如何使用gstat预测普通克里金法 [英] How to make Ordinary Kriging by using gstat predict

问题描述

我试图在R中使用gstat库编写代码以创建插值. 我已经阅读了gstat手册，并且基于互联网上的一些示例，我设法编写了这段代码(这只是一部分):

I am trying to write a code in R that use gstat library in order to create an interpolation. I have already read the gstat manual and based on some examples on internet I had managed to write this code (this is only a part):

 g <- gstat(id="tec", formula=TEC ~ 1, data=data)  ##I create an object
 v <- variogram(g) # plot the empirical variogram
 plot(v)
 mod<-vgm(sill=var(data$TEC),model="Sph",range=200,nugget=200) #create the variogram model

v.fit <- fit.variogram(v, model=mod,fit.method=1)  #fit the empirical variogram 
Theor_variogram=plot(variogram(g),v.fit,main="WLS Model") #plot the theoretical variogram
plot(Theor_variogram)
 ## Kriging interpolation
 p <- predict.gstat(g, model=v.fit, newdata=predGrid)

我的问题是，当我运行最后一个命令(预测)而不是使用普通克里格插值获得结果时，我得到了一个具有反距离权重(IDW)的结果. 我在gstat手册中读到:当未指定任何方差图时，反距离加权插值 是默认操作. 当指定了方差图时，默认的预测方法是普通的 克里金法."

My problem is that, when I run the last command (predict) instead of getting a result with ordinary kriging interpolation, I get one with inverse distance weighted (IDW). I read in the gstat manual that: "When no variograms are specified, inverse distance weighted interpolation is the default action. When variograms are specified the default prediction method is ordinary kriging."

但是，正如您在我的代码中看到的那样，我同时指定了经验方差和理论方差. 您知道为什么我会继续获得IDW而不是普通的克里金法吗?它可以与我拥有的坐标类型相关吗?例如，如果我的坐标彼此接近，或者感兴趣的区域太大? 任何帮助都将非常有用.

But, as you can see in my code, I specify both the empirical and theoretical variogram. Do you know why I keep getting IDW instead of ordinary kriging? Can it be related with the type of coordinates that I have? If for example I have coordinates close to each other, or if the region of interest is too big? Any help would be really useful.

先谢谢了 Dimitris

Thanks in advance Dimitris

推荐答案

您需要包括gstat对象的创建，而不是在预测阶段:

You need to include the creation of the gstat object, not in de prediction phase:

g <- gstat(id="tec", formula=TEC ~ 1, data=data, model = v.fit)

但是，我建议使用krige为gstat使用标准接口.这将gstat对象的构建和预测合并为一个函数.您很少需要自己构建gstat对象.例如:

However, I would recommend using the standard interface for gstat using krige. This combines the building of the gstat object and the prediction into one functions. Very rarely do you need to build gstat objects yourself. For example:

data(meuse)
coordinates(meuse) = ~x+y
data(meuse.grid)
gridded(meuse.grid) = ~x+y
m <- vgm(.59, "Sph", 874, .04) 
# OK:
x <- krige(log(zinc)~1, meuse, meuse.grid, model = m)

您还可以使用automap包(我是作者)，并让方差图模型自动拟合到数据中.例如，使用meuse数据集:

You could also use the automap package (which I am the author of) and let the variogram model be automatically fitted to the data. For example using the meuse dataset:

library(automap)
kr = autoKrige(log(zinc)~1, meuse, meuse.grid)

这将自动构建样本方差图，并为该样本方差图拟合方差图模型.

This will automatically build a sample variogram, and fit a variogram model to that sample variogram.

这篇关于如何使用gstat预测普通克里金法的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！