Dpt of Econometrics and Statistics, University of the Basque Country, L Agirre etorb 83, E48015 Bilbao, Spain.
Environ Pollut. 2011 Oct;159(10):2947-53. doi: 10.1016/j.envpol.2011.04.026. Epub 2011 May 10.
Underlying levels of atmospheric pollutants, assumed to be governed by smoothing mechanisms due to atmospheric dispersion, can be estimated from global emissions source databases on greenhouse gases and ozone-depleting compounds. However, spatial data may be contaminated with noise or even missing or zero-valued at many locations. Therefore, a problem that arises is how to extract the underlying smooth levels. This paper sets out a structural spatial model that assumes data evolve across a global grid constrained by second-order smoothing restrictions. The frequency-domain approach is particularly suitable for global datasets, reduces the computational burden associated with two-dimensional models and avoids cumbersome zero-inflated skewed distributions. Confidence intervals of the underlying levels are also obtained. An application to the estimation of global levels of atmospheric pollutants from anthropogenic emissions illustrates the technique which may also be useful in the analysis of other environmental datasets of similar characteristics.
在大气污染物的基础水平下，可以根据温室气体和消耗臭氧化合物的全球排放源数据库来估计，这些基础水平被认为是由大气扩散引起的平滑机制所控制。然而，空间数据在许多地方可能受到噪声的污染，甚至缺失或为零值。因此，出现的一个问题是如何提取基础的平滑水平。本文提出了一种结构空间模型，该模型假设数据在一个由二阶平滑限制约束的全球网格上演变。频域方法特别适用于全球数据集，减少了与二维模型相关的计算负担，并避免了繁琐的零膨胀偏态分布。还获得了基础水平的置信区间。对人为排放的大气污染物全球水平的估计应用说明了该技术，该技术在分析其他具有类似特征的环境数据集时也可能有用。
