Skip to Main Content
Machine Learning for Macroscale Ecological Niche Modeling - a Multi-Model, Multi-Response Ensemble Technique for Tree Species Management Under Climate ChangeAuthor(s): Anantha M. Prasad
Source: In: Humphries, Grant; Magness, Dawn R.; Huettmann, Falk, eds. Machine Learning for Ecology and Sustainable Natural Resource Management. Switzerland: Springer Nature: 123-139.
Publication Series: Book Chapter
Station: Northern Research Station
Download Publication (773.0 KB)
DescriptionMachine learning has come a long way in recent decades due to huge increases in computing power and the availability of robust public platforms for statistical analysis (e.g., R Core Team 2016). Machine learning techniques have benefited from advances in statistical learning and vice versa (Hastie et al. 2009; Slavakis et al. 2014), resulting in impressive applications of big data in imaging, astronomy, medicine, finance and to a lesser extent in ecology (Van Horn and Toga 2014; Zhang and Zhao 2015; Belle et al. 2015; Hussain and Prieto 2016; Hampton et al. 2013). A healthy relationship with computer science and engineering has invigorated the field even more, resulting in a variety of techniques suitable for diverse applications. One successful and frequently used method is ensemble learning, where learning algorithms independently construct a set of classifiers or regression-estimates and classify or regress newer data points by either taking a weighted vote (classifiers) or an average (regression) of their predictions (Zhou 2012). A majority of the ensemble learning problems deal with classification due to the binary, or in some cases multinomial, response that is of interest. However, in the field of ecology, and especially in tree species abundance modeling, we have access to continuous data thanks to the Forest Inventory Analysis (FIA) in the United States (Woudenberg et al. 2010) that lends itself to a regression approach. Valuable information can be lost if the continuous data are classified a priori into classes. Therefore, it is best to solve the problem in a regression context, and classify the results later to retain most of the information in the response. I will choose the regression approach for this reason and also to highlight this less used aspect of statistical learning.
- Check the Northern Research Station web site to request a printed copy of this publication.
- Our on-line publications are scanned and captured using Adobe Acrobat.
- During the capture process some typographical errors may occur.
- Please contact Sharon Hobrla, firstname.lastname@example.org if you notice any errors which make this publication unusable.
- We recommend that you also print this page and attach it to the printout of the article, to retain the full citation information.
- This article was written and prepared by U.S. Government employees on official time, and is therefore in the public domain.
CitationPrasad, Anantha M. 2018. Machine Learning for Macroscale Ecological Niche Modeling - a Multi-Model, Multi-Response Ensemble Technique for Tree Species Management Under Climate Change. In: Humphries, Grant; Magness, Dawn R.; Huettmann, Falk, eds. Machine Learning for Ecology and Sustainable Natural Resource Management. Switzerland: Springer Nature: 123-139. Chapter 6. https://doi.org/10.1007/978-3-319-96978-7_6.
- Approximating prediction uncertainty for random forest regression models
- Current state of the art for statistical modeling of species distributions [Chapter 16]
- Instance annotation for multi-instance multi-label learning
XML: View XML