Advancements in environmental statistics concerning multiple data sources

This work considers three different applications of environmental spatial statistics. Two of them relate to integrating data coming from different sources, while the other does the opposite and instead breaks down an aggregated response into several components of interest. The contributions of this work are separated by application into three parts. The first part is motivated by the interest in studying the biotic responses of species during the Last Glacial Maximum (LGM) due to rapid anthropogenic climate change. During this period, species retreated to highly spatially restricted geographic regions where survival was possible, known as glacial micro-refugia, from which they migrated and expanded when conditions became more suitable. Several distinct sources of evidence have contributed to developing a new understanding of how these regions might have impacted the sustainability of the natural populations of many species. Pollen records in Eastern Beringia (EB) have been used to explore the possibility that the region harbored glacial refugia for several plants from the arctic tundra and/or the boreal forest biomes common to the region. Our study focuses on alnus viridis and picea glauca, two predominant species of arcto-boreal vegetation. We propose to integrate genomic, SDM, and existing fossil data in a hierarchical Bayesian modeling (HBM) framework to determine whether multiple refugia existed in isolated geographic areas. This study demonstrates how the flexibility of HBMs makes the formal synthesis of such disparate data sources feasible. Our results highlight the regions of plausible refugia that can guide future investigations into studying the role of glacial refugia during climate change. The second part reverses the data integration of the first in hopes of utilizing present technologies better for the purposes of crop monitoring. The amount of carbon assimilated by plants through photosynthesis, called Gross Primary Productivity (GPP), is the largest carbon flux between the terrestrial biosphere and the atmosphere and, if quantified accurately, can grant insights into understanding several ecosystem functions as well as the impact of climate change to crop yields, in particular corn and soy. Recently, satellite-based measurements of solar-induced chlorophyll fluorescence (SIF) have been used as a strong proxy to measure GPP. SIF values will depend on the type of vegetation land cover; thus, the observed values can be decomposed into the specific vegetation type components to obtain its particular SIF yield information. We propose to implement a spatially varying coefficient regression model where the coefficients represent the specific SIF yields. For each land type coefficient, we induce spatial smoothness by penalizing the square deviations among adjacent sites according to some data-driven threshold value. The adjacent sites are chosen according to a minimum spanning tree (MST) in order to reduce redundancy in site pairing. Special characteristics of the data impose additional challenges, such as a non-negativity constraint on the estimations, as well as the presence of deterministic information that changes the structure of the MST. This study is able to retrieve accurate and fast results for the two main crops of interest when compared to other similar methods. Finally, the third part returns to data integration in handling the change of support problem in spatial statistics. Environmental applications are highly dependent on accurate and complete datasets of world temperature. However, the recollection of these datasets is limited by technology and poses additional challenges that must be handled before doing any environmental analysis. In particular, the change of support problem occurs when trying to combine data that has been collected at different resolutions. Some data is collected in situ, whereas other can be collected by satellites that cover wider areas. This study uses INLA to accurately handle data coming from the Integrated Global Radiosonde Archive (IGRA) and from the TIROS Operational Vertical Sounder (TOVS), both from the National Oceanic and Atmospheric Administration (NOAA), from 1990 to 1993. Both datasets are considered to measure the same latent process but in different ways that must be integrated to produce a complete picture of global temperature. This must also be done close to real-time, so an alternative model is also proposed to speed computations at the cost of accuracy. We are able to obtain similar results from both approaches as well as provide accurate uncertainty measurements for both. These results can then be used for future applications of environmental studies.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Statistics

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Campos, Mauricio

Contributors dc:contributor

• Li, Bo
• Simpson, Douglas
• Park, Trevor
• Bravo de Guenni, Lelys

• Data Integration
• Inla
• Spatial Process
• Glacial Refugia
• Solar-induced Fluorescence
• Spatially Varying Coefficients
• Remote Sensing

Statement dc:rights

• Copyright 2023 Mauricio Campos

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/121468

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F121468