<title>Evaluation of satellite change detection methods for monitoring forest harvest and silvicultural activity in Maine industrial forests</title>

Abstract

ABSTRACT Change detection is a major application of satellite remote sensing. The idea is to analyze change in spectral patternsover a particular geographic area at different points of time. The information might be gathered by different satellite platforms(multi-sensor), in various wavebands (multi-spectral) and on several acquisition dates (multi-temporal). For forestry field applications, change detection might provide useful information for forest resources management, inventory, evaluation, planning, and monitoring. This study incorporated a multi-temporal approach for detecting forest change due to clearcut, partial cut andrelease operation treatments in a Maine study area. Most forest change detection studies include only two dates of imagery.However, in this investigation, three date satellite images from 1983, 1988 and 1991 were examined simultaneously in a singlestep analysis approach. Two change detection methods, the Normalized Difference Vegetation Index (NDVI) and the PrincipalComponents Analysis (PCA) were evaluated and a new method, Principal Factor Analysis (PFA) was introduced. A maximumlikelihood classification algorithm was used to categorize change/no change events and the results were compared to a foreststand exam and history database. The Khat statistic was chosen as the criteria to evaluate the accuracy of each classificationmethod while pairwise significance tests were constructed to compare results between methods. The Standardized variant ofPrincipal Factor Analysis (SPFA) produced the best results followed by Principal Components Analysis and NormalizedDifference Vegetation Index.Keywords: Satellite Change Detection, Landsat, Forest Monitoring, Normalized Difference Vegetation Index, PrincipalComponents Analysis, Principal Factor Analysis, Accuracy Assessment