Performance of habitat suitability models for the endangered black-capped vireo built with remotely-sensed data

详细信息	查看全文 | 推荐本文 |

• 作者：Chad B. Wilsey^a ; ^{cbwilsey@uw.edu} ; Joshua J. Lawler^a ; ^{jlawler@uw.edu} ; David A. Cimprich^b ; ^{david.a.cimprich2.civ@mail.mil}
• 关键词：Vegetation ; Aerial imagery ; Aerial altimetry ; LiDAR ; cforest ; Variable importance ; Black-capped vireo
• 刊名：Remote Sensing of Environment
• 出版年：2012
• 期刊代码：79_00344257
• 类别：ear
• 出版时间：16 April, 2012
• 卷：119
• 期：Complete
• 页码：35-42
• 文件大小：465 K

摘要

Identifying suitable habitat is critical to endangered species management and recovery. However, this basic task is often complicated by the rarity of the species in question and the limited availability of high quality environmental data. The endangered Black-capped Vireo (Vireo atricapilla) occupies early seral shrub communities generated by fire. The ability to predict vegetation structure is particularly important for mapping vireo habitat because this species occupies transitional, non-equilibrium vegetation types. We use presence data for territorial male vireos collected throughout the Fort Hood Military Reservation, Texas, to construct habitat suitability models using vegetation type (mapped from aerial imagery), soil data, and laser altimetry (LiDAR)-derived measures of vegetation structure. LiDAR produces a three-dimensional, high-resolution representation of vegetation structure across broad spatial scales. We built models that incorporated LIDAR outputs as well as the other habitat predictors using a non-parametric machine-learning algorithm, cforest. Models built using a single predictor class (vegetation structure or type or soil) performed similarly across 25 bootstrapped training and test datasets (median accuracies 76%, 74%, and 79%, respectively). Models incorporating two predictor classes performed better (80-81%) and only slightly worse than the full model (82%). Furthermore, vegetation type and soil data were more important predictors of habitat suitability than structural measures in the full model. Predictive maps suggest that models incorporating vegetation type and soil would be most useful for habitat conservation and management applications on Fort Hood. The addition of LiDAR-derived variables to the model further distinguishes suitable habitats from potentially suitable but presently overgrown areas. In the absence of detailed vegetation data, models based on structural measures performed well when combined with soil data. This could be useful in a broader regional context in which vireos occupy a greater variety of vegetation types with a common structure.