Fig 1.
(A) Extent indicator of study location in northern Manitoba, (B) supratidal salt marsh study location within Wapusk National Park, (C) sample RGB photo of habitat surveyed by drone. Image acquired at 75 m above ground level.
Table 1.
Proportion land cover type classification of drone (Trimble UX5) imagery at three altitudes.
Proportion values are obtained from the enumeration of pixel types for each land cover class across all 5 plots. Accuracy and kappa statistics presented as mean ± SD, along with the range of values.
Fig 2.
Comparisons between RGB drone imagery and classified product.
(A) Example RGB imagery at 120 m AGL (B) final classified image. Post processing tools failed to eliminate the patch of darker barren surface and incorrectly classified the patch as non-shrub vegetation (indicated by the red circle). Shadows along the edge of the vegetation patch were improperly classified as shrubs (indicated by red arrows).
Fig 3.
Plotted proportional values and Pearson’s correlation coefficients between three methods of data acquisition (ground transects, drone transects and drone pixel counts).
Data presented for each cover type (barren, non-shrubs, and shrubs). Each point represents proportional cover data collected within the same cell (n = 92) for each method. Drone imagery collected at 75 m AGL. Red dashed line represents 1:1 relationship.
Fig 4.
Mixed model estimates of proportion land cover type from three different methods of data collection.
Ground transects data collected as linear transects, drone transects are the same transects overlaid on classified drone imagery (see Methods section) with land cover values were extracted every meter, and drone pixel counts based on the enumeration of pixels for each land cover type as a proportion of all pixels in each cell. Drone estimates made from imagery collected at 75 m AGL. Cover data obtained from 92 cells across 5 study plots.
Table 2.
Coefficient estimates from Beta GLMM for each cover type (barren, non-shrub, and shrub) as measured by drone pixel counts at altitudes of 75, 100 and 120m AGL.
Estimates obtained from 92 observations (cells) across 5 different study plots.
Table 3.
Coefficient estimates from Beta GLMM for each cover type (barren, non-shrub, and shrub) as measured by drone transects at altitudes of 75, 100 and 120m AGL.
Estimates obtained from 92 observations (cells) across 5 different study plots.
Table 4.
Coefficient estimates from Beta GLMM for each cover type (barren, non-shrub, and shrub) as measured by ground based transects, drone based transects and drone pixel counts at 75 m AGL.
Estimates obtained from 92 observations (cells) across 5 different study plots.