我有两组栅格,两者都有相同的x,y,z范围。我做了两个堆栈:stacka和stackb。我想计算沿着时间线的两个堆栈之间的每个网格单元中的Pearson相关系数(PCC)。我做了一个更简单的例子(原谅我用愚蠢的方式创建栅格)
a1<-c(1,1,1,1,1,1,1,1,NA)
a2<-c(2,2,2,2,1,2,2,NA,2)
a3<-c(3,3,3,3,3,2,NA,3,3)
b1<-c(2,2,2,2,2,2,2,2,2)
b2<-c(3,3,3,3,3,3,3,3,3)
b3<-c(4,4,4,4,4,4,4,4,4)
matrixa1<-matrix(a1,3,3)
matrixa2<-matrix(a2,3,3)
matrixa3<-matrix(a3,3,3)
matrixb1<-matrix(b1,3,3)
matrixb2<-matrix(b2,3,3)
matrixb3<-matrix(b3,3,3)
rastera1<-raster(matrixa1)
rastera2<-raster(matrixa2)
rastera3<-raster(matrixa3)
rasterb1<-raster(matrixb1)
rasterb2<-raster(matrixb2)
rasterb3<-raster(matrixb3)
stacka<-stack(rastera1,rastera2,rastera3)
stackb<-stack(rasterb1,rasterb2,rasterb3)
a_bar<-calc(stacka,mean,na.rm=TRUE)
b_bar<-calc(stackb,mean,na.rm=TRUE)
numerator<-setValues(rastera1,0)
denominator1<-numerator
denominator2<-numerator
for(i in 1:noflayers){
numerator<-numerator+(stacka[[i]]-a_bar)*(stackb[[i]]-b_bar)
denominator1<-denominator1+(stacka[[i]]-a_bar)^2
denominator2<-denominator2+(stackb[[i]]-b_bar)^2
}
pearsoncoeff<-numerator/sqrt(denominator1*denominator2)
最后,我有一个栅格,每个网格单元都充满了PCC。问题是,数据a是间歇性的,某些网格中的某些网格是NA。所以最终产品有一些空白。我的算法在遇到NA时吐出“NA”。我在计算中需要一些像na.rm=TRUE
这样的选项,所以输出会计算任何月份的值。
我能想到的方法是使用is.na(stacka[[nlayers]][nrows,ncols]==FALSE
并在stackb中找到相应的对,但这是基于单元格的,这需要大量的计算机时间。
答案 0 :(得分:3)
我编辑了Paulo在计算中处理NAs的推荐方法,它似乎在一系列测试中运行得很快,包括上面的数据集:
stack.correlation <- function(stack1, stack2, cor.method){
# output template
cor.map <- raster(stack1)
# combine stacks
T12 <- stack(stack1,stack2)
rnlayers=nlayers(T12)
# the function takes a vector, partitions it in half, then correlates
# the two sections, returning the correlation coefficient.
stack.sequence.cor <- function(myvec,na.rm=T){
myvecT1<-myvec[1:(length(myvec)/2)]
myvecT2<-myvec[(length(myvec)/2+1):length(myvec)]
return(cor(myvecT1,myvecT2, method = cor.method, use="complete.obs"))
}
# apply the function above to each cell and write the correlation
# coefficient to the output template.
cor.map <- stackApply(T12, indices = rep(1, rnlayers),
fun = stack.sequence.cor, na.rm = FALSE)
return(cor.map)
}
cor_r=stack.correlation(stacka, stackb, "pearson")
答案 1 :(得分:3)
一种稍微简单的方法:
library(raster)
a1 <- raster(matrix(c(1,1,1,1,1,1,1,1,NA),3,3))
a2 <- raster(matrix(c(2,2,2,2,1,2,2,NA,2), 3, 3))
a3 <- raster(matrix(c(3,3,3,3,3,2,NA,3,3), 3, 3))
b1 <- raster(matrix(c(2,2,2,2,2,2,2,2,2), 3, 3))
b2 <- raster(matrix(c(3,3,3,3,3,3,3,3,3), 3, 3))
b3 <- raster(matrix(c(4,4,4,4,4,4,4,4,4), 3, 3))
sa <- stack(a1, a2, a3)
sb <- stack(b1, b2, b3)
funcal <- function(xy) {
xy <- na.omit(matrix(xy, ncol=2))
if (ncol(xy) < 2) {
NA
} else {
cor(xy[, 1], xy[, 2])
}
}
s <- stack(sa, sb)
calc(s, funcal)