Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos
Autor(a) principal: | |
---|---|
Data de Publicação: | 2013 |
Tipo de documento: | Tese |
Idioma: | por |
Título da fonte: | LOCUS Repositório Institucional da UFV |
Texto Completo: | http://locus.ufv.br/handle/123456789/1649 |
Resumo: | In digital soil mapping the terrain attributes have been used as the main environmental predictor variables. Other variables related to pedogenic processes, such as climate changes, has not been used often. In this context, the objectives of this study were: (1) to evaluate the method of cokriging ordinary kriging compared to the spatial distribution of rainfall in the State of Espírito Santo, (2) identify soil mapping at different scales, which variables most relevant for better performance prediction of soil classes, whose study two distinct areas, (3) evaluate the performance of the NaiveBayes classifier algorithm, neural network multilayerperceptron - MLP SimpleCart and J48. In the evaluation of kriging and cokriging were used data from 108 rain gauges and secondary variables such as altitude and distance from the sea. For the evaluation of the variables and the mapping algorithms classifiers medium scale (1:100,000) the study was conducted in the river basin Muqui North, south of Espirito Santo. We used 598 training instances (10 soil classes) and 45variáveis related to different factors of soil formation, such as terrain attributes, geology, geomorphology, climate, water balance and indices derived from bands 1, 3, and 4 5 Landsat 5 TM. These variables were subjected to different methods of feature selection based on correlation - CFS, in consistency - CSE, information gain - and IA "ReliefF", available in the software Weka 3.6.8. This software was applied SimpleCart classifier to evaluate the effectiveness of the prediction with the subsets of selected variables. The evaluation of the four binders was performed using the 45 variables and the algorithm selected by "ReliefF." For detailed mapping conducted in Rural Settlement Sezínio Fernandes, Linhares, ES, we used 259 training instances (three soil classes) and 19 predictor variables (terrain attributes, climate and water balance) in the predictions made by the classifiers SimpleCart, J48, MLP and NaiveBayes. The predictions were evaluated based on cross- validation and comparisons of maps made with the conventional map reference. The interpolation results suggest that cokriging to be preferred to the use of regular grids for sampling secondary variables. The results of the selection of attributes for mapping river basin Muqui North indicated that the algorithms "ReliefF" and CSE, both limited xto 10 attributes were those with less complex trees and without significant loss in accuracy prediction compared to group 45 variables. The classification accuracy, indicated by Kappa of 0.60 was considered very good. The variables selected by "ReliefF" were geology, geomorphology and especially terrain attributes and elements of the water balance, as water surplus, water deficit and potential evapotranspiration. Algorithms NaiveBayes, MLP and SimpleCart showed similar performance prediction (Kappa 0.60 to 0.66), higher than the J48. The highest agreement with the reference map obtained by the MLP algorithm, followed by SimpleCart, J48 and NaiveBayes was 55, 52, 51 and 48%, respectively. The predictions of soil from Settlement Sezinio Fernandes variables water surplus, water deficit and air temperature were relevant. However, the small amplitude values presented by climatic variables and water balance are probably not sufficient to provide different pedogenetic conditions in the study area. There were no significant differences between the Kappa values (0.77 to 0.82) of the three classification algorithms. The greatest agreement with the conventional map was obtained for the algorithm J48, followed by NaiveBayes and SimpleCart,using only the terrain attributesas predictors variables. Decision trees for producing results more easily understood and presented in general accuracies similar to NaiveBayes classifiers and neural network MLP, may be regarded as of great potential to consolidate the digital soil mapping. |
id |
UFV_825b2b83bd81d08025ba195a51023f6d |
---|---|
oai_identifier_str |
oai:locus.ufv.br:123456789/1649 |
network_acronym_str |
UFV |
network_name_str |
LOCUS Repositório Institucional da UFV |
repository_id_str |
2145 |
spelling |
Cunha, Alexson de Mellohttp://lattes.cnpq.br/4853651139461402Fernandes Filho, Elpídio Ináciohttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4703656Z4Ferreira Neto, José Ambrosiohttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4723804D6Lani, João Luizhttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4783076P1Burak, Diego Langhttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4706238E8Soares, Vicente Paulohttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4781715A9Francelino, Márcio Rochahttp://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4794183U42015-03-26T12:52:53Z2014-03-242015-03-26T12:52:53Z2013-08-16CUNHA, Alexson de Mello. Selection of environmental variables and classification algorithms for digital soil mapping. 2013. 132 f. Tese (Doutorado em Fertilidade do solo e nutrição de plantas; Gênese, Morfologia e Classificação, Mineralogia, Química,) - Universidade Federal de Viçosa, Viçosa, 2013.http://locus.ufv.br/handle/123456789/1649In digital soil mapping the terrain attributes have been used as the main environmental predictor variables. Other variables related to pedogenic processes, such as climate changes, has not been used often. In this context, the objectives of this study were: (1) to evaluate the method of cokriging ordinary kriging compared to the spatial distribution of rainfall in the State of Espírito Santo, (2) identify soil mapping at different scales, which variables most relevant for better performance prediction of soil classes, whose study two distinct areas, (3) evaluate the performance of the NaiveBayes classifier algorithm, neural network multilayerperceptron - MLP SimpleCart and J48. In the evaluation of kriging and cokriging were used data from 108 rain gauges and secondary variables such as altitude and distance from the sea. For the evaluation of the variables and the mapping algorithms classifiers medium scale (1:100,000) the study was conducted in the river basin Muqui North, south of Espirito Santo. We used 598 training instances (10 soil classes) and 45variáveis related to different factors of soil formation, such as terrain attributes, geology, geomorphology, climate, water balance and indices derived from bands 1, 3, and 4 5 Landsat 5 TM. These variables were subjected to different methods of feature selection based on correlation - CFS, in consistency - CSE, information gain - and IA "ReliefF", available in the software Weka 3.6.8. This software was applied SimpleCart classifier to evaluate the effectiveness of the prediction with the subsets of selected variables. The evaluation of the four binders was performed using the 45 variables and the algorithm selected by "ReliefF." For detailed mapping conducted in Rural Settlement Sezínio Fernandes, Linhares, ES, we used 259 training instances (three soil classes) and 19 predictor variables (terrain attributes, climate and water balance) in the predictions made by the classifiers SimpleCart, J48, MLP and NaiveBayes. The predictions were evaluated based on cross- validation and comparisons of maps made with the conventional map reference. The interpolation results suggest that cokriging to be preferred to the use of regular grids for sampling secondary variables. The results of the selection of attributes for mapping river basin Muqui North indicated that the algorithms "ReliefF" and CSE, both limited xto 10 attributes were those with less complex trees and without significant loss in accuracy prediction compared to group 45 variables. The classification accuracy, indicated by Kappa of 0.60 was considered very good. The variables selected by "ReliefF" were geology, geomorphology and especially terrain attributes and elements of the water balance, as water surplus, water deficit and potential evapotranspiration. Algorithms NaiveBayes, MLP and SimpleCart showed similar performance prediction (Kappa 0.60 to 0.66), higher than the J48. The highest agreement with the reference map obtained by the MLP algorithm, followed by SimpleCart, J48 and NaiveBayes was 55, 52, 51 and 48%, respectively. The predictions of soil from Settlement Sezinio Fernandes variables water surplus, water deficit and air temperature were relevant. However, the small amplitude values presented by climatic variables and water balance are probably not sufficient to provide different pedogenetic conditions in the study area. There were no significant differences between the Kappa values (0.77 to 0.82) of the three classification algorithms. The greatest agreement with the conventional map was obtained for the algorithm J48, followed by NaiveBayes and SimpleCart,using only the terrain attributesas predictors variables. Decision trees for producing results more easily understood and presented in general accuracies similar to NaiveBayes classifiers and neural network MLP, may be regarded as of great potential to consolidate the digital soil mapping.No mapeamento digital de solos têm sido utilizados os atributos do terreno como as principais variáveis preditivas ambientais. Outras variáveis, relacionadas aos processos pedogenéticos, como as climáticas, não tem sido geralmente utilizadas. Nesse contexto, os objetivos deste estudo foram: (1) comparar o método da cokrigagem ordinária em com o da krigagem na espacialização da precipitação pluvial no Estado do Espírito Santo; (2) identificar no mapeamento de solos, em escalas diferentes, quais as variáveis mais relevantes para um melhor desempenho de predição das classes de solo, tendo como estudo duas áreas distintas; (3) avaliar o desempenho dos algoritmos classificadores NaiveBayes, rede neural multilayerperceptron - MLP, SimpleCart e J48.Na avaliação da krigagem e cokrigagem utilizaram-se dados de 108 postos pluviométricos e variáveis secundárias como altitude e distância do mar. Para a avaliação das variáveis e algoritmos classificadores no mapeamento de média escala (1:100.000) o estudo foi realizado na bacia do rio Muqui do Norte, sul do Estado do Espírito Santo. Utilizaram-se 598 amostras de treinamento (10 classes de solos) e 45variáveis relacionadas a diferentes fatores de formação dos solos, tais como: atributos do terreno, geologia, geomorfologia, clima, balanço hídrico e índices derivados das bandas 1, 3, 4 e 5 do sensor TM Landsat 5. Essas variáveis foram submetidas a diferentes métodos de seleção de atributos baseadas em correlação CFS; em consistência CSE; ganho de informação - IA e ReliefF , disponíveis no software Weka 3.6.8. Nesse software foi aplicado o classificador SimpleCart para avaliar a efetividade da predição com os subconjuntos de variáveis selecionadas. A avaliação dos quatro classificadores foi realizada com o uso das 45 variáveis e as selecionadas pelo algoritmo ReliefF . Para o mapeamento detalhado realizado no Assentamento Rural Sezínio Fernandes, em Linhares, ES, utilizaram-se 259 amostras de treinamento (3 classes de solos) e 19 variáveis preditivas (atributos do terreno, climáticas e balanço hídrico) nas predições feitas pelos classificadores SimpleCart, J48,MLP e NaiveBayes. As predições foram avaliadas com base na validação cruzada e comparações dos mapas elaborados com o mapa convencional de referência. Os resultados da interpolação sugerem que se deve preferir a cokrigagem e o uso de grades regulares para amostragem viiide variáveis secundárias. Os resultados da seleção de atributos para mapeamento da bacia do rio Muqui do Norte indicaram que os algoritmos ReliefF e o CSE, ambos limitados a 10 atributos, foram os que apresentaram árvores menos complexas e sem perda significativa na exatidão da predição em relação ao grupo de 45 variáveis. A exatidão da classificação, indicada pelo Kappa de 0,60, foi considerada muito boa. As variáveis selecionadas pelo ReliefF foram geologia, geomorfologia e principalmente atributos do terreno e elementos do balanço hídrico, como excedente hídrico, deficiência hídrica e evapotranspiração potencial. Os algoritmos NaiveBayes, MLP e SimpleCart apresentaram desempenhos de predição semelhantes (Kappa 0,60 a 0,66), superiores ao J48. A maior concordância com o mapa de referência obtida pelo algoritmo MLP, seguido do SimpleCart, J48 e NaiveBayes foi de 55, 52, 51 e 48%, respectivamente. Nas predições de classes de solos do Assentamento Sezinio Fernandes as variáveis excedente hídrico, deficiência hídrica e temperatura do ar foram relevantes. No entanto, as pequenas amplitudes de valores apresentadas pelas variáveis climáticas e balanço hídrico não são provavelmente suficientes para propiciar condições pedogenéticas diferenciadas na área de estudo. Não houve diferenças significativas entre os valores de Kappa (0,77 a 0,82) dos três algoritmos de classificação. As maiores concordâncias com o mapa convencional foram obtidas para o algoritmo J48, seguido do NaiveBayes e do SimpleCart, utilizando somente atributos de terreno como variáveis preditivas. As árvores de decisão por produzirem resultados de mais fácil entendimento e apresentarem em geral exatidões semelhantes aos classificadores NaiveBayes e rede neural MLP, podem ser consideradas como de grande potencial para se consolidarem no mapeamento digital de solos.application/pdfporUniversidade Federal de ViçosaDoutorado em Solos e Nutrição de PlantasUFVBRFertilidade do solo e nutrição de plantas; Gênese, Morfologia e Classificação, Mineralogia, Química,Mapeamento de solosRedes neuraisÁrvore de decisãoGeoprocessamentoSoil mappingNeural networksDecision TreeGeoprocessingCNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLOSeleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solosSelection of environmental variables and classification algorithms for digital soil mappinginfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/doctoralThesisinfo:eu-repo/semantics/openAccessreponame:LOCUS Repositório Institucional da UFVinstname:Universidade Federal de Viçosa (UFV)instacron:UFVORIGINALtexto completo.pdfapplication/pdf6172771https://locus.ufv.br//bitstream/123456789/1649/1/texto%20completo.pdf2e935edd7d5f1314c4001429329a77fbMD51TEXTtexto completo.pdf.txttexto completo.pdf.txtExtracted texttext/plain261486https://locus.ufv.br//bitstream/123456789/1649/2/texto%20completo.pdf.txt4c536d65dea8cbb300fa1039ad829630MD52THUMBNAILtexto completo.pdf.jpgtexto completo.pdf.jpgIM Thumbnailimage/jpeg3428https://locus.ufv.br//bitstream/123456789/1649/3/texto%20completo.pdf.jpgc2810b6c7bbd182f71dcad50ca57d3e8MD53123456789/16492016-04-07 23:11:09.329oai:locus.ufv.br:123456789/1649Repositório InstitucionalPUBhttps://www.locus.ufv.br/oai/requestfabiojreis@ufv.bropendoar:21452016-04-08T02:11:09LOCUS Repositório Institucional da UFV - Universidade Federal de Viçosa (UFV)false |
dc.title.por.fl_str_mv |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
dc.title.alternative.eng.fl_str_mv |
Selection of environmental variables and classification algorithms for digital soil mapping |
title |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
spellingShingle |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos Cunha, Alexson de Mello Mapeamento de solos Redes neurais Árvore de decisão Geoprocessamento Soil mapping Neural networks Decision Tree Geoprocessing CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO |
title_short |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
title_full |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
title_fullStr |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
title_full_unstemmed |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
title_sort |
Seleção de variáveis ambientais e de algoritmos de classificação para mapeamento digital de solos |
author |
Cunha, Alexson de Mello |
author_facet |
Cunha, Alexson de Mello |
author_role |
author |
dc.contributor.authorLattes.por.fl_str_mv |
http://lattes.cnpq.br/4853651139461402 |
dc.contributor.author.fl_str_mv |
Cunha, Alexson de Mello |
dc.contributor.advisor-co1.fl_str_mv |
Fernandes Filho, Elpídio Inácio |
dc.contributor.advisor-co1Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4703656Z4 |
dc.contributor.advisor-co2.fl_str_mv |
Ferreira Neto, José Ambrosio |
dc.contributor.advisor-co2Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4723804D6 |
dc.contributor.advisor1.fl_str_mv |
Lani, João Luiz |
dc.contributor.advisor1Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4783076P1 |
dc.contributor.referee1.fl_str_mv |
Burak, Diego Lang |
dc.contributor.referee1Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4706238E8 |
dc.contributor.referee2.fl_str_mv |
Soares, Vicente Paulo |
dc.contributor.referee2Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4781715A9 |
dc.contributor.referee3.fl_str_mv |
Francelino, Márcio Rocha |
dc.contributor.referee3Lattes.fl_str_mv |
http://buscatextual.cnpq.br/buscatextual/visualizacv.do?id=K4794183U4 |
contributor_str_mv |
Fernandes Filho, Elpídio Inácio Ferreira Neto, José Ambrosio Lani, João Luiz Burak, Diego Lang Soares, Vicente Paulo Francelino, Márcio Rocha |
dc.subject.por.fl_str_mv |
Mapeamento de solos Redes neurais Árvore de decisão Geoprocessamento |
topic |
Mapeamento de solos Redes neurais Árvore de decisão Geoprocessamento Soil mapping Neural networks Decision Tree Geoprocessing CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO |
dc.subject.eng.fl_str_mv |
Soil mapping Neural networks Decision Tree Geoprocessing |
dc.subject.cnpq.fl_str_mv |
CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO |
description |
In digital soil mapping the terrain attributes have been used as the main environmental predictor variables. Other variables related to pedogenic processes, such as climate changes, has not been used often. In this context, the objectives of this study were: (1) to evaluate the method of cokriging ordinary kriging compared to the spatial distribution of rainfall in the State of Espírito Santo, (2) identify soil mapping at different scales, which variables most relevant for better performance prediction of soil classes, whose study two distinct areas, (3) evaluate the performance of the NaiveBayes classifier algorithm, neural network multilayerperceptron - MLP SimpleCart and J48. In the evaluation of kriging and cokriging were used data from 108 rain gauges and secondary variables such as altitude and distance from the sea. For the evaluation of the variables and the mapping algorithms classifiers medium scale (1:100,000) the study was conducted in the river basin Muqui North, south of Espirito Santo. We used 598 training instances (10 soil classes) and 45variáveis related to different factors of soil formation, such as terrain attributes, geology, geomorphology, climate, water balance and indices derived from bands 1, 3, and 4 5 Landsat 5 TM. These variables were subjected to different methods of feature selection based on correlation - CFS, in consistency - CSE, information gain - and IA "ReliefF", available in the software Weka 3.6.8. This software was applied SimpleCart classifier to evaluate the effectiveness of the prediction with the subsets of selected variables. The evaluation of the four binders was performed using the 45 variables and the algorithm selected by "ReliefF." For detailed mapping conducted in Rural Settlement Sezínio Fernandes, Linhares, ES, we used 259 training instances (three soil classes) and 19 predictor variables (terrain attributes, climate and water balance) in the predictions made by the classifiers SimpleCart, J48, MLP and NaiveBayes. The predictions were evaluated based on cross- validation and comparisons of maps made with the conventional map reference. The interpolation results suggest that cokriging to be preferred to the use of regular grids for sampling secondary variables. The results of the selection of attributes for mapping river basin Muqui North indicated that the algorithms "ReliefF" and CSE, both limited xto 10 attributes were those with less complex trees and without significant loss in accuracy prediction compared to group 45 variables. The classification accuracy, indicated by Kappa of 0.60 was considered very good. The variables selected by "ReliefF" were geology, geomorphology and especially terrain attributes and elements of the water balance, as water surplus, water deficit and potential evapotranspiration. Algorithms NaiveBayes, MLP and SimpleCart showed similar performance prediction (Kappa 0.60 to 0.66), higher than the J48. The highest agreement with the reference map obtained by the MLP algorithm, followed by SimpleCart, J48 and NaiveBayes was 55, 52, 51 and 48%, respectively. The predictions of soil from Settlement Sezinio Fernandes variables water surplus, water deficit and air temperature were relevant. However, the small amplitude values presented by climatic variables and water balance are probably not sufficient to provide different pedogenetic conditions in the study area. There were no significant differences between the Kappa values (0.77 to 0.82) of the three classification algorithms. The greatest agreement with the conventional map was obtained for the algorithm J48, followed by NaiveBayes and SimpleCart,using only the terrain attributesas predictors variables. Decision trees for producing results more easily understood and presented in general accuracies similar to NaiveBayes classifiers and neural network MLP, may be regarded as of great potential to consolidate the digital soil mapping. |
publishDate |
2013 |
dc.date.issued.fl_str_mv |
2013-08-16 |
dc.date.available.fl_str_mv |
2014-03-24 2015-03-26T12:52:53Z |
dc.date.accessioned.fl_str_mv |
2015-03-26T12:52:53Z |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/doctoralThesis |
format |
doctoralThesis |
status_str |
publishedVersion |
dc.identifier.citation.fl_str_mv |
CUNHA, Alexson de Mello. Selection of environmental variables and classification algorithms for digital soil mapping. 2013. 132 f. Tese (Doutorado em Fertilidade do solo e nutrição de plantas; Gênese, Morfologia e Classificação, Mineralogia, Química,) - Universidade Federal de Viçosa, Viçosa, 2013. |
dc.identifier.uri.fl_str_mv |
http://locus.ufv.br/handle/123456789/1649 |
identifier_str_mv |
CUNHA, Alexson de Mello. Selection of environmental variables and classification algorithms for digital soil mapping. 2013. 132 f. Tese (Doutorado em Fertilidade do solo e nutrição de plantas; Gênese, Morfologia e Classificação, Mineralogia, Química,) - Universidade Federal de Viçosa, Viçosa, 2013. |
url |
http://locus.ufv.br/handle/123456789/1649 |
dc.language.iso.fl_str_mv |
por |
language |
por |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
eu_rights_str_mv |
openAccess |
dc.format.none.fl_str_mv |
application/pdf |
dc.publisher.none.fl_str_mv |
Universidade Federal de Viçosa |
dc.publisher.program.fl_str_mv |
Doutorado em Solos e Nutrição de Plantas |
dc.publisher.initials.fl_str_mv |
UFV |
dc.publisher.country.fl_str_mv |
BR |
dc.publisher.department.fl_str_mv |
Fertilidade do solo e nutrição de plantas; Gênese, Morfologia e Classificação, Mineralogia, Química, |
publisher.none.fl_str_mv |
Universidade Federal de Viçosa |
dc.source.none.fl_str_mv |
reponame:LOCUS Repositório Institucional da UFV instname:Universidade Federal de Viçosa (UFV) instacron:UFV |
instname_str |
Universidade Federal de Viçosa (UFV) |
instacron_str |
UFV |
institution |
UFV |
reponame_str |
LOCUS Repositório Institucional da UFV |
collection |
LOCUS Repositório Institucional da UFV |
bitstream.url.fl_str_mv |
https://locus.ufv.br//bitstream/123456789/1649/1/texto%20completo.pdf https://locus.ufv.br//bitstream/123456789/1649/2/texto%20completo.pdf.txt https://locus.ufv.br//bitstream/123456789/1649/3/texto%20completo.pdf.jpg |
bitstream.checksum.fl_str_mv |
2e935edd7d5f1314c4001429329a77fb 4c536d65dea8cbb300fa1039ad829630 c2810b6c7bbd182f71dcad50ca57d3e8 |
bitstream.checksumAlgorithm.fl_str_mv |
MD5 MD5 MD5 |
repository.name.fl_str_mv |
LOCUS Repositório Institucional da UFV - Universidade Federal de Viçosa (UFV) |
repository.mail.fl_str_mv |
fabiojreis@ufv.br |
_version_ |
1801213102137540608 |