Recovery of lizard assemblages ten years after reduced-impact logging in central-eastern Amazonia

Understanding changes in species composition due to human-induced habitat modification and environmental filtering is essential for formulating effective conservation strategies. Species turnover resulting from reduced-impact logging (RIL) is expected in the short term, generally with species adapted to open areas replacing those dependent on old-growth forest. However, little is known about how RIL activities influence assemblages after the perturbation ceased. We sampled lizards across an edaphic and vegetation-structure gradient in 64 plots in the Brazilian Amazon to test the hypothesis that changes in assemblage composition and proportion of heliothermic species are due to canopy openness resulting from ceased RIL activities and individual tree falls, or to other environmental gradients. Contrary to expectations, canopy openness did not significantly affect the overall composition of lizard assemblages, but nearby unforested areas influenced assemblage composition, resulting in a higher proportion of heliothermic species. The composition of lizard assemblages was also significantly influenced by the distance to the nearest water body, vegetation height, and soil sand content. However, leaf-litter height did not have a detectable impact on the composition of lizard assemblages. We conclude that short-term changes in species composition due to habitat modification by RIL do not persist in the long term after the perturbation ceases, and the assemblages may recover as vegetation regenerates. Although lizard species respond to spatial and temporal variation in environmental characteristics, we found evidence that lizard assemblages recover as reduced-impact logging (RIL) activities cease and vegetation regenerates.

Methods
We sampled lizards across an edaphic and vegetation-structure gradient in 64 plots. In the southern portion of the study region, inside the Tapajós National Forest (TNF), we sampled 40 plots distributed across four research modules established as part of the Brazilian Program for Biodiversity Research (PPBio) for biodiversity monitoring. These modules follow the Rapid Assessment and Long-term Ecological Research (RAPELD) system (Magnusson et al., 2013). Each module is 1 km in width and 5 km in length. Within each module, there are 10 evenly spaced 250-m-long plots that follow terrain contours (Magnusson et al., 2013). The minimum distance between plots in the same module is 1 km. In the north of the study area, inside the Alter do Chão APA, we also surveyed 24 plots, 19 located in forest fragments surrounded by a savanna matrix and five in continuous forest (Figure 1). Although these plots were 250 meters long, they did not follow the terrain contour. However, since the area was relatively flat, this deviation resulted in minimal differences in variability within each plot.

We used data collected by a portable Light Detection and Ranging (LIDAR) device (Model LD90-3100VHS-FLP, Riegl, Horn, Austria), which employs laser scanning to measure variables that quantify canopy openness (our local-clearing index) and average vegetation height in the plots (see Torralvo et al., 2020; Torralvo et al., 2021). We estimated canopy openness by calculating the percentage of sky shots, which represents the number of emitted pulses that do not return to the apparatus.

Measures of unforested areas, the proportion of open areas in the landscape, were obtained in a previous study (Torralvo et al., 2020; Torralvo et al., 2021) by using a satellite-based layer with 30 m x 30 m pixels made available through the MapBiomas project (Souza et al., 2020, see full description in http://mapbiomas.org), which summarized the data on area without forest until 2019. Using the raster R-package (Hijmans, 2020), we extracted the proportions of areas without forest within 500-meter buffers around each plot. In our study region, unforested areas are predominantly found in the northern part.

Soil collection and analyses were carried out following the PPBio protocol (available at https://ppbio.inpa.gov.br/manuais). We used data on the quantity (g/kg) of clay and sand (sum of fine and coarse sand) to 5 cm from the surface. Leaf-litter height was measured with a ruler at six points separated by 50-meter intervals in each plot and the mean values per plot were used in the analyses. We measured the distances from each plot to the nearest water body using the Euclidean distance matrix tools of QGIS 3.16.2 software (QGIS Development Team, 2020), applied to combined hydrographic shapefiles from public repositories (IBGE, 2021)