identified in inter-annual variability of the models. Conclusions: A LUR model applied separately for each year had an inter-annual variability of 21% in terms of R2. When applying regression coefficients of a base case model to each year, the variability remains about the same (23%) but the models are able to explain a larger portion of the variance.