Comparação das técnicas Kernel e Krigagem indicativa na predição de valores de variáveis espacialmente distribuídas - estudos de caso
Carregando...
Data
2020-02-28
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O Kernel e a Krigagem Indicativa são duas técnicas utilizadas na análise espacial para estimar a ocorrência de determinado fenômeno em uma área de estudo. Com o conhecimento da distribuição espacial para casos de doenças epidêmicas é possível criar estratégias de controle de transmissão. O objetivo deste trabalho foi estudar ambas técnicas de estatística espacial usando conjuntos de dados epidemiológicos, dengue e malária, de duas cidades, Rio Claro-SP e Porto Velho-RO, respectivamente, a fim de observar os locais com maiores ocorrências e exemplificar para comparação das técnicas para verificar qual teve melhor precisão nas predições. A partir desses resultados é possível verificar as áreas em que ocorreram os casos e tomar providências para diminuição ou erradicação das mesmas. A técnica Krigagem Indicativa gerou mapas que representam a probabilidade do valor de corte a ser superado e a técnica kernel gerou mapas que representam a intensidade de ocorrência em cada ponto. Ambas foram eficientes ao mostrar as áreas de concentração das endemias, mas a falta de variáveis adicionais nos bancos de dados, nos impôs algumas limitações ao fazer as comparações das técnicas; com isso, o melhor meio de compará-las foi através dos critérios de informação Akaike (AIC) e Bayesiano (BIC). Os valores encontrados nos critérios de informação para ambas as técnicas foram condizentes com os encontrados na literatura, mas para uma comparação não foi tão eficaz, pois apresentou valores muito distantes, porém com um critério visual dos mapas, umas das técnicas se destacou para os dois conjuntos de dados. Ao final concluímos que essas técnicas não são comparáveis, mas sim complementares, pois a técnica Kernel mostrou a intensidade das endemias e a técnica Krigagem Indicativa mostrou a probabilidade desses valores serem maiores que um ponto de corte determinado.
The Kernel and estima the Kriging indicative are two techniques used in spatial analysis to estimate the occurrence of a certain phenomenon on a field of study. With knowledge on spatial distribution for cases of epidemic diseases, it is possible to create transmission control strategies. The objective of this paper was to study both spatial statistics techniques using epidemiological data sets, dengue and malaria, from two cities, Rio Claro-SP and Porto Velho-RO, respectively, in order to observe the places which present the highest occurrence rates and to exemplify comparing the techniques and to observe which ones presented a better accuracy on their predictions. From these results it is possible to check the areas where the cases occur and take steps to reduce or eradicate them. The Indicative Kriging technique generated maps that represent a probability of the cut value being surpassed and the Kernel technique generated maps that represent the intensity of occurrence at each point. Both of them were efficient in showing areas of concentration of endemics, but the lack of additional variables in the databases ended up causing some restrictions when making comparing the techniques; therefore, the best means of comparison was through the use of Akaike (AIC) and Bayesian (BIC) information. The values found in the information criteria for both techniques were similar to those found in literature, but as a means of comparison it was not as effective, as it shows very distant values, but with a visual criteria of the maps, one of the techniques was highlighted for the two data sets. In the end, we conclude that these techniques are not comparable, but rather complementary, since a kernel technique showed intensity of endemic diseases and an Indicative Kriging technique showed the probability of these values being greater than a determined cutoff point.
The Kernel and estima the Kriging indicative are two techniques used in spatial analysis to estimate the occurrence of a certain phenomenon on a field of study. With knowledge on spatial distribution for cases of epidemic diseases, it is possible to create transmission control strategies. The objective of this paper was to study both spatial statistics techniques using epidemiological data sets, dengue and malaria, from two cities, Rio Claro-SP and Porto Velho-RO, respectively, in order to observe the places which present the highest occurrence rates and to exemplify comparing the techniques and to observe which ones presented a better accuracy on their predictions. From these results it is possible to check the areas where the cases occur and take steps to reduce or eradicate them. The Indicative Kriging technique generated maps that represent a probability of the cut value being surpassed and the Kernel technique generated maps that represent the intensity of occurrence at each point. Both of them were efficient in showing areas of concentration of endemics, but the lack of additional variables in the databases ended up causing some restrictions when making comparing the techniques; therefore, the best means of comparison was through the use of Akaike (AIC) and Bayesian (BIC) information. The values found in the information criteria for both techniques were similar to those found in literature, but as a means of comparison it was not as effective, as it shows very distant values, but with a visual criteria of the maps, one of the techniques was highlighted for the two data sets. In the end, we conclude that these techniques are not comparable, but rather complementary, since a kernel technique showed intensity of endemic diseases and an Indicative Kriging technique showed the probability of these values being greater than a determined cutoff point.
Descrição
Palavras-chave
Critérios de Informação, Semivariograma, Raio de Influência, Dengue, Malária, Rio Claro, Porto Velho