UserWarning The nodata value has been automatically changed

[aldotapia]

Hi,

I have a function for raster mosaicking and normalized difference index computation:

def mosaic_nd_raster(raster_list,layer1, layer2):
    '''function to compute normalized difference and mosaic  files with rioxarray library'''
    raster_single = []

    for raster in raster_list:
        with rioxr.open_rasterio(raster, masked = True) as src:
            lyr1 = getattr(src,layer1)
            lyr2 = getattr(src,layer2)
            raster_single.append((lyr1-lyr2)/(lyr1+lyr2))

    raster_mosaic = merge_arrays(raster_single)
    raster_mosaic = raster_mosaic.where((raster_mosaic <= 1) & (raster_mosaic >= -1))
    raster_mosaic = raster_mosaic.where(raster_mosaic != raster_mosaic.rio.nodata)
    raster_mosaic.rio.write_nodata(raster_mosaic.rio.nodata, encoded=True, inplace=True)
    return raster_mosaic

Then I write the file to my computer:

mos = mosaic_nd_raster(selected_files,'250m 16 days NIR reflectance', '250m 16 days MIR reflectance')
mos = mos * 1000 # for scaling the values
temporal_raster = 'some/path/name.tif'
mos.rio.to_raster(temporal_raster, compress='LZW')

Although the function works, the following message raises:

path/to/ee/lib/python3.10/site-packages/rioxarray/raster_writer.py:110: UserWarning: The nodata value (3.402823466e+38) has been automatically changed to (3.4028234663852886e+38) to match the dtype of the data.
  warnings.warn(

I included raster_mosaic.rio.write_nodata for updating the masked value, but I still get the warning. I can even difference between 3.402823466e+38 and 3.4028234663852886e+38.

What am I missing? I'm using MODIS hdf scenes with this function.

Also, If I save the values to integer datatype, I lose the NoData pixels resulting in 0 instead if NaN

Thanks in advance

[snowman2]

The warning is informing you that when changing the nodata value to the dtype of you data, it is not equal to the original value:

rioxarray/rioxarray/raster_writer.py

Lines 109 to 113 in 570150f

	if not numpy.isnan(nodata) and original_nodata != nodata:
	warnings.warn(
	f"The nodata value ({original_nodata}) has been automatically "
	f"changed to ({nodata}) to match the dtype of the data."
	)

This is likely due to loss of precision. I recommend selecting a nodata value appropriate for the datatype you are using.

[aldotapia]

Excellent, I modified the code and there is no warning message now:

def mosaic_nd_raster(raster_list,layer1, layer2):
    '''function to compute normalized difference and mosaic files with rioxarray library'''
    raster_single = []

    for raster in raster_list:
        with rioxr.open_rasterio(raster, masked = True) as src:
            lyr1 = getattr(src,layer1)
            lyr2 = getattr(src,layer2)
            nd = 1000 * (lyr1 - lyr2) / (lyr1 + lyr2)
            nd.rio.set_nodata(-32768)
            raster_single.append(nd)
    raster_mosaic = merge_arrays(raster_single)
    raster_mosaic = raster_mosaic.where((raster_mosaic <= 1000) & (raster_mosaic >= -1000))
    raster_mosaic = raster_mosaic.where(raster_mosaic != raster_mosaic.rio.nodata)
    return raster_mosaic

Thank you!