Withdraw
Loading…
Information driven causal structure, evolution, and functionality of complex biosphere - atmosphere systems
Hernandez Rodriguez, Leila Constanza
This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/115507
Description
- Title
- Information driven causal structure, evolution, and functionality of complex biosphere - atmosphere systems
- Author(s)
- Hernandez Rodriguez, Leila Constanza
- Issue Date
- 2022-01-26
- Director of Research (if dissertation) or Advisor (if thesis)
- Kumar, Praveen
- Doctoral Committee Chair(s)
- Kumar, Praveen
- Committee Member(s)
- Dominguez, Francina
- Bernacchi, Carl
- Jain, Atul
- Gupta, Hoshin
- Department of Study
- Civil & Environmental Eng
- Discipline
- Civil Engineering
- Degree Granting Institution
- University of Illinois at Urbana-Champaign
- Degree Name
- Ph.D.
- Degree Level
- Dissertation
- Keyword(s)
- Hydrology
- causality
- information theory
- causal structure
- evolution
- structure-function
- biosphere-atmosphere interactions
- land cover heterogeneity
- flux footprint
- eddy covariance
- turbulence
- complex systems
- Abstract
- At the biosphere-atmosphere interface, an eddy covariance flux tower acts as an integrator of information from several processes occurring in the ecosystem at different space and time scales. These high frequency observations, typically at 10Hz or 20Hz, contain information about multiple variables such as wind speed and direction, humidity, temperature, water vapor (H2O), and carbon dioxide (CO2). These variables together capture a complex system of interactions where time-dependent non-linear interdependencies sustain the dynamics and are able to influence the behavior of each component. Such interdependencies arise due to the dynamic connectivity of the system through which fluctuations in one variable propagate to other variables at a lagged time. At the tower, the near-instantaneous measurements are averaged for the estimation of the exchange of fluxes using the eddy covariance technique, resulting in continuous measurements of the ecosystem-scale scalar exchange of heat, water, and CO2. These observations are used in research and management worldwide, mainly to understand the temporal and spatial variations at the ecosystem scale. However, these observations also carry information about the influence of the composition inside its flux footprint, the temporally dynamic source/sink area that contributes to the measured fluxes. It also contains causal information regarding the dynamics of multivariate interactions at turbulence scales, which has not yet been explored in a multivariate context. In this dissertation, the primary goal of the research is to understand how the information carried in the observations from an eddy covariance tower captures changes in the structure, evolution, and functionality of the system at different scales. The specific goals of this dissertation are to (i) determine in what way the composition of the heterogeneous organization of the land cover inside the flux footprint contributes to the observations measured at the tower; (ii) develop a framework for the study of the evolution of the causal structure of multivariate interactions at the turbulence scale by exploiting the high frequency measurements, and (iii) determine the relationship between the evolution of the causal structure and its relationship to the system’s functionality. Based on data from a 25m tall eddy covariance flux tower that sees maize (corn) and soybean fields in the agricultural Midwest of the US, which comprises a “patchwork quilt” of different crops, we determine how this mosaic crop pattern impacts the exchange of water, heat, and CO2 fluxes measured at the tower. Then, we determine the relative contribution of the different land cover types to the total flux. Our hypothesis is that the human-induced spatial reorganization of the landscape results in an exchange of heat, water and CO2 that is different from the exchange that would result if the vegetation species were randomly (statistically homogeneously) distributed. To address this hypothesis, we estimate the spatial extent contributing to every flux measurement from April 2016 to April 2019 and combine them with an ecohydrological model to obtain the temporally varying contribution of fluxes for different land covers. Then we estimate how they contribute to the overall flux, which depends on how the crop fields are distributed. We find that the tower mostly “sees" fields located between 168m and 268m away although areas as far away as several kilometers contribute to the observations. As a result of crop rotation, maize fields contributed more than soybean fields during the 2016 and 2018 growing seasons, and vice versa during 2017. We compare our results to a hypothetical case where all vegetation is randomly distributed on the surface. We find that the mosaic organization of crop fields that the tower “sees” account for up to 24.5% more CO2 flux () than if the vegetation was randomly distributed on the landscape. We find that the knowledge of footprint contributions combined with ecohydrological model results helps explain how the contributions from different crops to the observed fluxes at the tower vary from year to year. To address the second goal, we explore how the eddy covariance measurements of high frequency (i.e., 10Hz) multivariate data also contain information about the interdependency between interacting variables such as horizontal and vertical wind speed, humidity, temperature, and CO2. These measurements offer an opportunity to move beyond the traditional spectral analyses to explore causal dependency among them at turbulence scales. We visualize the temporal dependencies among lagged variables that illustrate its dynamics in a Directed Acyclic Graph (DAG) representation for time series. Causal associations among variables are represented by links in a network of interactions denominated “causal structure, whose representation is a DAG. We estimate the network of inter-dependencies using DAGs from multivariate high frequency time series and quantify its evolutionary dynamics. Our hypothesis is that there are different types of causal structures embedded in the day-time turbulence of the land-atmosphere exchange that can be recognized as the system evolves. To test our hypothesis, we develop a novel approach to find patterns of similar behavior using the DAGs, which is based on a distance-based classification to characterize the structural differences between the DAGs and a -means clustering approach that selects an unbiased number of clusters. We explore a large range of feasible dynamics of the system by using data during a clear sky day and during the solar eclipse in 2017. We compare the sequences of DAGs to investigate structural differences in causal interactions among the variables. We find that in this multivariate system, temperature, horizontal and vertical wind velocities, along with scalars such as CO2 and H2O, create a causal structure of interdependencies that sustain information flow and evolves inside a range of dynamical behavior instead of being in a fixed configuration. Our results show well-defined clusters of similar structural connectivity and the emergence of patterns of similar dynamics while the system evolves. This causal history approach for investigating the evolution of multivariate dynamics, combined with the novel approach of the evolution of the causal structure provides a methodological framework to understand how dependence in turbulence is manifest at high frequencies. Finally, we explore the relationship between the causal structure and its associated function, using the representative DAGs in each cluster of similar dynamics. We determine if functional differences are a reflection of structural differences, or in spite of structural differences, there are no functional differences. We based our analysis on the hypothesis that causal relationships are affected through information flow, and the interaction arising due to information flow sustains the dynamics of the ecohydrological system at high frequency, its functionality. To test our hypothesis, we build upon our causal structure analysis to estimate the information flow in each cluster using a multivariate causal history approach. We characterize functionality as the nature of interactions as discerned through redundant, unique, and synergistic components of information flow. Redundant information is overlapping information provided by multiple sources to a target, unique information is only provided by a single target, and synergistic information is provided only when two or more sources are present together. We find that in turbulence at the biosphere-atmosphere interface, the variables that control the dynamic character of the atmosphere as well as the thermodynamics, are driven by non-local conditions, while the scalar transport associated with CO2 and H2O are mainly driven by short-term local conditions. In the absence of a causal dynamical law that is able to describe the behavior of complex systems in the geosciences, we develop a novel approach to account for the evolution of the structural causal model. The approach implemented here for investigating the functionality of the system through the evolution of information flow, supported by the novel approach of the evolution of the system’s causal structure, provides a methodological framework to understand how a complex system evolves. In our implementation we use the information carried in the continuous time series of observations at a flux tower to better understand the behavior of multivariate turbulence at the biosphere-atmosphere interface. With the increasing availability of observational data at different temporal and spatial scales, and with the rapid advance of data-driven approaches based in observations, the implementation of causality in machine learning and other artificial intelligence (AI) applications would help us recognize causal patterns of behavior and improve the understanding of the dynamics of complex systems to inform Earth systems models, prediction, and ultimately decision-making.
- Graduation Semester
- 2022-05
- Type of Resource
- Thesis
- Copyright and License Information
- Copyright 2022 Leila Hernandez Rodriguez
Owning Collections
Graduate Dissertations and Theses at Illinois PRIMARY
Graduate Theses and Dissertations at IllinoisManage Files
Loading…
Edit Collection Membership
Loading…
Edit Metadata
Loading…
Edit Properties
Loading…
Embargoes
Loading…