You are here

Landscape applications of machine learning: Comparing random forests and logistic regression in multi-scale optimized predictive modeling of American marten occurrence in northern Idaho, USA [Chapter 9]

Posted date: December 04, 2018
Publication Year: 
2018
Authors: Cushman, Samuel A.; Wasserman, Tzeidle N.
Publication Series: 
Book Chapter
Source: In: Humphries, G.; Magness, D.; Huettmann, F., eds. Machine Learning for Ecology and Sustainable Natural Resource Management. Springer, Cham. p. 185-203

Abstract

The American marten (martes americana) is a species that is dependent on old conifer forest at middle to high elevations and is highly sensitive to habitat loss and fragmentation in a scale dependent fashion (e.g., Hargis et al. 1999; Wasserman et al. 2012a, b), and forest management is often influenced by considerations of how management will affect extent and pattern of marten habitat. Due to their dependence on extensive, unfragmented forest landscapes and microhabitat structures associated with late successional forest (Buskirk and Ruggiero 1994; Hargis et al. 1999), American marten are sensitive to fragmentation of late seral forest habitats, such as that resulting from timber harvest and associated extraction routes and road building (e.g., Cushman et al. 2011). Previous studies have consistently shown that American marten habitat requirements include forests with high canopy cover (Hargis and McCullough 1984; Wynne and Sherburne 1984), abundant near ground structure (Chapin et al. 1998; Godbout and Ouellet 2008), high prey densities (Fuller and Harrison 2005), and sufficient snow depth to provide subnivean spaces during winter (Wilbert et al. 2000). These habitats are thought to provide opportunities for foraging, resting, denning, thermoregulation, and avoiding predation. Perturbations, such as timber harvest, remove canopy cover, reduce coarse woody debris, change mesic sites into xeric sites, remove riparian dispersal zones, and change prey communities (Buskirk and Ruggiero 1994). American marten avoid areas with even relatively low levels of forest fragmentation and rarely use sites where more than 25% of forest cover has been removed (Hargis et al. 1999). Highly contrasting edge habitats, such as borders between late successional forest and harvested patches, and areas of open canopy are strongly avoided (Buskirk and Ruggiero 1994; Hargis et al. 1999; Cushman et al. 2011).

Citation

Cushman, Samuel A.; Wasserman, Tzeidle N. 2018. Landscape applications of machine learning: Comparing random forests and logistic regression in multi-scale optimized predictive modeling of American marten occurrence in northern Idaho, USA [Chapter 9]. In: Humphries, G.; Magness, D.; Huettmann, F., eds. Machine Learning for Ecology and Sustainable Natural Resource Management. Springer, Cham. p. 185-203.