Convert timeseries stack of GTiff raster to single NetCDF

Question

Moving from gdal-dev mailing list:

On Mon, Sep 2, 2013 at 7:09 PM, David Shean wrote:

Hi list, I'm trying to package a timeseries of GTiff rasters with identical projection/extent/resolution as a single NetCDF file for distribution. I've spent the past hour consulting the online doc and playing with gdal_translate, gdalbuildvrt and gdalwarp without any success.

Is there an easy way to do this using existing gdal command line utilities? I figured I'd ask before resorting to a custom solution using the NetCDF Python API.

Thanks. -David

On Tue, Sep 3, 2013 at 10:15 AM, Etienne Tourigny wrote:

what you want is probably outside the scope of gdal. It would require some clever metadata management so that gdal_translate puts them in a single file...

I would advise you convert them all to netcdf using gdal_translate and then use python-netcdf4 (not the one from numpy/scipy) to stack them in the temporal dimension.

On Tue, Sep 3, 2013, at 7:55 AM, "Signell, Richard" wrote:

David, If you post your question on the GIS stackexchange group https://gis.stackexchange.com/ I will provide an example code that should be helpful.

-Rich

====================

Update 9/3/13 17:04 PDT

Here is gdalinfo output for one of my input datasets:


gdalinfo 20120901T2024_align_x+22.19_y+3.68_z+14.97_warp.tif

Driver: GTiff/GeoTIFF
Files: 20120901T2024_align_x+22.19_y+3.68_z+14.97_warp.tif
Size is 10666, 13387
Coordinate System is:
PROJCS["unnamed",
    GEOGCS["WGS 84",
        DATUM["WGS_1984",
            SPHEROID["WGS 84",6378137,298.257223563,
                AUTHORITY["EPSG","7030"]],
            AUTHORITY["EPSG","6326"]],
        PRIMEM["Greenwich",0],
        UNIT["degree",0.0174532925199433],
        AUTHORITY["EPSG","4326"]],
    PROJECTION["Polar_Stereographic"],
    PARAMETER["latitude_of_origin",70],
    PARAMETER["central_meridian",-45],
    PARAMETER["scale_factor",1],
    PARAMETER["false_easting",0],
    PARAMETER["false_northing",0],
    UNIT["metre",1,
        AUTHORITY["EPSG","9001"]]]
Origin = (-211346.063781524338992,-2245136.291794800199568)
Pixel Size = (5.000000000000000,-5.000000000000000)
Metadata:
  AREA_OR_POINT=Area
Image Structure Metadata:
  COMPRESSION=LZW
  INTERLEAVE=BAND
Corner Coordinates:
Upper Left  ( -211346.064,-2245136.292) ( 50d22'39.70"W, 69d23'55.59"N)
Lower Left  ( -211346.064,-2312071.292) ( 50d13'22.38"W, 68d48'10.75"N)
Upper Right ( -158016.064,-2245136.292) ( 49d 1'33.33"W, 69d26'16.42"N)
Lower Right ( -158016.064,-2312071.292) ( 48d54'35.06"W, 68d50'27.28"N)
Center      ( -184681.064,-2278603.792) ( 49d38' 1.32"W, 69d 7'17.04"N)
Band 1 Block=256x256 Type=Float32, ColorInterp=Gray
  NoData Value=-32767

Following up on Luke's suggested approach.

The vrt generation works fine:

gdalbuildvrt -separate newtest.vrt *warp.tif

<VRTDataset rasterXSize="10666" rasterYSize="13387">
  <SRS>PROJCS["unnamed",GEOGCS["WGS 84",DATUM["WGS_1984",SPHEROID["WGS 84",6378137,298.257223563,AUTHORITY["EPSG","7030"]],AUTHORITY["EPSG","6326"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433],AUTHORITY["EPSG","4326"]],PROJECTION["Polar_Stereographic"],PARAMETER["latitude_of_origin",70],PARAMETER["central_meridian",-45],PARAMETER["scale_factor",1],PARAMETER["false_easting",0],PARAMETER["false_northing",0],UNIT["metre",1,AUTHORITY["EPSG","9001"]]]</SRS>
  <GeoTransform> -2.1134606378152434e+05,  5.0000000000000000e+00,  0.0000000000000000e+00, -2.2451362917948002e+06,  0.0000000000000000e+00, -5.0000000000000000e+00</GeoTransform>
  <VRTRasterBand dataType="Float32" band="1">
    <NoDataValue>-3.27670000000000E+04</NoDataValue>
    <ComplexSource>
      <SourceFilename relativeToVRT="1">20110619T2024_align_x+15.51_y+1.15_z+12.10_warp.tif</SourceFilename>
      <SourceBand>1</SourceBand>
      <SourceProperties RasterXSize="10666" RasterYSize="13387" DataType="Float32" BlockXSize="256" BlockYSize="256" />
      <SrcRect xOff="0" yOff="0" xSize="10666" ySize="13387" />
      <DstRect xOff="0" yOff="0" xSize="10666" ySize="13387" />
      <NODATA>-32767</NODATA>
    </ComplexSource>
  </VRTRasterBand>
  <VRTRasterBand dataType="Float32" band="2">
    <NoDataValue>-3.27670000000000E+04</NoDataValue>
    <ComplexSource>
      <SourceFilename relativeToVRT="1">20110802T2024_align_x+16.33_y+2.14_z+12.02_warp.tif</SourceFilename>
      <SourceBand>1</SourceBand>
      <SourceProperties RasterXSize="10666" RasterYSize="13387" DataType="Float32" BlockXSize="256" BlockYSize="256" />
      <SrcRect xOff="0" yOff="0" xSize="10666" ySize="13387" />
      <DstRect xOff="0" yOff="0" xSize="10666" ySize="13387" />
      <NODATA>-32767</NODATA>
    </ComplexSource>
  </VRTRasterBand>
...

But when I attempt to translate to nc, I get the following error:


gdal_translate -of netcdf newtest.vrt newtest.nc

Input file size is 10666, 13387
Warning 1: Variable has 0 dimension(s) - not supported.
0...10...20...30...40...50ERROR 1: netcdf error #-62 : NetCDF: One or more variable sizes violate format constraints .
at (netcdfdataset.cpp,SetDefineMode,1574)

ERROR 1: netcdf error #-39 : NetCDF: Operation not allowed in define mode .
at (netcdfdataset.cpp,IWriteBlock,1435)

ERROR 1: netCDF scanline write failed: NetCDF: Operation not allowed in define mode
ERROR 1: An error occured while writing a dirty block
...ERROR 1: netcdf error #-39 : NetCDF: Operation not allowed in define mode .
at (netcdfdataset.cpp,IWriteBlock,1435)

ERROR 1: netCDF scanline write failed: NetCDF: Operation not allowed in define mode
ERROR 1: netcdf error #-62 : NetCDF: One or more variable sizes violate format constraints .
at (netcdfdataset.cpp,~netCDFDataset,1548)

So upon closer inspection, it appears that gdal is unhappy with the polar stereographic projection I'm using (EPSG:3413). See lines 1570-1582 of netcdfdataset.cpp:

https://code.vpac.org/gitorious/gdal-netcdf-testing/gdal-netcdf-driver/blobs/8fa3582669969ad4d55e461f5846b3ed33727f63/gdal/frmts/netcdf/netcdfdataset.cpp

My projection has a latitude_of_origin specified but no standard parallels as expected by the netcdf driver.

What GDAL version? There were a number of changes in the NetCDF driver in GDAL>=1.9.0. That page specifically mentions changes to polar stereographic projection handling. You might be able to work around by overriding the projection with the gdal_translate -a_srs parameter and specifying a valid but equivalent projection string. See also (trac.osgeo.org/gdal/wiki/NetCDF_ProjectionTestingStatus) — user2856, Commented Sep 4, 2013 at 1:28
Thanks to both Rich and Luke for helpful input. I need to update to the latest GDAL release, evaluate the latest netcdf driver polar stereographic functionality, and follow up with gdal-dev on any lingering issues. While both answers will work, I like Rich's recipe, and will adopt for my own purposes. I know others will find this discussion useful - glad it's archived on SE. — David Shean, Commented Sep 4, 2013 at 20:27

Rich Signell · Accepted Answer · 2024-10-15 18:51:32Z

Here's some python code that does what you want, reading GDAL files that represent data at specific times and writing to a single NetCDF file that is CF-Compliant

#!/usr/bin/env python
'''
Convert a bunch of GDAL readable grids to a NetCDF Time Series.
Here we read a bunch of files that have names like:
/usgs/data0/prism/1890-1899/us_tmin_1895.01
/usgs/data0/prism/1890-1899/us_tmin_1895.02
...
/usgs/data0/prism/1890-1899/us_tmin_1895.12
'''

import numpy as np
import datetime as dt
import os
import gdal
import netCDF4
import re

ds = gdal.Open('/usgs/data0/prism/1890-1899/us_tmin_1895.01')
a = ds.ReadAsArray()
nlat,nlon = np.shape(a)

b = ds.GetGeoTransform() #bbox, interval
lon = np.arange(nlon)*b[1]+b[0]
lat = np.arange(nlat)*b[5]+b[3]


basedate = dt.datetime(1858,11,17,0,0,0)

# create NetCDF file
nco = netCDF4.Dataset('time_series.nc','w',clobber=True)

# chunking is optional, but can improve access a lot: 
# (see: http://www.unidata.ucar.edu/blogs/developer/entry/chunking_data_choosing_shapes)
chunk_lon=16
chunk_lat=16
chunk_time=12

# create dimensions, variables and attributes:
nco.createDimension('lon',nlon)
nco.createDimension('lat',nlat)
nco.createDimension('time',None)
timeo = nco.createVariable('time','f4',('time'))
timeo.units = 'days since 1858-11-17 00:00:00'
timeo.standard_name = 'time'

lono = nco.createVariable('lon','f4',('lon'))
lono.units = 'degrees_east'
lono.standard_name = 'longitude'

lato = nco.createVariable('lat','f4',('lat'))
lato.units = 'degrees_north'
lato.standard_name = 'latitude'

# create container variable for CRS: lon/lat WGS84 datum
crso = nco.createVariable('crs','i4')
csro.long_name = 'Lon/Lat Coords in WGS84'
crso.grid_mapping_name='latitude_longitude'
crso.longitude_of_prime_meridian = 0.0
crso.semi_major_axis = 6378137.0
crso.inverse_flattening = 298.257223563

# create short integer variable for temperature data, with chunking
tmno = nco.createVariable('tmn', 'i2',  ('time', 'lat', 'lon'), 
   zlib=True,chunksizes=[chunk_time,chunk_lat,chunk_lon],fill_value=-9999)
tmno.units = 'degC'
tmno.scale_factor = 0.01
tmno.add_offset = 0.00
tmno.long_name = 'minimum monthly temperature'
tmno.standard_name = 'air_temperature'
tmno.grid_mapping = 'crs'
tmno.set_auto_maskandscale(False)

nco.Conventions='CF-1.6'

#write lon,lat
lono[:]=lon
lato[:]=lat

pat = re.compile('us_tmin_[0-9]{4}\.[0-9]{2}')
itime=0

#step through data, writing time and data to NetCDF
for root, dirs, files in os.walk('/usgs/data0/prism/1890-1899/'):
    dirs.sort()
    files.sort()
    for f in files:
        if re.match(pat,f):
            # read the time values by parsing the filename
            year=int(f[8:12])
            mon=int(f[13:15])
            date=dt.datetime(year,mon,1,0,0,0)
            print(date)
            dtime=(date-basedate).total_seconds()/86400.
            timeo[itime]=dtime
           # min temp
            tmn_path = os.path.join(root,f)
            print(tmn_path)
            tmn=gdal.Open(tmn_path)
            a=tmn.ReadAsArray()  #data
            tmno[itime,:,:]=a
            itime=itime+1

nco.close()

GDAL and NetCDF4 Python can be a bit of a pain to build, but the good news is that they are part of most scientific python distributions (Python(x,y), Enthought Python Distribution, Anaconda, ...)

Update: I haven't done polar stereographic in CF-compliant NetCDF yet, but I should look something like this. Here I've assumed that central_meridian and latitude_of_origin in GDAL are the same as straight_vertical_longitude_from_pole and latitude_of_projection_origin in CF:

#!/usr/bin/env python
'''
Convert a bunch of GDAL readable grids to a NetCDF Time Series.
Here we read a bunch of files that have names like:
/usgs/data0/prism/1890-1899/us_tmin_1895.01
/usgs/data0/prism/1890-1899/us_tmin_1895.02
...
/usgs/data0/prism/1890-1899/us_tmin_1895.12
'''

import numpy as np
import datetime as dt
import os
import gdal
import netCDF4
import re

ds = gdal.Open('/usgs/data0/prism/1890-1899/us_tmin_1895.01')
a = ds.ReadAsArray()
ny,nx = np.shape(a)

b = ds.GetGeoTransform() #bbox, interval
x = np.arange(nx)*b[1]+b[0]
y = np.arange(ny)*b[5]+b[3]


basedate = dt.datetime(1858,11,17,0,0,0)

# create NetCDF file
nco = netCDF4.Dataset('time_series.nc','w',clobber=True)

# chunking is optional, but can improve access a lot: 
# (see: http://www.unidata.ucar.edu/blogs/developer/entry/chunking_data_choosing_shapes)
chunk_x=16
chunk_y=16
chunk_time=12

# create dimensions, variables and attributes:
nco.createDimension('x',nx)
nco.createDimension('y',ny)
nco.createDimension('time',None)
timeo = nco.createVariable('time','f4',('time'))
timeo.units = 'days since 1858-11-17 00:00:00'
timeo.standard_name = 'time'

xo = nco.createVariable('x','f4',('x'))
xo.units = 'm'
xo.standard_name = 'projection_x_coordinate'

yo = nco.createVariable('y','f4',('y'))
yo.units = 'm'
yo.standard_name = 'projection_y_coordinate'

# create container variable for CRS: x/y WGS84 datum
crso = nco.createVariable('crs','i4')
crso.grid_mapping_name='polar_stereographic'
crso.straight_vertical_longitude_from_pole = -45.
crso.latitude_of_projection_origin = 70.
crso.scale_factor_at_projection_origin = 1.0
crso.false_easting = 0.0
crso.false_northing = 0.0
crso.semi_major_axis = 6378137.0
crso.inverse_flattening = 298.257223563

# create short integer variable for temperature data, with chunking
tmno = nco.createVariable('tmn', 'i2',  ('time', 'y', 'x'), 
   zlib=True,chunksizes=[chunk_time,chunk_y,chunk_x],fill_value=-9999)
tmno.units = 'degC'
tmno.scale_factor = 0.01
tmno.add_offset = 0.00
tmno.long_name = 'minimum monthly temperature'
tmno.standard_name = 'air_temperature'
tmno.grid_mapping = 'crs'
tmno.set_auto_maskandscale(False)

nco.Conventions='CF-1.6'

#write x,y
xo[:]=x
yo[:]=y

pat = re.compile('us_tmin_[0-9]{4}\.[0-9]{2}')
itime=0

#step through data, writing time and data to NetCDF
for root, dirs, files in os.walk('/usgs/data0/prism/1890-1899/'):
    dirs.sort()
    files.sort()
    for f in files:
        if re.match(pat,f):
            # read the time values by parsing the filename
            year=int(f[8:12])
            mon=int(f[13:15])
            date=dt.datetime(year,mon,1,0,0,0)
            print(date)
            dtime=(date-basedate).total_seconds()/86400.
            timeo[itime]=dtime
           # min temp
            tmn_path = os.path.join(root,f)
            print(tmn_path)
            tmn=gdal.Open(tmn_path)
            a=tmn.ReadAsArray()  #data
            tmno[itime,:,:]=a
            itime=itime+1

nco.close()

Great code Rich! This is very useful, and I will use this in the future. Looks like your input projection is assumed to be geographic w/ units of lat/lon (EPSG:4326). I'm working with high-resolution data at polar latitudes, so this is not ideal, but I will try converting to WGS84. — David Shean, Commented Sep 4, 2013 at 1:00
lat/lon was just an example. You can use whatever you wish. What application(s) are you targeting? ArcGIS, just for archiving or what? — Rich Signell, Commented Sep 4, 2013 at 1:27
Well, I have many timeseries like this, and I'm evaluating options for efficient storage and analysis. But at the moment, I'm packaging data for ingestion by flow models. The modeling community, at least ice flow modeling, seems to like netcdf. — David Shean, Commented Sep 4, 2013 at 2:47
CF compliant link is now a 404, should it be replaced with cfconventions.org or a specific versioned metadata convention? — A Lee, Commented Jul 30, 2023 at 0:58

user2856 · Accepted Answer · 2013-09-03 23:55:04Z

It's easy to put them in a single NetCDF with GDAL utilities, example below. But you don't get the temporal dimension/other metadata of @RichSignell's answer. The tiffs just get dumped into subdatasets.

C:\remotesensing\testdata>dir /b ndvi*.tif
ndvi1.tif
ndvi2.tif
ndvi3.tif

C:\remotesensing\testdata>gdalbuildvrt -separate ndvi.vrt ndvi*.tif
0...10...20...30...40...50...60...70...80...90...100 - done.

C:\remotesensing\testdata>gdal_translate -of netcdf ndvi.vrt ndvi.nc
Input file size is 96, 88
0...10...20...30...40...50...60...70...80...90...100 - done.

C:\remotesensing\testdata>gdalinfo ndvi.nc
Driver: netCDF/Network Common Data Format
Files: ndvi.nc
Size is 512, 512
Coordinate System is `'
Metadata:
  NC_GLOBAL#Conventions=CF-1.5
  NC_GLOBAL#GDAL=GDAL 1.10.0, released 2013/04/24
  NC_GLOBAL#history=Wed Sep 04 09:49:11 2013: GDAL CreateCopy( ndvi.nc, ... )
Subdatasets:
  SUBDATASET_1_NAME=NETCDF:"ndvi.nc":Band1
  SUBDATASET_1_DESC=[88x96] Band1 (32-bit floating-point)
  SUBDATASET_2_NAME=NETCDF:"ndvi.nc":Band2
  SUBDATASET_2_DESC=[88x96] Band2 (32-bit floating-point)
  SUBDATASET_3_NAME=NETCDF:"ndvi.nc":Band3
  SUBDATASET_3_DESC=[88x96] Band3 (32-bit floating-point)
Corner Coordinates:
Upper Left  (    0.0,    0.0)
Lower Left  (    0.0,  512.0)
Upper Right (  512.0,    0.0)
Lower Right (  512.0,  512.0)
Center      (  256.0,  256.0)

C:\remotesensing\testdata>gdalinfo NETCDF:"ndvi.nc":Band1
Driver: netCDF/Network Common Data Format
Files: ndvi.nc
Size is 96, 88
Coordinate System is:
GEOGCS["GCS_GDA_1994",
    DATUM["Geocentric_Datum_of_Australia_1994",
        SPHEROID["GRS 1980",6378137,298.2572221010002,
            AUTHORITY["EPSG","7019"]],
        AUTHORITY["EPSG","6283"]],
    PRIMEM["Greenwich",0],
    UNIT["degree",0.0174532925199433]]
Origin = (115.810500000000000,-32.260249999999999)
Pixel Size = (0.000250000000000,-0.000250000000000)
Metadata:
  Band1#_FillValue=0
  Band1#grid_mapping=crs
  Band1#long_name=GDAL Band Number 1
  crs#GeoTransform=115.8105 0.00025 0 -32.26025 0 -0.00025
  crs#grid_mapping_name=latitude_longitude
  crs#inverse_flattening=298.2572221010002
  crs#longitude_of_prime_meridian=0
  crs#semi_major_axis=6378137
  crs#spatial_ref=GEOGCS["GCS_GDA_1994",DATUM["Geocentric_Datum_of_Australia_1994",SPHEROID["GRS 1980",6378137,298.2572221010002,AUTHORITY["EPSG","7019"]],AUTHORITY["EPSG","6283"]],PRIMEM["Greenwich",0],UNIT["degree",0.0174532925199433]]
  lat#long_name=latitude
  lat#standard_name=latitude
  lat#units=degrees_north
  lon#long_name=longitude
  lon#standard_name=longitude
  lon#units=degrees_east
  NC_GLOBAL#Conventions=CF-1.5
  NC_GLOBAL#GDAL=GDAL 1.10.0, released 2013/04/24
  NC_GLOBAL#history=Wed Sep 04 09:49:11 2013: GDAL CreateCopy( ndvi.nc, ... )
Corner Coordinates:
Upper Left  ( 115.8105000, -32.2602500) (115d48'37.80"E, 32d15'36.90"S)
Lower Left  ( 115.8105000, -32.2822500) (115d48'37.80"E, 32d16'56.10"S)
Upper Right ( 115.8345000, -32.2602500) (115d50' 4.20"E, 32d15'36.90"S)
Lower Right ( 115.8345000, -32.2822500) (115d50' 4.20"E, 32d16'56.10"S)
Center      ( 115.8225000, -32.2712500) (115d49'21.00"E, 32d16'16.50"S)
Band 1 Block=96x1 Type=Float32, ColorInterp=Undefined
  NoData Value=0
  Metadata:
    _FillValue=0
    grid_mapping=crs
    long_name=GDAL Band Number 1
    NETCDF_VARNAME=Band1

I tried this approach and it failed for my input data - I will post output above. — David Shean, Commented Sep 4, 2013 at 0:02
As a test, I used gdalwarp to reproject the EPSG:3413 multi-band vrt to EPSG:4326, then used gdal_translate to convert to netcdf4. As Luke suggests, this works without issue. As Etienne suggested in the original gdal-dev thread, there is limited control over metadata for this approach. — David Shean, Commented Sep 4, 2013 at 20:11
Nice and simple approach. Then, I usually change variable names and add attributes in NC file e.g. with NCO command line utility. — jurajb, Commented Mar 20 at 9:28

Bằng Rikimaru · Accepted Answer · 2020-11-16 13:37:24Z

I have a simple solution by using gdalwarp and cdo:

First warp the netCDF file to a tiff file (note: the time slices are translated to bands in 2D tiff file)

gdalwarp /home/rasdaman/Downloads/TT_201001_daymean.nc -t_srs EPSG:4326 -of netCDF warp.nc

Then, uses cdo tool https://code.mpimet.mpg.de/projects/cdo/wiki/Tutorial#Interpolation

with the provided information from gdalinfo to create a gridfile.txt

e.g: 

gridtype  = lonlat
gridsize  = 765320
xsize     = 1007
ysize     = 760
xname     = lon
xlongname = "longitude" 
xunits    = "degrees_east" 
yname     = lat
ylongname = "latitude" 
yunits    = "degrees_north" 
xfirst    = 0.011638990053852
xinc      = 0.033333333
yfirst    = 46.6918065
yinc      = 0.011623675593254

then run cdo to create a new netcdf file with projected XY domains

cdo -remapbil,gridfile.txt TT_201001_daymean.nc cdo1.nc

Stack Exchange Network

Convert timeseries stack of GTiff raster to single NetCDF

3 Answers 3

Your Answer

Not the answer you're looking for? Browse other questions tagged
python
gdal
netcdf
or ask your own question.

Linked

Hot Network Questions

Convert timeseries stack of GTiff raster to single NetCDF

3 Answers 3

Your Answer

Sign up or log in

Post as a guest

Not the answer you're looking for? Browse other questions tagged pythongdalnetcdf or ask your own question.

Linked

Related

Hot Network Questions

Not the answer you're looking for? Browse other questions tagged
python
gdal
netcdf
or ask your own question.