Alternative approaches to assessing the natural regeneration of Scots pine in a Mediterranean forest
Key message In modelling regeneration patterns, parametric regression is recommended because it can account for the spatial and temporal correlation present in the data, whereas decision trees allow more complex interactions and can be used to reduce the number of variables. Context The establishment of seedlings after regeneration fellings is key to guaranteeing the development and persistence of the forest. Depending on the objective pursued, data available or type of forest, a number of different methods have been employed to assess the relationship between seedling establishment and both environmental and stand factors. Most authors have conducted their analyses using parametric regression or point pattern analysis. Aim We analysed the way in which light, stand conditions, edaphic and topographic variables affect the regeneration of Pinus sylvestris L. in Central Spain. We used different methods to analyse the same data set. The strengths and weaknesses of each method were discussed. Methods We used two parametric approaches generalized linear mixed model regression using a negative binomial followed by the variant explanatory variables reduction prior to regression as well as three nonparametric approaches not commonly employed in forest regeneration nonmetric multidimensional scaling, regression trees and random forests algorithm. Results The parametric regression identified a larger number of variables associated with the regeneration process and the inclusion of a random effect in the model allowing the consideration of the spatial variability among plots. However, decision trees captured the complex interaction among variables, which typical parametric methods were unable to detect. Conclusion: Different statistical methods gave similar insights into the underlying ecological process. However, different statistical premises with inference implications can be noticed. This may give misinterpretation of the model depending on the nature of the data. The choice of a given method should be made according to the nature of the data and the achievement of desirable results. © 2015, INRA and Springer-Verlag France.
| Main Authors: | , , , , |
|---|---|
| Format: | journal article biblioteca |
| Language: | English |
| Published: |
BioMed Central
2015
|
| Subjects: | CHAID, Random forests, PCA, NMDS, Negative binomial, Generalized linear mixed models, |
| Online Access: | http://hdl.handle.net/20.500.12792/3017 http://hdl.handle.net/10261/293891 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Key message In modelling regeneration patterns, parametric regression is recommended because it can account for the spatial and temporal correlation present in the data, whereas decision trees allow more complex interactions and can be used to reduce the number of variables. Context The establishment of seedlings after regeneration fellings is key to guaranteeing the development and persistence of the forest. Depending on the objective pursued, data available or type of forest, a number of different methods have been employed to assess the relationship between seedling establishment and both environmental and stand factors. Most authors have conducted their analyses using parametric regression or point pattern analysis. Aim We analysed the way in which light, stand conditions, edaphic and topographic variables affect the regeneration of Pinus sylvestris L. in Central Spain. We used different methods to analyse the same data set. The strengths and weaknesses of each method were discussed. Methods We used two parametric approaches generalized linear mixed model regression using a negative binomial followed by the variant explanatory variables reduction prior to regression as well as three nonparametric approaches not commonly employed in forest regeneration nonmetric multidimensional scaling, regression trees and random forests algorithm. Results The parametric regression identified a larger number of variables associated with the regeneration process and the inclusion of a random effect in the model allowing the consideration of the spatial variability among plots. However, decision trees captured the complex interaction among variables, which typical parametric methods were unable to detect. Conclusion: Different statistical methods gave similar insights into the underlying ecological process. However, different statistical premises with inference implications can be noticed. This may give misinterpretation of the model depending on the nature of the data. The choice of a given method should be made according to the nature of the data and the achievement of desirable results. © 2015, INRA and Springer-Verlag France. |
|---|