Thesis OPTIMIZACIÓN DE LA RESPUESTA A LA FERTILIZACIÓN EN TRIGO USANDO HERRAMIENTAS DE ECONOMETRÍA ESPACIAL.
Loading...
Date
2013-04
Journal Title
Journal ISSN
Volume Title
Program
DEPARTAMENTO DE INDUSTRIAS. INGENIERÍA CIVIL INDUSTRIAL
Campus
Campus Vitacura, Santiago
Abstract
El objetivo de este trabajo es mejorar la precisión de los modelos de regresión para estimar la respuesta a la fertilización en trigo, y también el demostrar que, cuando se tienen variables regionales, no basta con un modelo de regresión lineal múltiple, si no que se deben tomar en cuenta los efectos espaciales. A partir de la base de datos de un predio ubicado en la VIII Región del Biobío, con un total de 785 observaciones y 24 variables, que incluyen: coordenadas, rendimiento, macronutrientes, micronutrientes, nutrientes intercambiables, texturas y otras propiedades del terreno, se realizaron dos modelos lineales como base:Lm1 y Lm2, los cuales difieren en solo dos variables, siendo una de estas la coordenada x. Los modelos lineales no fueron válidos, ya que se consideraron sesgados al poseer autocorrelación espacial en sus residuales. Éstos se repararon con modelos autorregresivos AR1 y AR2, con los cuales se obtuvo un R2corregido de 73,35% y 73,41% respectivamente. Por otra parte, se utilizó el algoritmo de Fuzzy K-Means para obtener el número óptimo de Clusters, dando como resultado 9 agrupaciones. Con el software Vesper se realizó la interpolación puntual de Kriging de la variable de respuesta, paralas distintas dosis de la variable insumo(nrate).De los resultados obtenidos, el mejor modelo fue aquel que incluyó Cluster, conun R2 corregido =84,86%. Además, se realizó una validación cruzada con 20 observaciones, obteniéndose un R2=57,80%y,para el cual, su valor óptimo generó una utilidad de 126.773[$/ha].
The aimof this work is to improve theregression modelprecision to estimate the response of the wheatfertilization. Also, this investigation highlights the fact that while regional variables are used, a multiple linear regression is not enough. Considering all this evidence,spatial effects must be considered.In order to convey thisinvestigation,two linear models,Lm1 and Lm2,were performedkeeping their main characteristics.Theydiffer in only two variables, being one of them the x coordinate.They were developed from the database of alandlocated in Region VIIIofBiobío, with a total of 785 observations and 24 variables which include: coordinates, yield, macronutrients, micronutrients, exchangeable nutrients, textures and other propertiesof land.The linear models were not valid since they were considered to be slanted while having spatial autocorrelation onits residual. These autocorrelationswere repaired by AR1 and AR2,which areautoregressivemodels,obtaininga correctedR2equalto 73,35% and 73,41% respectively.On the other hand, Fuzzy K-Meansalgorithm wasusedto obtainthe ideal number of Clusters, generating a result of ninegroups.With Vesper software, Krigingpunctual interpolation of the response variable was performed. Usingdifferent doses of the supplies (nrate) variable.Fromthe obtained results, of this investigation, the best model was theone that included Clusters, with a corrected R2= 84,86 %. Across-validation wasalso performedwith 20 observations, giving anR2= 57,80 %. In addition,theoptimal value of the Clusters generateda utility of 126.773[$/ha].
The aimof this work is to improve theregression modelprecision to estimate the response of the wheatfertilization. Also, this investigation highlights the fact that while regional variables are used, a multiple linear regression is not enough. Considering all this evidence,spatial effects must be considered.In order to convey thisinvestigation,two linear models,Lm1 and Lm2,were performedkeeping their main characteristics.Theydiffer in only two variables, being one of them the x coordinate.They were developed from the database of alandlocated in Region VIIIofBiobío, with a total of 785 observations and 24 variables which include: coordinates, yield, macronutrients, micronutrients, exchangeable nutrients, textures and other propertiesof land.The linear models were not valid since they were considered to be slanted while having spatial autocorrelation onits residual. These autocorrelationswere repaired by AR1 and AR2,which areautoregressivemodels,obtaininga correctedR2equalto 73,35% and 73,41% respectively.On the other hand, Fuzzy K-Meansalgorithm wasusedto obtainthe ideal number of Clusters, generating a result of ninegroups.With Vesper software, Krigingpunctual interpolation of the response variable was performed. Usingdifferent doses of the supplies (nrate) variable.Fromthe obtained results, of this investigation, the best model was theone that included Clusters, with a corrected R2= 84,86 %. Across-validation wasalso performedwith 20 observations, giving anR2= 57,80 %. In addition,theoptimal value of the Clusters generateda utility of 126.773[$/ha].
Description
Keywords
FERTILIZACIÓN EN TRIGO, REGIÓN DEL BIOBÍO