Probability of Land Cover Classification Estimates for the Kenai Peninsula Lowlands; 1973, 2002, and 2017

Abstract

These raster images represent continuous surfaces of estimated land cover types for the western Kenai Peninsula circa 1973, circa 2002, and circa 2017. This raster image represents the coded series of estimated land cover types for each pixel over the three distinct time periods. The estimated land cover types (Needleleaf Forest, Mixed Forest, Broadleaf Forest, Herbaceous, Wetland, Alpine, Barren, Shrub, Water) were originally derived from a random forest classifier executed in R (version 3.5.0). Predictor variables from training data included known landcover types deduced from high resolution aerial imagery, summer and winter spectral indices obtained from historical Landsat scenes, and topographic parameters derived from a digital elevation model. For each era (c. 1973, c. 2002, and c. 2017) 3,600 training points (400 points for each land cover type) were randomly distributed within training areas and training areas were opportunistically distributed to capture the regional and geomorphic extent of each land cover type to the extent possible given availability of aerial imagery. Each training point was assigned feature list values from the Landsat mosaics and digital elevation model while land cover was manually interpreted using high-resolution areal imagery. Model output included predicted landcover type and a corresponding probability score and were rasterized for each era – one raster image featuring land cover types and one raster image featuring land cover type probability. This raster image characterizes a coded series that represents the unique combinations (or series) of vegetation types that pixels experienced over time. Each cell's series value corresponds to a series of numbers (#_#_#) where the first number is the vegetation code from 1973, the middle number is the vegetation code from 2002 and the last number is the vegetation code from 2017. Raster values (1-687) can be translated into series using the vegetation_code_series_table.csv. Series numbers can be translated into vegetation types using the vegetation_class_table.csv. The code series raster image allows for the graphical depiction of certain landcover change types such as deforestation (e.g. Forest_Forest_Shrub) and has 30 meter resolution.