Premium
Periglacial distribution modelling with a boosting method
Author(s) -
Hjort Jan,
Marmion Mathieu
Publication year - 2008
Publication title -
permafrost and periglacial processes
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.867
H-Index - 76
eISSN - 1099-1530
pISSN - 1045-6740
DOI - 10.1002/ppp.629
Subject(s) - boosting (machine learning) , solifluction , parametric statistics , robustness (evolution) , permafrost , artificial neural network , machine learning , gradient boosting , artificial intelligence , computer science , geology , geomorphology , mathematics , statistics , random forest , biochemistry , chemistry , oceanography , glacial period , gene
We assessed the applicability of a boosting method in periglacial distribution modelling using empirically derived data on cryoturbation, sporadic permafrost and sorted solifluction from an area of 600 km 2 in sub‐Arctic Finland. The main aims were: (1) to compare the predictive ability of the generalised boosting method used with more common parametric techniques (generalised linear model) and machine‐learning methods (artificial neural networks) and (2) to assess the tenability of the explanatory variables highlighted by the generalised boosting method. The results showed the robustness of the boosting method in predicting the distribution of periglacial phenomena in the sub‐Arctic landscape. Furthermore, the environmental factors selected by the boosting method coincided well with the expected controls of the phenomena. The strengths of the generalised boosting method lie in its high predictive ability, flexibility in capturing complex process‐environment relationships and realistic model outcomes. Copyright © 2008 John Wiley & Sons, Ltd.