Error with ncdf files

[ramiromagno]

Hi Edzer!,

Thanks for the stars package!

I am struggling with reading an NC file. Could you take a look, please? I tried my best to sort the issue myself...

Download the file (1.1G) from Google Drive: https://drive.google.com/file/d/1GGt68l6HOo5Q4aGuX4ZLIRTRXgZSDKs3/view?usp=sharing.

With:

nc_file <- "chl_Omon_MPI-ESM1-2-HR_ssp245_r1i1p1f1_gn_201501-201912.nc"
nc <- stars::read_ncdf(.x = nc_file)
stars::st_as_stars(nc)

I get this error:

Error in seq.default(dims$start[ic], length = dims$count[ic]) : 
  'from' must be a finite number

I managed to narrowed it down to the fact that dims gets an NA in start at this location:

stars/R/ncdf.R

Line 165 in 76df605

dims <- .update_dims(dims, proxy_dimensions, coords, tdim)

# A tibble: 4 × 9
     id name  length unlim coord_dim coord_var axis  start count
  <int> <chr>  <dbl> <lgl> <lgl>     <chr>     <chr> <dbl> <dbl>
1     3 i        802 FALSE TRUE      ""        ""        2   802
2     2 j        404 FALSE TRUE      ""        ""        2   404
3     1 lev       40 FALSE TRUE      "lev"     "Z"      NA    40
4     0 time      60 TRUE  TRUE      "time"    "T"       1    60

[edzer]

I can reproduce that, but managed to read the file using

> nc <- stars::read_stars(.x = nc_file)
There were 22 warnings (use warnings() to see them)
> nc
stars object with 4 dimensions and 1 attribute
attribute(s), summary of first 1e+05 cells:
                                                Min.      1st Qu.      Median
Mass Concentration of Total... [kg/m^3] 1.837166e-07 7.611779e-07 1.62674e-06
                                                Mean      3rd Qu.        Max.
Mass Concentration of Total... [kg/m^3] 3.080384e-06 4.044461e-06 3.34308e-05
                                         NA's
Mass Concentration of Total... [kg/m^3] 18119
dimension(s):
     from  to  refsys                                      values x/y
x       1 802      NA          [802x404] 0.005916 [°],...,360 [°] [x]
y       1 404      NA          [802x404] -78.72 [°],...,89.76 [°] [y]
lev     1  40 udunits              [6,17) [m],...,[5720,6270) [m]    
time    1  60 POSIXct 2015-01-16 12:00:00,...,2019-12-16 12:00:00    
curvilinear grid

and

> x = read_mdim(nc_file)
> x
stars object with 4 dimensions and 1 attribute
attribute(s), summary of first 1e+05 cells:
                     Min.      1st Qu.       Median         Mean      3rd Qu.
chl [kg/m^3] 6.839074e-10 2.858014e-09 1.985313e-07 2.226392e-07 3.982834e-07
                     Max.  NA's
chl [kg/m^3] 4.758287e-06 65786
dimension(s):
     from  to   offset delta  refsys                    values x/y
i       1 802 -0.5 [1] 1 [1] udunits                      NULL [x]
j       1 404 -0.5 [1] 1 [1] udunits                      NULL [y]
lev     1  40       NA    NA      NA    [0,12),...,[5420,6020)    
time    1  60       NA    NA    Date 2015-01-16,...,2019-12-16    

The difference in summary statistics of the first 1e5 cells is, I guess, caused by a different dimension ordering of the two interfaces. As the lon/lat "cells" are defined by 4 corners, I'm not sure they are read in sensibly by read_stars. Another worry is the difference in lev intervals from both interfaces.