Análisis geoestadístico multivariado a través de métodos funcionales y curvas de Andrews
Autores
Dueñas Herrera, María Paula
Director
Tipo de contenido
Trabajo de grado - Maestría
Idioma del documento
EspañolFecha de publicación
2017-05-26
Título de la revista
ISSN de la revista
Título del volumen
Documentos PDF
Resumen
En geoestadística multivariada, usualmente hay dos caminos de análisis: usar kriging de manera individual sobre cada variable o emplear cokiging tendiendo en cuenta la codependencia espacial entre las variables. En literatura geoestadística, se ha demostrado que la segunda opción, en general, permite una disminución en la varianza de predicción. No obstante, su uso es limitado cuando el número de variables aumenta, pues la estimación de la covarianza espacial a través del modelo lineal de corregionalización se vuelve computacionalmente compleja. En este trabajo, se explora el uso del kriging ordinario funcional basado en curvas de Andrews como alternativa al kriging y al cokriging en el contexto multuvariado. Para la comparación de metodologías se hace un estudio de simulación, y posteriormente se lleva a cabo la aplicación de la propuesta en un conjunto de datos reales.
Abstratc. In multivariate geostatistics, there are usually two ways for modeling the data: applying kriging individually on each variable or using cokriging, which takes into account the spatial cross-dependence between the variables. It has been shown the second option, in general, allows a prediction variance reduction. However, the use of cokriging in practice can be limited when the number of variables increases, as the estimation of the linear model of coregionalization becomes complex. In this work, the use of ordinary functional kriging based on Andrews Curves is explored as an alternative in the multivariate scenerie. This approach is compared to kriging and cokriging based on a simulation study and at the end the proposed methodology is applied to a real dataset.
Abstratc. In multivariate geostatistics, there are usually two ways for modeling the data: applying kriging individually on each variable or using cokriging, which takes into account the spatial cross-dependence between the variables. It has been shown the second option, in general, allows a prediction variance reduction. However, the use of cokriging in practice can be limited when the number of variables increases, as the estimation of the linear model of coregionalization becomes complex. In this work, the use of ordinary functional kriging based on Andrews Curves is explored as an alternative in the multivariate scenerie. This approach is compared to kriging and cokriging based on a simulation study and at the end the proposed methodology is applied to a real dataset.