Abstract

Atmospheric dispersion is a process that involves many uncertainties in model parameters, inputs and source parameters. In this article, we present an uncertainty management procedure for the height release at source which is a key parameter in modeling the subsequent dispersal of contamination after a nuclear accident. When setting the initial parameters of a dispersal model, it is difficult to estimate the height of the release and this will obviously affect the consequences. This procedure reduces the risk of setting an erroneous height value by running mixed model. That is, we include several models in our analysis, each with a different release height. The Bayesian methodology assigns probabilities to each model representing its relative likelihood and updates these probabilities in the light of monitoring data. The effect this has is that the data give most weight to the most likely model and thus models, which consistently badly perform can be discarded. As an illustration we perform sequential learning with an atmospheric dispersion model on a real site under real atmospheric conditions using data from tracer experiments.

摘要

大气扩散是一个涉及模型参数、输入和源参数的许多不确定性的过程。在本文中，我们提出了源高度释放的不确定性管理程序，这是模拟核事故后污染随后扩散的关键参数。在设置扩散模型的初始参数时，很难估计释放的高度，这显然会影响后果。此过程通过运行混合模型降低了设置错误高度值的风险。也就是说，我们在分析中包含了几个模型，每个模型都有不同的释放高度。贝叶斯方法为每个模型分配代表其相对可能性的概率，并根据监测数据更新这些概率。这样做的效果是数据赋予最可能的模型最大的权重，因此可以丢弃一直表现不佳的模型。作为说明，我们使用来自示踪剂实验的数据在真实大气条件下的真实站点上使用大气扩散模型执行顺序学习。