Fig 1.
Northern Maine, U.S.A. study area with 5 km square sample landscape units superimposed. Harvesting trends and patterns of landscape change were calculated for forestland assumed available for harvest. Mapped forest composition classes demonstrate the spatial distribution of general forest types at the onset of our study period (1975). State and provincial boundaries displayed in the inset map were obtained from the National Atlas of the U.S. (Political Boundaries) and the Atlas of Canada (National Frameworks Data, Census Subdivisions and Population Ecumene).
Table 1.
Landsat images used to map forest harvesting (1973–2010) and forest composition (1975 and 2004).
Table 2.
Error matrix and accuracy estimates for validation classes aggregated from the 1973–2010 forest harvest time series.
Table 3.
Error matrix and accuracy estimates for the 2010 cumulative harvest map.
Table 4.
Error matrix and accuracy estimates for the 1975 forest type map.
Table 5.
Error matrix and accuracy estimates for the 2004 forest type map.
Fig 2.
Forest harvest trends and landscape change for two sample grid cells.
Mapped and modeled cumulative harvest time series for two arbitrary sample grid cells, expressed as a proportion of available forestland. The median cumulative harvest time series (n = 608) is shown for reference. Images of landscape conditions include cumulative harvest impact superimposed over the 1975 forest type map for a subset of time series dates, and the 2004 forest type map. Comparison of the 1975 and 2004 forest type data indicates areas where intervening harvests induced changes in forest type.
Fig 3.
Predominant patterns of harvesting.
(a) Dendrogram produced by agglomerative hierarchical clustering of modeled cumulative harvest time series. Six groups of landscape units were identified for subsequent analysis (sample sizes provided in parentheses). (b) Mean cumulative harvest area time series for each of the groups identified in (a), expressed as a proportion of available forestland. Vertical bars represent the interquartile range. Bars are provided at a subset of dates and are offset horizontally to improve visual clarity. (c) Spatial distribution of groups identified in (a). Hatching indicates outlying samples excluded from further analysis.
Fig 4.
Time series of regenerating and partially harvested forest area.
Cumulative time series of (a) regenerating forest area and (b) partially harvested forest area, expressed as a proportion of available forestland and averaged within groups identified by cluster analysis of modeled cumulative harvest time series. Vertical bars represent the interquartile range. Bars are provided at a subset of dates and are offset horizontally to improve visual clarity.
Fig 5.
Time series of intact mature forest configuration metrics.
Time series of cumulative change in (a) area-weighted mean patch size, (b) patch density, (c) edge density, and (d) area-weighted mean fractal dimension for intact mature forest, averaged within groups identified by cluster analysis of modeled cumulative harvest time series. Vertical bars represent the interquartile range. Bars are provided at a subset of dates and are offset horizontally to improve visual clarity.
Fig 6.
Time series of regenerating forest configuration metrics.
Time series of cumulative change in (a) area-weighted mean patch size, (b) patch density, (c) edge density, and (d) area-weighted mean fractal dimension for regenerating forest, averaged within groups identified by cluster analysis of modeled cumulative harvest time series. Vertical bars represent the interquartile range. Bars are provided at a subset of dates and are offset horizontally to improve visual clarity.
Fig 7.
Initial landscape composition and changes in composition, 1975–2004.
(a) Proportion of available forestland classified by 1975 forest type. (b) Change in forest type between 1975 and 2004, expressed as a proportion of forestland harvested prior to 2004 (negative values indicate loss; positive values indicate gain). Values were calculated for individual sample grid cells and then averaged within groups identified by cluster analysis of modeled cumulative harvest time series. Error bars represent the interquartile range.
Fig 8.
Changes in average harvest patch size through time.
Area-weighted mean patch size for (a) the stand-replacing harvest class and (b) the combined stand-replacing and partial harvest class, calculated at each time series interval. Values were averaged within groups identified by cluster analysis of modeled cumulative harvest time series. Vertical bars represent the interquartile range. Bars are provided at a subset of dates and are offset horizontally to improve visual clarity.