First draft of graphics and tables for shrubland paper
A living document with set pieces for the shrubland paper.
Mike Mahoney true
2022-01-23
Figures
NY Population
Figure 1: Population per square kilometer for each census block group in New York State, as reported in the 2019 5 year American Community Survey.
Probability by height
Figure 2: Predicted probability, from the logistic ensemble model, of shrubland as a function of pixel heights. Each point represents one of one million pixels randomly sampled from all LiDAR coverage areas. Shaded area between vertical lines represents the 1-5m height threshold used to define “shrubland” for this study. Trendline shown is a generalized additive model fit using penalized cubic regression splines.
Probability by class
Figure 3: Smoothed kernel density estimates of predicted probability of shrubland for both shrubland and non-shrubland pixels, calculated using a random sample of 1,000,000 pixels taken from the LiDAR patchwork prediction surface using the logistic ensemble model. Vertical lines indicate each of the four probability thresholds used to classify pixels. Colors represent the correct classification of the pixel.
LiDAR Shrub Map
Figure 4: Identified shrubland areas within each available LiDAR coverage. Shrubland was defined at a 1 meter resolution as being any area within a vegetated LCPRI land cover class and below 1067 meters elevation with a LiDAR-derived height between 1 and 5 meters. 30 meter pixels, used for analysis and modeling, were then defined as shrubland if more than 50% of their contained 1 meter pixels were classified as shrubland. In total, approximately 2.5% of 30 meter pixels were classified as shrubland.
LiDAR probabilities (modeled)
Figure 5: Predicted probability of shrubland for the boundaries of all used LiDAR coverages, from the logistic ensemble model. Predictions were made using data reflecting the same year as LiDAR acquisition; the map therefore represents a temporal patchwork of predictions. Pixels in non-vegetated LCPRI land cover classes (developed, water, ice/snow, and barren) or above 1067 meters in elevation were not mapped and are shown in white.
LiDAR Classifications
Figure 6: Predicted shrubland locations within each LiDAR coverage, from the logistic ensemble model. Predicted pixel probabilities were classified using either the Youden-optimal threshold (which maximizes both sensitivity and specificity) or a threshold chosen to target a certain level of specificity, using thresholds derived from the validation data set. Predictions were made using data reflecting the same year as LiDAR acquisition; the map therefore represents a temporal patchwork of predictions. Pixels in non-vegetated LCPRI land cover classes (developed, water, ice/snow, and barren) or above 1067 meters in elevation were not mapped and are shown in white.