Forest degradation is one of those elusive, hard-to-estimate concepts that is widely used in ecological literature. Ghazoul et al. (2015) suggested that a good definition of degradation should incorporate the ecosystem processes that underlie forest dynamics, like regeneration and tree mortality. Proper estimations of these processes for a landscape provide powerful information to link ecosystems processes and plant community dynamics that are usually measured at the plot level. These estimations are desirable in tropical areas where fast-pace forest degradation urges to have strategies for forest conservation. Here we tested the potential of canopy surface metrics derived from high-resolution aerial imagery to estimate forest mortality and regeneration. We hypothesized that differences on estimated metrics hold high correspondence with differences on structural and functional metrics derived from plot-level data. To test this hypothesis, we used orthophoto mosaics (using Unmanned Area Vehicles) to monitor changes in canopy cover and NDVI values in two Tropical Dry Forest landscapes in Colombia, which were previously assigned to a high or low degree of degradation based on plot and landscape-level metrics. For each mosaic of a site x period combination we conducted a spatial wavelet analyses (similar to Strand et al 2006) to quantify degradation in terms of spatial and temporal variability of canopy height, NDVI values, and gap sizes. Aerial photographs were obtained at four moments during dry and wet seasons between 2015 and 2016.
The distribution of canopy gaps sizes were not different between the two areas (F=3.184, p-value=0.0456), but they showed significant differences in their variance and their spatial pattern (SDhigh-degradation=0.1487 SDlow-degradation=0.4430, F-test p-value=0.05); gaps in the high-degradation area (HAD) were randomly distributed (z-score of 0.67800), their location did not correlate with topographic variables (R2-adjust <0.1 and non-significant F-statistics for all variable tested), and showed higher temporal variability (preliminary analyses). In contrast, canopy gap in low degradation areas (LDA) showed lower variance of size, dispersed spatial distribution (z-score of 2.6494), and low but significant correlations with some topographic variables (R2=0.2 p-value<0.001 for slope). Analyses of wavelet coefficients showed that NDVI signals HDA are less predictable (% of zero coefficients ranges from 25-50% for HDA vs [10-13%] for LDA). HDA showed higher heterogeneity in forest canopy surface, NDVI values and canopy gap patterns. These results demonstrate the utility of landscape-scale metrics derived from high-resolution low-cost imagery as indicators of forest processes such regeneration and mortality.