Premium
Effects of variable attribute weights on landform classification
Author(s) -
Deng Y. X.,
Wilson J. P.,
Sheng J.
Publication year - 2006
Publication title -
earth surface processes and landforms
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.294
H-Index - 127
eISSN - 1096-9837
pISSN - 0197-9337
DOI - 10.1002/esp.1401
Subject(s) - landform , terrain , fuzzy logic , computer science , matching (statistics) , sensitivity (control systems) , variable (mathematics) , artificial intelligence , data mining , pattern recognition (psychology) , mathematics , geology , statistics , geography , cartography , geomorphology , engineering , mathematical analysis , electronic engineering
Abstract This paper focuses on the attribute weight issue and advocates use of modiable attribute weights in terrain‐based environmental analysis and classication. A question was asked: ‘How much will the result of a terrain‐based environmental analysis be affected if the weights of used terrain attributes are changed?’ The literature on landform classication and the fuzzy k ‐means method was reviewed in particular to help clarify the background and importance of this weight assignment issue. As an example, the effects of modifying attribute weights were evaluated for fuzzy k ‐means landform classication in a case study area. A total of 102 classications were compared with each other and with a soil map, and comparison methods were specically designed to evaluate the differences between these classications. The results show that fuzzy k ‐means landform classication is sensitive to weight adjustments of adopted terrain attributes. The sensitivity is particularly high when the attribute weights started to be tuned away from the standard (i.e. uniform) weight of one. Better matching between landform classication and a soil map may be produced when attribute weights are tuned. In all, we advocate the widespread adoption of an exploratory attitude in assigning attribute weights for environmental analysis and classication. Copyright © 2006 John Wiley & Sons, Ltd.