Comparisons of forward-in-time and backward-in-time Lagrangian stochastic dispersion models for micro-scale atmospheric dispersion

Abstract

ABSTRACTLagrangian stochastic dispersion models are sometimes run in backward mode to estimate air emissions from different types of sources including area sources. The forward-in-time and backward-in-time Lagrangian stochastic (fLS and bLS) dispersion models may not result in the same estimates. The two models were compared under different atmospheric conditions in micro-scale applications. They are equivalent in a steady-state and horizontally homogeneous atmosphere in many features including estimating concentration at a point, using surface receptor, and prerunning the models. Although bLS shows better computational efficiency, it has a larger uncertainty in results due to the use of surface receptors. In a non-steady-state wind field, the two models show opposite transition trends when the wind fields experience a step change. Under sinusoidal-varying winds, the two models show different shapes of the predicated concentration curves. The normalized differences of the mean concentrations mainly increase with the receptor height when the source-receptor distance is fixed. A controlled methane release experiment was conducted to investigate the behaviors of the two models driven by real wind fields. The correlation coefficient between model-predicted concentrations is 0.95. The model-predicted (forward model) and measured concentrations show similar trend with a correlation coefficient of 0.70. The bLS model estimates larger peak concentrations than that fLS model under the same emission rate. The best-fitted results of the fLS and bLS models give recovery ratios of 1.1558 and 0.9675, respectively, which are better than that using a constant 15-min averaged wind (0.7922).Implications: There are large uncertainties and difficulties in quantification of fugitive air emissions from area sources such as landfills, agriculture, and industry sections. Lagrangian stochastic dispersion model is a versatile tool for these applications with the capability of near-field description and good efficiency. Backward models are usually used to estimate emission rates from area sources due to high computing efficiencies. But they may not result in the same estimate as the forward models due to factors involving model realization and input parameters. It is necessary to investigate the discrepancies to select the best model with minimal uncertainty in the results.