Invalid nodata value in gdal_rasterize when output is int64

[johanvdw]

What is the bug?

When converting a shape to a raster using gdal_rasterize, we notice that the wrong nodata value is used if the datatype is int64. The issue did not exist in gdal 3.2.2, and is present in 3.6 ->current master It seems no conincidence that this arrived after the support for int64 as a datatype

Steps to reproduce the issue

The command below returns incorrect nodata value (-9999) when inspecting the file

gdal_rasterize -q -a_nodata -9999.0 -at -a NR -of GTiff -te 250000.0 161520.0 250600.0 161840.0 -tr 20.0 20.0 gdal-testfile.shp output64.tif

The nodata value used inside the grid is 0 instead of -9999 . We expect only one value (1438) in the statistics.

 gdalinfo output64.tif -stats
....
Band 1 Block=30x16 Type=Int64, ColorInterp=Gray
  Minimum=0.000, Maximum=1438.000, Mean=257.642, StdDev=551.461
  NoData Value=-9999
  Metadata:
    STATISTICS_MINIMUM=0
    STATISTICS_MAXIMUM=1438
    STATISTICS_MEAN=257.64166666667
    STATISTICS_STDDEV=551.46123006417
    STATISTICS_VALID_PERCENT=100

Forcing output to be Int32 fixes this:

gdal_rasterize -q -a_nodata -9999.0 -at -a NR -of GTiff -te 250000.0 161520.0 250600.0 161840.0 -tr 20.0 20.0 gdal-testfile.shp output32.tif -ot Int32

[testcase-gdal.zip](https://github.com/user-attachments/files/15984721/testcase-gdal.zip)
gdalinfo output32.tif -stats
Band 1 Block=30x16 Type=Int32, ColorInterp=Gray
  Minimum=1438.000, Maximum=1438.000, Mean=1438.000, StdDev=0.000
  NoData Value=-9999
  Metadata:
    STATISTICS_MINIMUM=1438
    STATISTICS_MAXIMUM=1438
    STATISTICS_MEAN=1438
    STATISTICS_STDDEV=0
    STATISTICS_VALID_PERCENT=17.92

Versions and provenance

GDAL 3.10.0dev-de5a7f8c70, released 2024/06/25 (debug build)

Additional context

No response

[johanvdw]

Adding testcase files testcase-gdal.zip

[jratike80]

By the documentation https://gdal.org/programs/gdal_rasterize.html the default output type is Float64, not Int64. Forcing the utility to use the default yields correct result. So the issue seems to be limited to Int64 output, and the biggest bug is probably the changed default datatype.

gdal_rasterize -q -a_nodata -9999.0 -at -a NR -of GTiff -te 250000.0 161520.0 250600.0 161840.0 -tr 20.0 20.0 gdal-testfile.shp output64f.tif -ot Float64

[johanvdw]

I have checked and indeed when running in gdal 3.2.2 (debian bullseye, which was our reference) without any specification we get Type=Float64.