Can LiDAR data improve bird habitat suitability models?

Habitat suitability models are based on digital maps that very often describe the environment at a human scale and, hence miss ecological structures and features that are important for wildlife. LiDAR (Light Detection And Ranging) data, laser scanning acquired by remote sensing, can fill this gap by providing useful information not only on the spatial extent of habitat types but also information on the vertical height. The advantage of LiDAR derived variables lays also in the availability at a large scale, instead of just in the survey sites. In this work we evaluated the effect of three LiDAR derived variables (tree height, percentage of trees in open areas and length of ecotone) on the performance of habitat models, developed for four farmland bird species. For each species multiple runs of stepwise Logistic Regression (LR) and Maximum Entropy Models (Maxent) were performed. For each run we included and excluded the LiDAR variables and recorded the improvement in model performance using the AUC, AIC, Sensitivity, Specificity. Model results were applied in a GIS in order to create habitat suitability maps. Results for the RL models showed that for most of the species at least one LiDAR variable was selected and significant (p<0.05). Additionally the inclusion LIDAR data gave a positive percentage of contribution to the AUC of the Maxent models. The models calculated using LiDAR derived variables identified a smaller area on the map, with a better overlap with open areas, thus showing a more realistic spatial pattern. The interpretation of these variable is also more straightforward, both from the ecological point of view and when defining management guidelines.