Date of Award


Document Type

Campus Access Dissertation

Degree Name

Doctor of Philosophy (PhD)


Computer Science

First Advisor

Wei Ding

Second Advisor

Dan Simovici

Third Advisor

Nurit Haspel


Abundant satellite and ground observation data have been analyzed to extract hidden patterns from environmental study data to assist in applications such as the prediction of heavy precipitation and estimation of future crop yield. However, the high-dimensional spatio-temporal feature space, which is composed of variations of multiple variables and spatial, temporal influences, can result in intensive and expensive calculations. The non-linear and location-dependent analytic forms of relationships between independent and dependent variables are difficult to obtain from those data sets because of the complex physical/biological processes involved.

In this dissertation, we propose a machine learning framework named Non-Linear Spatio-Temporal Modeling (NSTM), which includes feature selection, feature optimization, and prediction to develop efficient models with less computational complexity, but that retain the characteristics of the data. Subspace Learning with Streaming Feature Selection (SLSFS) is modeled by Nearest-Sample Choosing (NSC) and Online Streaming Feature Selection (OSFS) to handle the imbalanced samples and high dimensional feature space. Following this, subspace learning is performed. A Non-linear Feature-Optimized Neural Network (NFO-NN), built by the Delta Feature Construction (DFC) process, is then used to obtain the deviation and normalize each feature, and finally a two-layered neural network non-linearly estimates the bias between two different time slots.

Two real-world cases have been studied to evaluate the proposed methodologies: 1) predict the daily and monthly extremely heavy precipitation with SLSFS, and then discover physical meanings between the hydro-climate teleconnection signals and terrestrial precipitation events, and 2) estimate the crop yield/loss, which is mostly affected by the precipitation, with NFO-NN. Both of our models have produced efficient and accurate results.


Free and open access to this Campus Access Dissertation is made available to the UMass Boston community by ScholarWorks at UMass Boston. Those not on campus and those without a UMass Boston campus username and password may gain access to this dissertation through resources like Proquest Dissertations & Theses Global or through Interlibrary Loan. If you have a UMass Boston campus username and password and would like to download this work from off-campus, click on the "Off-Campus UMass Boston Users" link above.