Análise estatística espacial na avaliação de produtividade no melhoramento genético do feijoeiro
Arquivos
Data
2009-02-18
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal de Viçosa
Resumo
A dependência espacial é a propensão pela qual o valor de uma variável observada em certa posição tende a assemelhar-se mais aos valores vizinhos do que ao restante das observações do conjunto amostral. O objetivo deste trabalho foi a avaliação da eficiência de alguns métodos de análise estatística espacial para melhorar a precisão experimental em ensaios de seleção de famílias em programa de melhoramento genético do feijoeiro. Foram utilizados dados para produtividade de oito experimentos montados em látice, conduzidos na estação experimental de Coimbra, pertencente ao Departamento de Fitotecnia da Universidade Federal de Viçosa (UFV). Para cada análise, foram analisados os métodos de análise de vizinhança: método das médias móveis, método de Papadakis, reaplicação das médias móveis, reaplicação do método Papadakis e o método com erros dependentes em comparação com a análise tradicional em látice, que desconsidera a dependência espacial entre parcelas quanto ao ajuste dos modelos e à classificação das famílias avaliadas. Foi verificada a existência da dependência espacial nos experimentos pelo teste de Durbin-Watson, semivariograma empírico e teórico com o ajustamento de um modelo geoestatístico para resíduos, verificando a coerência dos dados e percebendo a ocorrência da dependência espacial em seis dos oito experimentos avaliados (capítulo 1). Dos seis experimentos, cinco apresentaram dependência espacial para análise em blocos e um para análise em látice e em blocos, com alcance variando de 2,6 a 33,3 m. No capítulo 2, foram utilizados o teste de Durbin-Watson para verificar a dependência espacial entre resíduos para as diferentes metodologias, além do critério de informação de Akaike (AIC) e do teste da razão da verossimilhança (LRT) para comparar o ajuste dos modelos. Utilizaram-se o coeficiente de variação e a eficiência relativa para comparar a eficiência das metodologias testadas em relação à análise em látice. Também foram utilizados o coeficiente de correlação de Spearman e a eficiência de seleção para comparar as classificações de famílias do feijoeiro obtidas na análise em látice e pelos métodos de análise espacial. Verificou-se que os métodos de análise espacial ajustaram-se melhor aos dados em relação aos métodos que assumem erros independentes. Quanto à precisão experimental, a eficiência da análise em látice foi maior em relação à análise em blocos quando há maior dependência espacial. Os métodos de análise espacial, médias móveis, Papadakis e o método com erros dependentes apresentaram razoável eficiência, enquanto as metodologias propostas, reaplicações das médias móveis e Papadakis foram altamente eficientes em relação à análise em látice, sendo mais eficiente o método reaplicação de Papadakis cuja vizinhança é composta por uma parcela superior e uma inferior em relação à parcela referência. Os métodos testados apresentaram razoável nível de concordância em relação às famílias selecionadas na análise em látice. As metodologias de reaplicação de médias móveis e reaplicação de Papadakis mostram-se altamente eficientes para melhorar a precisão experimental, além de conseguir garantir a independência entre resíduos, sendo uma excelente alternativa de análise em relação à análise em látice.
Spatial dependence is a tendency that a variable´s value, observed in certain position, leads to approximate to its neighbors value than the value of the rest of the elements observed in the sample. The aim of this work was to evaluate the efficiency of some methods of geostatistic analyses to improve the experimental accuracy of tests that selects families for genetic improvement program of beans plantation. Date set were used to evaluate the productivity of eight experiments set in lattice design, located in Coimbra s Experimental Station controlled by the Plant Science Department at Federal University of Viçosa UFV. For each study was analyzed the neighborhood methods: Moving Means Method, Papadakis Method, Moving Means Replication, Papadakis Method Replication and the method that contains errors dependent in the comparison between traditional analyses in lattice which do not consider the spatial dependence among plots according to its fitting model, and the family classification in study. The existence of experiment s spatial dependence was observed through the Durbin-Watson s test, empirical and theoretical semivariogram with a geostatistic fitting model for residues checking the coherence of data and noticing the occurrence of spatial dependence in 6 out of 8 experiments in study (Chapter1). Five out of the six experiments showed spatial dependence for analyses in blocks and one for analyses in lattice and blocks, reaching 2.6 to 33.3m. In chapter 2, it was used the Durbin-Watson s test to observe the spatial dependence among residues for different methodologies, besides the Akaike Information Criteria (AIC) and the Maximum Likelihood Method to compare the fitting model. The variation coefficient and the relative efficiency were used to compare the efficiency of tested methodologies related to the analyses in lattice. The Spearman s Correlation Coefficient and the efficiency of selection were also used to compare the classification of beans plantation families acquired through the analyses in lattice and through the spatial analyses method. It was noticed that the spatial analyses method had a better fitting to the data than the methods that assume independent errors. The experiments accuracy, the analyses in lattice efficiency was higher compared to the analyses in blocks when there is higher spatial dependence. The spatial analyses method, Moving Means Method, Papadakis Method and the method with independent errors showed reasonable efficiency, while the proposed methodologies, Moving Means Replication and Papadakis were highly efficient than the analyses in lattice. Papadakis replication Method showed to be more efficient once its neighborhood is constituted by a superior and inferior plot compared to the reference plot. The tested methods showed reasonable concordance level compared to the selected families for analyses in lattice. The Moving Means Replication and Papadakis Replication methodologies showed highly efficient to improve the experiments accuracy, besides to guarantee the independence among residues, being an excellent alternative for analyses compared to the analyses in lattice.
Spatial dependence is a tendency that a variable´s value, observed in certain position, leads to approximate to its neighbors value than the value of the rest of the elements observed in the sample. The aim of this work was to evaluate the efficiency of some methods of geostatistic analyses to improve the experimental accuracy of tests that selects families for genetic improvement program of beans plantation. Date set were used to evaluate the productivity of eight experiments set in lattice design, located in Coimbra s Experimental Station controlled by the Plant Science Department at Federal University of Viçosa UFV. For each study was analyzed the neighborhood methods: Moving Means Method, Papadakis Method, Moving Means Replication, Papadakis Method Replication and the method that contains errors dependent in the comparison between traditional analyses in lattice which do not consider the spatial dependence among plots according to its fitting model, and the family classification in study. The existence of experiment s spatial dependence was observed through the Durbin-Watson s test, empirical and theoretical semivariogram with a geostatistic fitting model for residues checking the coherence of data and noticing the occurrence of spatial dependence in 6 out of 8 experiments in study (Chapter1). Five out of the six experiments showed spatial dependence for analyses in blocks and one for analyses in lattice and blocks, reaching 2.6 to 33.3m. In chapter 2, it was used the Durbin-Watson s test to observe the spatial dependence among residues for different methodologies, besides the Akaike Information Criteria (AIC) and the Maximum Likelihood Method to compare the fitting model. The variation coefficient and the relative efficiency were used to compare the efficiency of tested methodologies related to the analyses in lattice. The Spearman s Correlation Coefficient and the efficiency of selection were also used to compare the classification of beans plantation families acquired through the analyses in lattice and through the spatial analyses method. It was noticed that the spatial analyses method had a better fitting to the data than the methods that assume independent errors. The experiments accuracy, the analyses in lattice efficiency was higher compared to the analyses in blocks when there is higher spatial dependence. The spatial analyses method, Moving Means Method, Papadakis Method and the method with independent errors showed reasonable efficiency, while the proposed methodologies, Moving Means Replication and Papadakis were highly efficient than the analyses in lattice. Papadakis replication Method showed to be more efficient once its neighborhood is constituted by a superior and inferior plot compared to the reference plot. The tested methods showed reasonable concordance level compared to the selected families for analyses in lattice. The Moving Means Replication and Papadakis Replication methodologies showed highly efficient to improve the experiments accuracy, besides to guarantee the independence among residues, being an excellent alternative for analyses compared to the analyses in lattice.
Descrição
Palavras-chave
Dependência espacial, Médias móveis, Papadakis, Seleção, Spatial dependence, Moving means, Papadakis method, Selection
Citação
FERES, Andréia Luiza Gonzaga. Geostatistic analyses in the assessment the beans families productivity. 2009. 89 f. Dissertação (Mestrado em Estatística Aplicada e Biometria) - Universidade Federal de Viçosa, Viçosa, 2009.