使用 TERRA package 从插值栅格计算长期平均月降雨量？

Question

I have a SpatRaster object in R called IDW3, estimated using IDW interpolation method. I have nlyr = 240 , containing 12 months x 20 years. I need to calculate the long-term mean monthly rainfall from the layers, so that I get nlyr = 12 at the end, in which each layer represents one calendar month (Jan - Dec).

I have tried using the code below, following this thread calculating long term daily means from a RASTER in R , but I want to verify the code I used.

Any thoughts and comments please?

idw3
#> class       : SpatRaster 
#> dimensions  : 723, 449, 240  (nrow, ncol, nlyr)
#> resolution  : 100, 100  (x, y)
#> extent      : 624698.7, 669598.7, 640507.8, 712807.8  (xmin, xmax, ymin, ymax)
#> coord. ref. :  
#> sources     : May 1998_masked_idw3.asc  
#>               May 1999_masked_idw3.asc  
#>               May 2000_masked_idw3.asc  
#>               ... and 237 more source(s)
#> names       :     Jan 1998,     Jan 1999,     Jan 2000,     Jan 2001,     #> Jan 2002,     Jan 2003, ... 
#> min values  :           ? ,           ? ,           ? ,           ? ,           ? ,           ? , ... 
#> max values  :           ? ,           ? ,           ? ,           ? ,           ? ,           ? , ... 

## CALCULATE THE LONGTERM MONTHLY MEANS
# get the months substring
month.ltm <- substr(my, 1,3)

# calculate the ltm using tapp funtion in terra
idw3.ltm <- tapp(idw3, month.ltm, mean)
names(idw3.ltm)
#> [1] "May" "Apr" "Aug" "Jan" "Sep" "Jul" "Jun" "Feb" "Dec"
#> [10] "Nov" "Oct" "Mar"

Answer 1

It's been some time since you posted this, but I'd try to solve this for the next one asking. Since your example does not seem to be fully reproducible, let my use my own data for this purpose.

I used GPCC v2022 data - the last two decades of monthly data at 0.5° resolution to be precise - from German Weather Service.

library(terra)
#> terra 1.5.21

# define filenames
files <- c("full_data_monthly_v2022_2001_2010_05.nc",
           "full_data_monthly_v2022_2011_2020_05.nc")

# create SpatRaster object
nc_data <- rast(files)

# get variable names
varnames(nc_data)
#>  [1] "precip"                       "numgauge"                    
#>  [3] "infilled_numgauges"           "interpolation_error"         
#>  [5] "interpolation_error_infilled" "diff_new_old_method"         
#>  [7] "precip"                       "numgauge"                    
#>  [9] "infilled_numgauges"           "interpolation_error"         
#> [11] "interpolation_error_infilled" "diff_new_old_method"

# subset dataset to precipitation only
nc_precip <- nc_data["precip"]

# sneak peek
nc_precip
#> class       : SpatRaster 
#> dimensions  : 360, 720, 240  (nrow, ncol, nlyr)
#> resolution  : 0.5, 0.5  (x, y)
#> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
#> coord. ref. : lon/lat WGS 84 
#> sources     : full_data_monthly_v2022_2001_2010_05.nc:precip  (120 layers) 
#>               full_data_monthly_v2022_2011_2020_05.nc:precip  (120 layers) 
#> varnames    : precip (gpcc full data monthly product version 2022, precipitation per grid) 
#>               precip (gpcc full data monthly product version 2022, precipitation per grid) 
#> names       : precip_1, precip_2, precip_3, precip_4, precip_5, precip_6, ... 
#> unit        : mm/month, mm/month, mm/month, mm/month, mm/month, mm/month, ... 
#> time        : 2001-01-01 to 2020-12-01

As you can see, this dataset is quite similar to yours in terms of information at least (except for crs, extent and resolution). A stack of SpatRaster objects with nlyr = 240 containing monthly precipitation data. What differs most notably is the time attribute ranging from 2001-01-01 to 2020-12-01.

However, basically I approached your issue constructing an appropriate time-based index vector as input to tapp using fun = mean :

# get timestamps from your SpatRaster object
tst <- terra::time(nc_precip) 

# calculate monthly means
lta <- tapp(nc_precip, index = 1:12, fun = mean)

# tidy your names a little bit
names(lta) <- format(tst, "%B") |> unique()

# inspect result
lta
#> class       : SpatRaster 
#> dimensions  : 360, 720, 12  (nrow, ncol, nlyr)
#> resolution  : 0.5, 0.5  (x, y)
#> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
#> coord. ref. : lon/lat WGS 84 
#> sources     : memory  
#>               memory  
#>               memory  
#>               ... and 9 more source(s)
#> names       :   January,  February,     March,     April,       May,      June, ... 
#> min values  :         0,         0,         0,         0,         0,         0,     ... 
#> max values  :  979.1880,  852.0020,  720.6245,  739.8225,  884.2455, 1590.6805,     ...

The result seems plausible from my point of view, but since this is the first time I used tapp , I want to make sure the function behaves as expected by re-calculating manually:

# init an empty list for temporary storage purposes
lta <- list()

# loop monthly and calculate the long-term mean
for (i in 1:12) {
  
  idx <- seq(from = i, by = 12, length.out = n_years)
  
  lta[[i]] <- nc_precip[[idx]] |> terra::mean()
}

# create a SpatRast object with nlyr = 12
lta <- terra::rast(lta)

lta
#> class       : SpatRaster 
#> dimensions  : 360, 720, 12  (nrow, ncol, nlyr)
#> resolution  : 0.5, 0.5  (x, y)
#> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
#> coord. ref. : lon/lat WGS 84 
#> sources     : memory  
#>               memory  
#>               memory  
#>               ... and 9 more source(s)
#> names       :      mean,      mean,      mean,      mean,      mean,      mean, ... 
#> min values  :         0,         0,         0,         0,         0,         0, ... 
#> max values  :  979.1880,  852.0020,  720.6245,  739.8225,  884.2455, 1590.6805, ...

Same results, phew.

Edit:

After some weird behaviour yesterday which cannot be reproduced today I can confirm that using index = months.abb gives you the same results as using index = "months" (as suggested by Roger below in the comments):

tapp(nc_precip, index = month.abb, fun = mean)
#> class       : SpatRaster 
#> dimensions  : 360, 720, 12  (nrow, ncol, nlyr)
#> resolution  : 0.5, 0.5  (x, y)
#> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
#> coord. ref. : lon/lat WGS 84 
#> source      : memory 
#> names       :       Jan,       Feb,       Mar,       Apr,       May,       Jun, ... 
#> min values  :         0,         0,         0,         0,         0,         0, ... 
#> max values  :  979.1880,  852.0020,  720.6245,  739.8225,  884.2455, 1590.6805, ...

tapp(nc_precip, index = "months", fun = mean)
#> class       : SpatRaster 
#> dimensions  : 360, 720, 12  (nrow, ncol, nlyr)
#> resolution  : 0.5, 0.5  (x, y)
#> extent      : -180, 180, -90, 90  (xmin, xmax, ymin, ymax)
#> coord. ref. : lon/lat WGS 84 
#> source      : memory 
#> names       :        X1,        X2,        X3,        X4,        X5,        X6, ... 
#> min values  :         0,         0,         0,         0,         0,         0, ... 
#> max values  :  979.1880,  852.0020,  720.6245,  739.8225,  884.2455, 1590.6805, ...

Answer 2

You can use tapp for that. Here are some example data

library(terra)
r <- rast(ncols=10, nrows=10, nlyr=24)
values(r) <- 1:size(r)

If the data are ordered by year, and then by month, you can now do

x <- tapp(r, 1:12, mean)

In other cases you may have to create another index to match the layers that need to be combined. If your data has a time-stamp, there are some shortcuts. In this case you can use index="months"

time(r) <- as.Date("2000-01-15") + 0:23 * 30.5
y <- tapp(r, "months", mean)