Degree Type

Creative Component

Semester of Graduation

Summer 2019

Department

Statistics

First Major Professor

Dr. Jarad Niemi

Degree(s)

Master of Science (MS)

Major(s)

Statistics

Abstract

The main objective of this research is to build an emulator (i.e., surrogate) for the (Water Erosion Prediction Project) WEPP computer model, a soil erosion prediction technology used by the United States Department of Agriculture (USDA). The WEPP is a continuous simulation computer program that predicts soil loss and sediment deposition by considering various functional, quantitative and categorical inputs.

The emulator is built using Gaussian processes (GP) with both scalar and functional inputs. Three different GP models: (GP with scalar inputs, GP both scalar and functional inputs and GP with functional inputs) were employed and trained on the WEPP simulated data. Weight and nugget parameters in the covariance matrix of the GP model were estimated by the maximum likelihood method i.e eBayes approach. Particularly, we assumed weight parameters are built on trigonometric basis vectors.

GP model with functional inputs showed the best performance, while it's is the most computationally expensive among three approaches. This model predicted almost the same values as true response values for the inputs from the WEPP simulated training dataset.

Copyright Owner

Kuttubekova, Gulzina

Creative Commons License

Creative Commons Attribution 4.0 License
This work is licensed under a Creative Commons Attribution 4.0 License.

File Format

PDF

Share

COinS