Landslides are common natural disturbances in tropical montane forests. While the geomorphic drivers of landslides in the Andes have been studied, factors controlling post-landslide forest recovery across the steep climatic and topographic gradients characteristic of tropical mountains are poorly understood. Here we use a LiDAR-derived canopy height map coupled with a 25-year landslide time series map to examine how landslide, topographic, and biophysical factors, along with residual vegetation, affect canopy height and heterogeneity in regenerating landslides. We also calculate aboveground biomass accumulation rates and estimate the time for landslides to recover to mature forest biomass levels. We find that age and elevation are the biggest determinants of forest recovery, and that the jump-start in regeneration that residual vegetation provides lasts for at least 18 years. Our estimates of time to biomass recovery (31.6-37.1 years) are surprisingly rapid, and as a result we recommend that future research pair LiDAR with hyperspectral imagery to estimate forest aboveground biomass in frequently disturbed landscapes. Synthesis: Using a high-resolution LiDAR dataset and a time-series inventory of 608 landslides distributed across a wide elevational gradient in Andean montane forest, we show that age and elevation are the most influential predictors of forest canopy height and canopy variability. Other features of landslides, in particular the presence of residual vegetation, shape post-landslide regeneration trajectories. LiDAR allows for a detailed analysis of forest structural recovery across large landscapes and numbers of disturbances, and provides a reasonable upper bound on aboveground biomass accumulation rates. However, because this method does not capture the effect of compositional change through succession on aboveground biomass, wherein high-wood density species gradually replace light-wooded pioneer species, it overestimates aboveground biomass. Given previously estimated stem turnover rates along this elevational gradient, we posit that aboveground biomass recovery takes at least three times as long as our recovery time estimates based on LiDAR-derived structure alone.

The bulk of the dataset was collected by overlaying landslide polygons onto various raster datafiles, then extracting the mean values of each parameter for each landslide. A portion of the dataset was generated by a support vector machine model. Methods for how each variable was collected are detailed in the manuscript and briefly in the "Information" tab of the Excel file. Where not noted, data were processed using various spatial data packages in R statistical software.

There is additional information about each tab and each variable in the dataset on the "Information" tab of the Excel file.