Date of Award


Degree Name

Doctor of Philosophy


Environmental Science and Engineering


Craig E. Tweedie


Ecosystems are responding to a variety of human-induced, interlinked stressors that have emerged from changing climate, alteration to the global water cycle, sea-level rise, and land use and land cover change, among others. Quantifying these changes and their associated impacts on ecosystems requires a huge amount of long-term data. Due to advances in data collection techniques, such as remote sensing platforms, environmental sensors, synthesized datasets, and various software technologies, the volume and variety of long-term ecological data being collected has tremendously increased. Although there are several complex models and analyses that are increasingly parametrized with data from such sensors, there still exists a huge gap in managing, analyzing, visualizing, integrating, and sharing ecological data. The overarching goal of this dissertation is to develop ecoinformatics tools that will contribute to the advancement of global change science through: I) mitigating the challenges of new infrastructures for Big Data archiving, management and sharing, and analysis by developing a flexible system that supports multiple and novel data usage and visualization and II) attempt to utilize multi-sensor cross-correlation to detect rare soil moisture events in temporal data using some Machine Learning and Deep Learning (DL) models. To actualize the first objective, we developed webâ??based analytic tools capable of integrating spectral reflectance data from multiple instruments in the NASA Arctic-Boreal Vulnerability Experiment (ABoVE) study region using an open-source software â?? R shiny. R-HyperSpectral will help to dynamically view, interact, and discover optical properties of boreal and tundra plant communities. We also developed a multi-data fusion tool called rDataFusion, which is capable of aggregating heterogeneous data sets collected from a range of automated and semi-automated sensors and manual observations over a decade-long period. rDataFusion was developed using R shiny. Lastly, to achieve the second objective, we deployed several Machine and Deep Leaning techniques for optimal rare soil moisture events detection in the Chihuahuan Desert, New Mexico. Specifically, the machine and deep learning techniques used for this study include both classification and regression methods, including a Decision Tree Classifier (DTC), Logistic Regression Classifier (LR), Random Forest Regression (RF), and the Long Short-Term Memory (LSTM) method of Artificial Neural Network (ANN). Of all these methods used, the DTC performed the best, with prediction accuracy of 88.8%, closely followed by LSTM model with 88%. The LR recorded a prediction accuracy of approximately 80%. Lastly, through the tools we developed, data will be available for ecological and environmental science researchers to analyze and further understand ecosystem changes over multiple temporal scales and levels of biological organization and interaction. Furthermore, the analysis and prediction of rare soil moisture events in the dryland ecosystem unveils a pathway to understanding soil moisture events and the key drivers in drylands.




Recieved from ProQuest

File Size

124 p.

File Format


Rights Holder

Ifeanyi H. Nwigboji