Dissertations, Theses, and Capstone Projects

Date of Degree

2-2017

Document Type

Dissertation

Degree Name

Ph.D.

Program

Earth & Environmental Sciences

Advisor

Ines Miyares

Committee Members

Sean Ahearn

Andrew Maroko

Subject Categories

Environmental Monitoring | Physical and Environmental Geography | Remote Sensing

Keywords

Land history, Multi-temporal Landsat data, Google-Earth, Oil Palm Plantation, Mill, LC/LD Model

Abstract

The aim of this study is to reconstruct the history of land conversion to oil palm plantation in tropical Asia using multi-temporal satellite data. A new method was constructed with a newly developed computer model, Land Change Detection and Land Definition Model (LC/LD Model) to map out spatio-temporal distribution of land changes. A comprehensive, cloud-free Landsat dataset was created from all the available Landsat data from 1988 to 2015. The pixel-based dataset was converted into a polygon-based dataset by applying the multi-temporal image segmentation method. The representation of the spectral information was also reduced to a single index of IB45, the ratio of the near-infrared (Band 4) to mid-infrared (Band 5) bands, which was the most suitable index for detecting and tracking the transformation of land to oil palm plantation. To extract targeted land changes and land uses from a given temporal profile, land change scenarios were assumed and temporal segmentation method was developed for Land Change Detection Model (LCM). The segmented profiles were then evaluated by using bio-physical metrics in the Land Definition Model (LDM) to determine the land uses. The two-tiered LC/LD Model could detect not only large-scale land changes caused by private companies but also small-scale changes caused by smallholders, which is supposedly the most uncertain factor for the future development of oil palm plantation. Relationships between local factors and two land change phenomena, conversion to oil palm plantation and deforestation, were investigated using quantitative assessments such as Logistic Regression analysis. The results explicitly showed the positive impacts of proximities to 1) pre-existing oil palm plantations and 2) nearest mill and negative impacts of 1) elevation and 2) slope on the occurrence of small oil palm plantations. These findings strongly imply that oil palm development in the neighborhood initiated further development in nearby areas. The accessibility to mills also increased the chance of oil palm development. From topographical aspects, flat and low altitude land was more favored than steep and high altitude one. The results also indicated that large size enterprise plantations were more responsible for directly converting untouched natural land than smallholders and were the main contributor to deforestation. In contrast, smallholders mainly converted preexisting farmland to oil palm plantations.

Share

COinS