Lidar-Based Models of Understory Bird Habitat in a Tropical Forest

Abstract

Where a poorly understood group of wildlife species seems to be declining quickly, a rapid assessment of the species’ habitat requirements may be needed in order to make the most optimal management decisions possible. We used bird presence data and a combination of field data and full-waveform lidar data to predict and interpret the distributions of declining insectivorous understory bird species at the La Selva Biological Station in Costa Rica. Raw lidar waveforms were used to create metrics of multi-dimensional forest structure which take into account not only horizontal structure such as patches and their arrangement or fragmentation, but also the vertical structure of vegetation such as canopy height and the distribution of canopy layers. Habitat models for four species of understory insectivore were developed using MaxEnt and validated using a jackknife approach, while guild diversity was estimated across the landscape using multiple logistic regression. Habitat projections for individual species showed high and significant predictive ability in jackknife tests. Results of habitat modeling showed significant differences between species in terms of which habitat variables were most important, but percent cover, distance to forest edge, foliage height diversity, and canopy height were consistently important. Metrics derived from canopy height profiles were consistently more useful predictors than metrics from the raw lidar waveforms. General metrics such as canopy height, elevation, and distance to edge were generally more useful predictors than understory-specific metrics, which could indicate that understory insectivores respond more strongly to climate & habitat patch size than to understory structure at a micro level. Alternatively, large-footprint lidar may be unable to adequately represent the aspects of understory structure which impact understory birds. Overall, however, models which included canopy height profile metrics significantly improved upon models which did not, indicating that inclusion of measures of multi-dimensional forest structure which account for the understory may add value to lidar-based habitat models for many wildlife species.