Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Geostatistical Methods for Spatio-Temporal Forecasting and Spatial-Causal Inference
This dissertation focuses on problems in spatio-temporal forecasting and causal inference with geostatistical data. It explores three specific problems in this area: spatio-temporal forecasting, spatial confounding, and spatial-causal inference.Chapter 2 introduces a novel approach for site-specific yield forecasting. Precision agriculture involves analyzing spatio-temporal data to identify within-field variability and to help optimize resources. We propose a two-stage approach for forecasting site-specific yield in the case of a short time series and high-resolution spatial data. We implement a clustering approach to perform noise reduction followed by a Bayesian hierarchical model to obtain yield forecasts. Implementing our proposed method at three different sites in Nebraska, we demonstrate that our method provides fine resolution yield maps that accurately represent the observed yield maps.Chapter 3 investigates the effect of spatial confounding on the bias and precision of regression coefficients estimates. Spatial confounding is used to describe the multicollinearity among the spatial covariates and the spatial random effect in spatial generalized linear mixed models. While there is extensive literature studying the effect of spatial confounding bias in the case of continuous covariates, not much work has been done in scenarios where the covariate of interest is binary. Since theoretical derivations for bias are hard to come by under the binary covariate setting, we resort to numerical studies. We show that spatial smoothness plays an important role in understanding spatial confounding bias.Chapter 4 proposes a novel method to adjust for the spatial confounding bias under a spatial-causal inference setting. We propose an improved doubly robust estimator to obtain causal estimates by combining tools in spatial statistics and causal inference literature. We demonstrate that our proposed method balances all observed confounders and adjusts for the unobserved confounders through simulation studies. Our method has the lowest bias and the lowest standard error under most settings compared to alternatives. We show that adjusting for spatial confounding is essential for reliable inference by analyzing the effect of emission control technologies on ambient ozone concentrations.
Statistics|Environmental engineering|Geographic information science|Atmospheric sciences|Agricultural engineering
Pokal, Sayli Vijaykumar, "Geostatistical Methods for Spatio-Temporal Forecasting and Spatial-Causal Inference" (2021). ETD collection for University of Nebraska-Lincoln. AAI28862620.