Lidar and Multispectral Data for Assessing Texas and Oklahoma Songbird Habitat and Density

Submitted by atripp on

National Wildlife Refuges (NWR) in Texas and Oklahoma manage forested habitats to support priority bird populations in the West Gulf Coastal Plain and Ouachitas Bird Conservation Regions. Airborne laser altimetry or light detection and ranging (lidar) can capture details of forest structure that determine bird species diversity, densities, and distributions. Point cloud information from leaf-on multi-return lidar generates a variety of forest metrics such as tree canopy cover, density, and height, as well as other less conventional parameters representing vertical and horizontal habitat structure, such as height rumple and surface to volume ratio. PlanetScope multispectral imagery with channels in the blue (455–515 nanometers), green (500–590 nanometers), red (590–670 nanometers), and near-infrared (780–860 nanometers) wavelengths provides a high spatial resolution (3 meter) and frequency acquired (daily) data source that can add plant phenology and composition information when fused with lidar. Forest management relies on continuous forest inventory (CFI) plots and songbird point counts to monitor habitat conditions and species abundance. Synthesis methods in development will pool data resources for estimating forest inventory parameters, habitat conditions, bird detection bias, availability, and density across each of five refuges.

Combined and consistent field surveys and remotely sensed data will provide robust models of species habitat relationships and value-added data layers useful for making land management decisions. Preliminarily, recursive feature elimination (RFE) and ‘tree bag’ functions reduced over 300 lidar grid and strata metrics to less than 10 non-correlated predictor variables for estimating forest inventory parameters across Little River and Caddo Lake NWR. Lidar and PlanetScope data summarized for 1/5th acre CFI plots (n = 50) and regression tree models showed good performance for estimating basal area (R2 = 0.59, RMSE = 29.6) and trees per acre (R2 = 0.67, RMSE = 22.0). Future work will use these and other metrics to estimate songbird habitat relationships and density. Project outputs will range from standardized data collection protocols to mapped forest inventory parameters, bird density estimates, and information on habitat preferences. Remote sensing applications and songbird models will help inform forest management actions most likely to maintain varied habitat conditions and songbird populations.

Preliminary maps of a) Caddo Lake National Wildlife Refuge (CLNWR) in eastern Texas showing Recursive Feature Elimination (RFE) and regression tree model results that combined continuous forest inventory plots and light detection and ranging (lidar) and PlanetScope predictors to estimate b) tree basal area and c) trees per acre.


Author Name
Steven Sesnie; Jim Mueller; Paige Schmidt; Randy Stewart
Author Email;;;