An artificial neural network approach to spatial habitat modelling with interspecific interaction

被引:205
作者
Özesmi, SL
Özesmi, U
机构
[1] Univ Minnesota, Conservat Biol & MacArthur Interdisciplinary Prog, Minneapolis, MN 55455 USA
[2] Univ Minnesota, Dept Forest Resources, St Paul, MN 55108 USA
关键词
red-winged blackbird; Agelaius phoeniceus; marsh wren; Cistothorus palustris; artificial neural networks; neural interpretation diagram; logistic regression; spatial model; habitat selection; interspecific interaction; freshwater marsh; wetlands;
D O I
10.1016/S0304-3800(98)00149-5
中图分类号
Q14 [生态学(生物生态学)];
学科分类号
071012 ; 0713 ;
摘要
Spatial models for habitat selection of marsh-breeding bird species were developed using artificial neural networks. The habitat models included single and multiple species artificial neural networks for red-winged blackbird (Agelaius phoeniceus) and marsh wren (Cistothorus palustris). Data for the study came from two diked wetland basins on southwestern Lake Erie, USA. The single species artificial neural network model performed better than the logistic regression model except in the presence of interspecific interaction with marsh wren. A multiple species habitat model was developed using data from one basin that included both nesting red-winged blackbirds and marsh wrens. The multiple species neural network model performed better than the logistic regression model in the presence of interspecific interaction and could simultaneously predict the nest locations of both species. Using neural interpretation diagrams, relevances, and sensitivity analyses, we determined the mechanisms of habitat selection in red-winged blackbirds and marsh wrens. Habitat selection in marsh-breeding bird species was a non-linear process that could not be sufficiently understood in terms of general linear models. The non-linear structure of the neural network model not only predicts habitat selection better, but with a critical evaluation through neural interpretation diagrams, relevances and sensitivity analyses can lead to a better understanding of the mechanisms of habitat selection. As predictive tools single species neural network models were too specific in cases where the system changed to include interacting species. In these cases linear models might be better predictors but not necessarily provide a better understanding. In the presence of two interacting species we suggest the use of a neural network model that is trained with a data set from a wetland where all the interacting species are present. In the absence of the interacting species the dynamics will change and habitat selection can be based on an entirely different set of rules and relationships. Therefore, the decision on which neural network model to use must be based on an in-depth understanding of the ecology of the system under study. (C) 1999 Elsevier Science B.V. All rights reserved.
引用
收藏
页码:15 / 31
页数:17
相关论文
共 55 条
  • [1] ALBERS PH, 1978, WILSON BULL, V90, P619
  • [2] Anderson J. A, 1995, INTRO NEURAL NETWORK, P650
  • [3] SPATIAL MODELING OF THE BARN-OWL TYTO-ALBA HABITAT USING LANDSCAPE CHARACTERISTICS DERIVED FROM SPOT DATA
    ANDRIES, AM
    GULINCK, H
    HERREMANS, M
    [J]. ECOGRAPHY, 1994, 17 (03) : 278 - 287
  • [4] [Anonymous], IEEE ASSP MAGAZINE
  • [5] SPATIAL-ANALYSIS OF SANDHILL CRANE NESTING HABITAT
    BAKER, BW
    CADE, BS
    MANGUS, WL
    MCMILLEN, JL
    [J]. JOURNAL OF WILDLIFE MANAGEMENT, 1995, 59 (04) : 752 - 758
  • [6] Stochastic models that predict trout population density or biomass on a mesohabitat scale
    Baran, P
    Lek, S
    Delacoste, M
    Belaud, A
    [J]. HYDROBIOLOGIA, 1996, 337 (1-3) : 1 - 9
  • [7] BERNSTEIN NP, 1980, OHIO J SCI, V80, P15
  • [8] Bishop C.M., 1995, NEURAL NETWORKS PATT, P482, DOI DOI 10.2307/2965437
  • [9] Block William M., 1993, Current Ornithology, V11, P35
  • [10] Revealing dynamics of ecological systems from natural recordings
    Boudjema, G
    Chau, NP
    [J]. ECOLOGICAL MODELLING, 1996, 91 (1-3) : 15 - 23