Master's Thesis Presentation: Multi-Temporal Land-Cover Classification and Accuracy Comparison Using a Rule-Based Forest Succession Logic

Event Type: 
Timothy Johnson, MS, Environmental Informatics
Friday, April 11, 2014 - 2:49pm to 3:10pm
1028 Dana Building
Event Sponsor: 
School of Natural Resources and Environment

Presenter: Timothy Johnson, MS, Environmental Informatics

Advisers: Kathleen Bergen and Joshua Newell

Primorsky Krai is a unique area where a very rich and important ecosystem provides vital life to often endangered flora and fauna. This southern boreal forest is located near the Pacific Ocean providing a moderating climate for a unique blend of taiga and deciduous tree species. Forest management of this region has changed over the past 35 years, during the timescale of this analysis, as a result of the Soviet Union break up and new timber demand from nearby China. The Primorsky Krai region is specifically valuable due to its unique mammal species, notably the Siberian Tiger, the only location on earth where they are still found. It is very important to preserve this ecosystem.

To analyze how this forested region has changed, time-series Landsat data from 1976, 1989, 1999, and 2009 were classified using a hybrid classification method. Resulting maps indicate four important times during Russian history, during the Soviet Union (1976), near the time of transition (1989), in a post-Soviet transitioning economy (1999), and during a more recent time of new management practices and new global forest demands (2009). Maps were compared to analyze land-cover change over the three periods 1976-1989, 1989-1999, and 1999-2009. Change direction from one land cover to another were analyzed further and checked for illogical changes. By comparing multiple dates with one another using a combination of forest succession logic and a more general process of elimination, classification results were improved. This technique could prove useful with time series land cover analysis. Specifically, where very remote areas without ground truth data are classified, where shadowing effects from mountainous or hilly terrain present issues with results, and/or when illogical/noise changes occur for other broader reasons.