Data and analysis of vegetation structure across elevations in the Tropical Andes

We established eight locations at ~500 m-elevation intervals (223-4053 m) along a ~3800 m elevation gradient on the Eastern slopes of the Ecuadorian Andes, between 0.5 to 0.7° S latitude. The locations encompassed 7 ecosystem types (Sierra, 1999), most in national parks or reserves. At each location (June-August 2019), we set up six to eight 100m-transects, separated from one another by at least 0.5 Km. In order to maximize the diversity of habitat types surveyed, two of the transects were set up along the forest edge (road or river edge), the rest in the forest interior. We used land surveys to assess (a) total height and vertical and horizontal fill of the vegetation along the transects; (b) complexity in the vertical arrangement of fill patterns; (c) diversity of vegetation structural types at three scales (plant life form; type of structure; texture) along a horizontal axis; and (d) complexity in the overall vegetation configuration at stand and microhabitat levels using digital photographs. Here, we used vegetation height to represent vegetation volume; vegetation fill to represent vegetation structural density; and all other measurements to represent different aspects of vegetation structural complexity.
We used a rangefinder (Nikon Forestry Pro) to estimate stand height and vertical vegetation fill along the transects. We estimated canopy height at 10 m intervals and vertical vegetation fill at the 0, 50, and 100 m points along the transects (referred to as ‘random’ points). At these points, we assessed (to a precision of 10%) the percent vegetation fill, from the ground to the canopy in 1 m3 voxels. As these points often fell in open space, we also estimated, from the ground to the canopy in 1 m3 voxels, the fill of epiphytes growing within a 30 cm buffer area around the trunk of the closest tree to each point (referred to here as ‘closest tree’). In locations above 3000 m where vegetation such as mosses and “almohadillas” typically grow hunkered down against the ground, we considered the ground to be below such plants in order to account for them in vegetation fill measures. For both the random points and closest tree, we estimated the total vegetation fill along a transect by multiplying average canopy height times the average fill per m3 voxel times the 100m transect length. For epiphytes, we estimated the total fill at each transect by multiplying the average tree height times the average vegetation fill over trees times the number of trees per 100m.
We quantified vertical fill patterns using the effective number of layers method, which, as with species diversity indices, uses information theory measures that take into account stand height (number of 1m “layers”) and the fill of the various “layers” by plant material (“effective” number of layers). For each vertical column of vegetation fill at random points and the closest tree, we calculated the effective number of layers as the Shannon entropy index on the vertical vegetation fill measurements across all 1 m3 voxels.
We also used the line intercept method to characterize the diversity of structural forms along the transects at three spatial scales: plant life-form (tree, shrub, herbaceous, rosette, grass, liana, vine, moss, mushroom), type of structure (trunk, stem, root, leaf, flower, fruit, moss, and mushroom), and type of micro-structure (smooth, rugged, grooved, reticulate, spiny, barbed, bumpy, hairy). For every element that touched a 100m tape along a transect (waist height, at the lower elevations; ankle, above 3000 m, where most vegetation was below 1m) , we recorded the type of structure at each of the spatial scales. With data on the diversity and relative abundance of structural types at each scale, we then calculated Shannon entropy indices per transect at each of the three spatial scales.
Following the method of Proulx and Parrot (2008), we used digital photographs to characterize the structural complexity of the vegetation at stand and microhabitat levels. Proulx and Parrot applied information theory measures to assess the complexity captured across pixels in vegetation photographs at two spatial scales. They showed that complexity in light intensity variation at the microhabitat scale correlates with plant diversity and that variation in chroma and light intensity in stand-level photographs picks up distinct signatures for different seasons. We took stand- and microhabitat-level photographs at 0, 50, and 100 m along the transects. Stand-level photographs were taken at chest height, perpendicular to the vertical line, and in the four cardinal directions. We took micro-habitat photographs of the closest vegetation element at each of these locations, also at chest height and in the four cardinal directions. We placed the camera 30-cm away from the subject, centering it on the object as much as possible to fill a field of view of 50 x 37.5 cm. In transects where the vegetation did not grow taller than chest height (e.g. many above 3000 m), we took the photographs at knee or ankle height. We took additional photographs of vegetation elements along the trunks of the tree closest to the 0, 50, and 100 m marks along the transect. Photographs along tree trunks were taken at 5 m intervals from the ground to the top of the tree using the camera’s zoom to photograph a field of view of approximately 50 x 37.5 cm, independent of the height that was being photographed. Photographs were 16.1 MP (4608 x 3456 P) taken with a Canon PowerShot SX60 HS using 400 ISO, a F3.4 aperture, and the aperture time, recorded for each photo, set to automatic. We took the photographs on rainless days, recording how sunny it was on a 1-4 scale: (1) overcast; (2) overcast, with no projected shadows, but it is possible to distinguish the position of the sun; (3) partially overcast, but it is possible to see projected shadows; and (4) the sun is fully exposed. We also recorded the day and time of each photograph and used this information to calculate the sun’s azimuth and zenith. We included exposure time, lens aperture, zenith, azimuth, and sun index in statistical models to control for these factors.
Following Proulx and Parrot (2008), we calculated the Shannon entropy index on various measures of heterogeneity and spatial distribution of pixels in the hue (H), chroma (S), and intensity (V) channels of each photograph. These measures included dominance (or marginal entropy, ME), which is based on the frequency of pixel values, independent of location, and contagion (or joint entropy, JE), which is based on the pattern of 2x2-pixel matrices. From the standardized difference between JE and ME in each photograph we calculated the mean information gain (MIG) for each channel. MIG ranges from zero, for uniform patterns, to 1, for random ones (Proulx & Parrot, 2008). The degree to which the slope of the relationship between ME and MIG deviates from 1 is indicative of non-random spatial structure in the images (Proulx & Parrot, 2008). In analyzing the channels, we tried various combinations of photograph resolution and number of bins (N) in which to group pixel values at the HSV channels and chose the combination that maximized the standard deviation of MIG values across photographs (10% resolution and N=10). As complexity arises from structured disorder, it is expected to peak at intermediate values of MIG (Proulx & Parrot, 2008). To assess the MIG values at which this occurs, we calculated the related mean mutual information index (MMI), which, opposite to MIG, yields zero for complete randomness and 1 for complete uniformity (Proulx & Parrot, 2008). The product of these two indices (MIG x MMI) yields a convex function, Γ, which peaks at maximum complexity (Proulx & Parrot, 2008). We used Γ values, scaled to range between 0.0 and 1.0, as our measurement of structural complexity in the photographs. To facilitate interpretation, we averaged Γ values across the HSV channels to obtain a single measure of complexity per digital photograph.