Conservation biologists and natural resource managers often require detailed, accurate information on natural resources or biodiversity elements such as species, landscapes, and ecosystems. Their patterns of occurrence and their responses to environmental disturbance or change are dynamic over space and time and may be mediated by complex ecological processes. In most cases, our ability to directly measure or comprehensively map biodiversity elements is limited by human or financial resources, and logistical challenges such as difficulties in accessing terrain or short field seasons. In other situations, we might want to make quantitative inferences about, say, the kinds of environments that are most suitable for the persistence of an endangered species, or the influence of landscape modification on its highest-quality habitat. In these cases, developing models that explain and predict the patterns of biodiversity elements can help provide guidance at scales and resolutions that are not available through direct measurement. For example, Goetz et al. (2007) employed lidar data to predict the bird species richness across a 5,315 ha temperate forest reserve, the Patuxent National Wildlife Refuge (PWNR) in the eastern United States. In this study, Goetz et al. derived and mapped several measures of forest canopy structure, including canopy height, and three descriptors of the vertical distribution of canopy elements. In addition to lidar, they also used optical remotely sensed data from two dates of Landsat ETM+ to derive NDVI during the growing season and the difference between the NDVI of leaf-on and leaf-off conditions (growing season versus winter). Testing three different quantitative statistical models (stepwise multiple linear regression, generalized additive models, and regression trees) to predict bird species richness, the authors used field survey data on the birds of the PWNR that were collected at a series of fixed points across the reserve as the training data for the response variable (bird species richness). To calibrate the model, they combined the habitat descriptors with the survey data, usually reserving 25 percent of the survey data to validate each model’s results.