R的不同层和几个netcdf文件的平均值

Question

I have 15 netCDF files (.nc) for each year from 2000 to 2014. In one nc file, I have hourly data of one variable in 8760 layers. The 3 dimensions are: time (size 8760), latitude (size 90) and longitude (size 180) (2° resolution).

I want to compute the average of my variable between 8am and 7pm from april to september and for the period 2000-2014.

For one .nc file, this correspond to the average between

• layer time from 2169 (ie 01/04/2000 8am) to 2180 (ie 01/04/2000 7pm) (to i=2169 to i+11),
• then from 2193 (ie 02/04/2000 8am) to 2204 (ie 02/04/2000 7pm) (i+22, i+33)
• etc....
• ... and from 6537 (ie 30/09/2000 8am) to 6548 (ie 30/09/2000 7pm)
• And then the average of all nc. files.

The result should be presented in one .nc file of 3 dimensions : - time (only one value as average), - latitude (size 90) and - longitude (size 180) (2° resolution)

then I can draw the map of the variable averaged over 2000-2014 (Apr to Sept, from 8am to 7pm). I am able to read each nc file, do a map for each hour ofeach nc file, but I have know idea of how to make the mean required. If anybody can help me, that would be great.

name of my variable : dname <- "sfvmro3"

Here is my code as a fist reading:

ncin <- nc_open("sfvmro3_hourly_2000.nc")
print(ncin)

lon <- ncvar_get(ncin, "lon")
lon[lon > 180] <- lon[lon > 180] - 360
nlon <- dim(lon)
head(lon)

lat <- ncvar_get(ncin, "lat", verbose = F)
nlat <- dim(lat)
head(lat)

print(c(nlon, nlat))

t <- ncvar_get(ncin, "time")
tunits <- ncatt_get(ncin, "time", "units")
nt <- dim(t)

dname <- "sfvmro3"
var.array <- ncvar_get(ncin, dname)*10^9  # from mol.mol-1 to ppb
dlname <- ncatt_get(ncin, dname, "long_name")
dunits <- ncatt_get(ncin, dname, "units")
fillvalue <- ncatt_get(ncin, dname, "_FillValue")
var.array[var.array == fillvalue$value] <- NA
dim(var.array)

tustr <- strsplit(tunits$value, " ")
tdstr <- strsplit(unlist(tustr)[3], "-")
tyear = as.integer(unlist(tdstr)[1])
tmonth = as.integer(unlist(tdstr)[2])
tday = as.integer(unlist(tdstr)[3])
chron = chron(t, origin = c(tmonth, tday, tyear))

Here are the details on one of the yearly file.nc:

 4 variables (excluding dimension variables):
    double time_bnds[bnds,time]   
    double lat_bnds[bnds,lat]   
    double lon_bnds[bnds,lon]   
    float sfvmro3[lon,lat,time]   
        standard_name: mole_fraction_of_ozone_in_air
        long_name: Ozone Volume Mixing Ratio in the Lowest Model Layer
        units: mole mole-1
        original_name: O_x
        original_units: 1
        history: 2016-04-22T05:20:31Z altered by CMOR: Converted units from '1' to 'mole mole-1'.
        cell_methods: time: point (interval: 30 minutes)
        cell_measures: area: areacella
        missing_value: 1.00000002004088e+20
        _FillValue: 1.00000002004088e+20
        associated_files: ...

 4 dimensions:
    time  Size:8760   *** is unlimited ***
        bounds: time_bnds
        units: days since 1850-01-01
        calendar: noleap
        axis: T
        long_name: time
        standard_name: time
    lat  Size:90
        bounds: lat_bnds
        units: degrees_north
        axis: Y
        long_name: latitude
        standard_name: latitude
    lon  Size:180
        bounds: lon_bnds
        units: degrees_east
        axis: X
        long_name: longitude
        standard_name: longitude
    bnds  Size:2

26 global attributes:
    institution: aaaa
    institute_id: aaaa
    experiment_id: aaaa
    source: aaaa
    model_id: aaaa
    forcing: HG, SA, S
    parent_experiment_id: N/A
    parent_experiment_rip: N/A
    branch_time: 0
    contact: aaa
    history: aaa
    initialization_method: 1
    physics_version: 1
    tracking_id: aaa
    product: output
    experiment: aaa
    frequency: hr
    creation_date: 2016-04-22T05:20:31Z
    Conventions: aaa
    project_id: aaa
    table_id:aaa
    title: aaaa
    parent_experiment: N/A
    modeling_realm: aaa
    realization: 1
    cmor_version: 2.7.1

Answer 1

I know two diferent possible solutions for your problem. One is base on taking the average for each .nc file and then take a weight average of that, the other is to get a really large array and average using that array.

• First possible solution

Each .nc that you read will give you and array, array1, array2 and so on. Also for each array you will have a time series associated to one dimension of the array. This meaning that time_serie1 has all the different times in POSIXct format for array1. So first you have to build in that vector. One you have that you can get a vector index of the times you want to use for average. For this I would use lubridate package but it is not necessary.

index1 <- month(time_serie1) < 10 & month(time_serie1) > 3 # this make an index from april to septembre
index1 <- index1 & hour(time_serie1) <= 19 & hour(time_serie1) >= 8 # then you add the hour restriction
mean1 <- apply(array1[,,index1],1:2,mean)

This code will give you a 2D array with the mean for the first year, you can put your arrays and time_series into list and loop it. Then you will have for each year a 2d array of the mean for that year and you can average this arrays. The part of "weight" average that I said is because if you do this and in your average you include February your's means will have be done taking different amount of days, for your example it is not necesary, but if you use February then you have to weight the amount of data used for each mean value.

• Second possible solution

For this solution is almost the same than the other one, but I like it more. You can merge all your arrays into a big array doing it in order so the time index is in increasing order, I will call this array BigArray. Then merge the Time series associated with each array, I will call it BigTime. And the look for the indexes you want to average and it is done. The big advantage is that you don't have to make a loop with the data in a list, and that you don't have to care about February changing size.

Index <- month(BigTime) < 10 & month(BigTime) > 3 # this make an index from april to septembre
Index <- Index & hour(BigTime) <= 19 & hour(BigTime) >= 8 # then you add the hour restriction
Mean <- apply(BigArray[,,Index],1:2,mean) 

And then it is done the mean for your values.

In both possibles a 2d array is build, if you want a 3d array with one dimension (time) having only one value chase add that dimension. And if you want to look for more information taking mean of specific time values is normally call composite technique in Meteorology Science.

I hope this solve your problem.