您如何使用循环在多个站点上执行 Mann-kendall 测试？

Question

I've just started using R and would like to use the modifiedmk package to perform tests on monthly groundwater level data. My dataframe (GL) looks something like this

GL

well    year    month   value
684     1994    Jan     8.53
684     1995    Jan     8.74
684     1996    Jan     8.88
684     1997    Jan     8.24
1001    2000    Jan     9.1
1001    2001    Jan     9.2
1001    2002    Jan     9.54
1001    2003    Jan     9.68
2003    1981    Jan     55.2
2003    1982    Jan     55.8
2003    1983    Jan     56.4
2003    1984    Jan     53.2

First I have created a list of wells and a results_list file for printing the results

well_list <- unique(GL$well)
results_list <- vector("list", length(well_list))

Then I have created what I think is a loop

for(i in well_list){
  results_list[[grep(i, well_list)]] <- MannKendall(GL[,4])
}
names(results_list) <- well_list

but I kept getting this error

Error in Kendall(1:length(x), x) : length(x)<3

I can get this code to work (which I took from another post here) but I don't want to perform the pre-whitening method

for(i in well_list){
tempDF <- GL1[GL$well == i, ]
c<-acf(tempDF$value,lag.max=1)
t <- dim(c$acf)
tempDF$prewhit1<-c$acf[[t[1], t[2], t[3]]]*tempDF$value
prewhitseries<-data.frame(with(tempDF,(tempDF$value[-1] - prewhit1[ 
length(prewhit1)])))
autocordata<-cbind(tempDF[-1,],prewhitseries)
results_list[[grep(i, well_list)]] <- MannKendall(autocordata[,5])}
names(results_list) <- well_list

Answer 1

Something like this should do. split() splits along the well column, creating a list with a vector for each well. Only vectors of length 3 or more are kept. MannKendall() is then run on each of the remaining vectors using lapply()

library(Kendall)

tt <- read.table(text="
well    year    month   value
684     1994    Jan     8.53
684     1995    Jan     8.74
684     1996    Jan     8.88
684     1997    Jan     8.24
1001    2000    Jan     9.1
1001    2001    Jan     9.2
1001    2002    Jan     9.54
1001    2003    Jan     9.68
2003    1981    Jan     55.2
2003    1982    Jan     55.8
2003    1983    Jan     56.4
2003    1984    Jan     53.2
2004    1984    Jan     53.2", header=TRUE)

tt.wells <- split(tt$value, tt$well)
tt.wells <- tt.wells[lengths(tt.wells) >= 3]

lapply(tt.wells, MannKendall)

# $`684`
# tau = 0, 2-sided pvalue =1

# $`1001`
# tau = 1, 2-sided pvalue =0.089429

# $`2003`
# tau = 0, 2-sided pvalue =1

Answer 2

Maybe with split/lapply it is easier and more readable.

First, runs the tests.

sp <- split(GL, GL$well)
results_list <- lapply(sp, function(DF){
  tryCatch(MannKendall(DF[, 4]),
           error = function(e) e)
})

Now, get the good ones, with no errors.

bad <- sapply(results_list, inherits, "error")

And inspect them.

results_list[bad]
#named list()

results_list[!bad]
#$`684`
#tau = 0, 2-sided pvalue =1
#
#$`1001`
#tau = 1, 2-sided pvalue =0.089429
#
#$`2003`
#tau = 0, 2-sided pvalue =1