Análise de coordenadas principais de matrizes de vizinhança (PCNM) e o efeito da distância de truncamento nos padrões espaciais gerados

• Ano de defesa: 2023
• Autor(a) principal: Leão Neto, Wilson Mamedes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Mato Grosso; Brasil; Instituto de Biociências (IB); UFMT CUC - Cuiabá; Programa de Pós-Graduação em Ecologia e Conservação da Biodiversidade
• Palavras-chave em Português: PCNM; Distância de truncamento; Ecologia e conservação; Correlação espacial; Partição de variância; CNPQ::CIENCIAS BIOLOGICAS::ECOLOGIA; Truncation value; Ecology and conservation; Spatial correlation; Partitioning variation
• Link de acesso: http://ri.ufmt.br/handle/1/5930

The ecologists Pierre Legendre and Daniel Borcard have proposed in their paper of 2002 a method, the Principal Coordinate of Neighbour Matrices (PCNM), to detect and quantify the spatial correlation patterns on several spatial scales. The PCNM works by using a distance matrix in which a predefined distance value (truncation value/threshold value) is used to determine the type of connection between spatial areas (the truncation of the distance matrix). In the original model it’s used the Minimum Spanning Tree type of connection as truncation value. Although the Minimum Spanning Tree is the default truncation value in R program and is widely used in ecology analysis for the application of PCNM, it is possible to use other types of connections or values as threshold for PCNM that are better adapted for the data. As such, this paper tries to test the hypothesis that different truncation values have a great influence in the results of the spatial filters produced by PCNM analysis and that likely the arbitrary choose of a truncation value can lead researchers to different conclusions. To test the hypothesis, we utilized five different geographic sampling patterns data, with biological data as associated to them. We applied PCNMs with several different truncation values to all five geographic sampling data types to analyze their response to the biological data in a Redundancy Analysis (RDA) and the filters generated by the PCNMs. Our results show that PCNMs is highly affected by the choose of the truncation value, particularly the number of fine scale filters that the PCNMs are able to produce. This has a high effect over the power of explanation when they are applied to analyze the spatial pattern of the biological data in RDA. For some species analyzed, such as Trichoptera and Oribatida, the PCNMs varied from non-explanation to an explanation of almost 100%, of the data distribution. Also, the truncation value that grants the analyse of biological distribution in RDA with PCNMs a high explanation can vary a lot between specie in the sampling data used. So, the usage of a predefined truncation value, such as the Minimum Spanning Tree, may underestimate the spatial autocorrelation in data. The change of the truncation values for PCNMs lead na increase of the adjusted R², with significant P value (<0.05), for many of the species analyzed, including the ones in which the PCNM with the threshold produced by the Minimum Spanning Tree could find no explanation. We conclude that the truncation value has great influence over the number of filters produced by PCNMs and the power of explanation of those filters. In order to make a good application of this method it’s necessary a better understanding of the kind of sampling area and its spatial distribution, as well a broad understanding of the PCNM components. We hope that our study helps other researchers to have a better understanding of this method.