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.
Application of Generalized Additive Models on High-Density Time Series Sensor Data in Horticulture & Generalized Non-Linear Mixed Model (GNLMM) Estimation and Inference with Pseudo-Likelihood
This dissertation consists of two sections. The first section focuses on application of generalized additive models (GAM) on high-density time-series sensory data in horticulture. The proposed methodology is studied in the context of a greenhouse experiment with the growth of basil plants. Sensors were placed in the soilless mix to record volumetric water content, electrical conductivity, and mix temperature at frequent time intervals. This results in large amounts of data in multiple dimensions. Efficiently extracting time-related patterns from such dataset is challenging. A series of generalized additive models (GAMs) are implemented to study hidden trends to predict basil growth. With the GAM, we can detect recurring daily cycles in different growth stages; extract representative daily patterns of growth stages; learn about the time-related correlation among features. The information obtained from GAMs was then linked to a plant growth response, specifically changes in height. The second section discusses generalized nonlinear mixed model (GNLMM) estimation and inference with focus on pseudo-likelihood (PL) algorithm. GNLMMs can account for nonlinear form of fixed and random effects in the predictor with data assumed to follow various distributions, not just normal. GNLMM has application potentials in medical studies and agricultural science. PL and integral approximation are among the methods to estimate GNLMM. Integral approximation has been well studied and can be implemented in both SAS and R, though with some limitations. However, research concerning the application of PL on GNLMM and corresponding software development is scarce. In this section, the PL algorithm for GNLMM is derived, and an R package is developed to implement it. The computation stability, estimation accuracy of PL and quadrature are compared for different types of simulated data (counts, proportions) with two kinds of nonlinear trend (asymptotic, S-shaped). It is found that PL can achieve accurate estimates of the fixed effects, and less biased variance component estimates than those obtained by integral approximation for some data scenarios.
Wu, Qianmei, "Application of Generalized Additive Models on High-Density Time Series Sensor Data in Horticulture & Generalized Non-Linear Mixed Model (GNLMM) Estimation and Inference with Pseudo-Likelihood" (2021). ETD collection for University of Nebraska - Lincoln. AAI28489887.