Author : Sara Zapata-Marin
Publisher :
ISBN 13 :
Total Pages : 0 pages
Book Rating : 4.:/5 (135 download)
Book Synopsis Land-use Regression and Spatio-temporal Hierarchical Models for Environmental Processes by : Sara Zapata-Marin
Download or read book Land-use Regression and Spatio-temporal Hierarchical Models for Environmental Processes written by Sara Zapata-Marin and published by . This book was released on 2022 with total page 0 pages. Available in PDF, EPUB and Kindle. Book excerpt: "Land-use regression is a popular method used to describe the spatial variability of different environmental processes using local variables. However, there are situations in which there might be some complex spatio-temporal structure left after accounting for land-use variables.In this work, three different Bayesian hierarchical models are proposed to model the spatial and spatio-temporal dispersion of air pollutants and aeroallergens within cities. Bayesian inference can easily accommodate complex interactions while naturally accounting for uncertainties in the estimation of unknowns in the model when performing predictions.In the first study, a spatial hierarchical model is used to analyze the concentration of volatile organic compounds (VOCs) in Montreal, Canada. The data consists of concentration measurements of five VOCs measured over two-week periods for three monitoring campaigns between 2005 and 2006 over 130 locations in the city. The five VOCs of interest are: benzene, decane, ethylbenzene, hexane, and trimethylbenzene. Four different models are fitted to each of the five VOCs. These models extend land-use regression by accounting for any spatial structure left after including the covariates while also capturing the across campaign variation through an indicator variable or campaign-specific coefficients. Predicted surfaces are obtained for each campaign. For all VOCs higher levels are found during the December campaign, and the predicted areas with the highest levels correspond to multiple sections of major highways.For the second and third studies, we have available data on the daily and weekly measurements of pollen concentration in Toronto, Canada collected in 2018. The measurements consist of tree, weed, grass, and total pollen concentration at 18 monitoring sites and were obtained daily for eleven of these sites and weekly for the other seven sites.In the second study, the weekly concentration of each of the four pollen types is modeled. Instead of considering the temporal window that only has positive values, that is, removing the zeros, a hurdle model is proposed to account for the high number of measurements equal to zero. This structure allows for the estimation of the probability of the pollen concentration being equal to zero at any given week, which provides further information on temporal windows with positive concentrations of the different types of pollen. Additionally, a dynamic linear model is used to capture the weekly trend of pollen concentration in the city.In the third study, the daily concentration of total pollen is modeled. Rather than aggregating the data to the weekly scale, a temporal misalignment model is proposed to account for the difference in scale and to take advantage of the daily measurements. Using the properties of dynamic linear models and the multivariate normal distribution, a spatio-temporal model to account for temporal misalignment is proposed. This model allows to estimate the fine-scale measurements at locations where only coarse-scale observations were available. Additionally, the model is fitted to artificial data with different temporal structures, including trend and seasonality.The predicted surfaces obtained in these three studies will help inform future health-related studies. Furthermore, the methods proposed here are flexible, easily adaptable, and can improve our understanding of similar environmental processes. All codes are publicly available such that the implementation of the proposed approach in similar situations is easily achieved"--
