How to exclude NoData values when calculating the flow direction of DEM?

[mht2953658596]

When using PyShed to calculate the flow direction of DEM, I found that the flow direction was also calculated for areas where DEM values were NoData. How can I avoid this situation?

When I set NoData using the following code:

nodata = gdal.Open(raster_path)tGetRasterBand(1).ReadAsArray().GetNoDataValue()
grid = Grid.from_raster(raster_path, nodata=nodata)
dem = grid.read_raster(self.raster_path, band=1, nodata=grid.nodata)
flooded_dem = grid.fill_depressions(dem)
inflated_dem = grid.resolve_flats(flooded_dem)
fdir = grid.flowdir(inflated_dem)

it reports the following error:

[mht2953658596]

If I don't set the NoData value, it will report the same error.

[mht2953658596]

@cheginit

[phillipaarestrup]

I have found the same issue and also when loading some rasters. I think some of this is related to a new update in the Numpy package from 1.2x -> 2.0.0 (released in june). I only have these issues with the new version of Numpy and not when I downgrade it to 1.2x.

Your screenshow shows that you get a typeerror in sview.py which is a result of this function call np.can_cast(viewfinder.nodata, obj.dtype, casting='safe') in line 85

See also the discussion here on fx the function can_cast, which is used for checks in the pysheds package: numpy/numpy#24856

[mht2953658596]

@phillipaarestrup But my NumPy version is 1.24.3, and I am still encountering this issue.

[JaweedNazary]

I tried NumPy 2.0 and the nodata representation error pops up when running flodir. However, with NumPy 1.26, it is working pretty good,

[JaweedNazary]

@mht2953658596 try converting your raster to float before using the code.

[col16]

Here's a workaround: you can pass the parameter nodata_out=np.int64(0):

fdir = grid.flowdir(inflated_dem, nodata_out=np.int64(0))

The same issue affects the accumulation function, and the workaround is the same:
acc = grid.accumulation(fdir, nodata_out=np.int64(0))

[hydro-coop]

@col16 thank you for providing a workaround. I used it for both the flow direction and accumulation functions and was finally able to avoid the 'nodata' TypeError.

Continuing with the workflow, I still received the error message when trying to run the snap_to_mask function. Is there a workaround for this function or for the catchment function (unsure yet if that will raise the same issue). I tried the same as above but was unsuccessful.

Thank you in advance.

[hydro-coop]

Update: I found that downgrading to Numpy 1.26 allowed me to complete the delineation without using any provided workarounds.