How to not overwrite with NoData values when updating a raster file using a numpy array?

Question

I have a numpy array (called out_window) with the following values:

out_window:

[[[499 498 481]
  [494 490 464]
  [483 480 -1]
  [482 -1 -1]]]

This numpy array was obtained by clipping a raster file using a polygon feature and altering only those elements with value different to -1 (-1 is NoData)

I'm trying to update that same raster file with those new values.

import rasterio
import geopandas

raster_in = r'C:\.......\raster_in_file.tif'
raster_out = r'C:\.......\raster_out_file.tif'
copy(raster_in, raster_out)
polyg = gpd.read_file(gdb_vml_polygs, driver='FileGDB', layer=fc_vml_polygs)

out_window, out_transform = mask(raster_in, polyg['geometry'], all_touched=True, crop=True)

out_meta = raster_in.meta.copy()                        

#this is not relevant
width = ...
height = ...
x = ...
y = ...

with rasterio.open(raster_out, "r+", **out_meta) as dest:
    dest.write(out_window, window=Window(x, y, width, height))
dest.close()

The problem is that the instruction dest.write(.....) is overwriting the raster with all 12 values of the out_window array. I do not want to overwrite the raster with the -1 values. Those 3 elements of the raster should be left unaltered.

Is there any way to do this?

Answer 1

Issue was overcome by adding a line to read original values using window with same dimensions and same top left cell position than array produced by clipping with polygon.

#Reading tif file
raster = rasterio.open(raster_path)

#copy of raster metadata
out_meta = raster.meta.copy()  

#clipping raster using a polygon
out_window, out_transform = mask(raster, polygon, all_touched=True, crop=True)

#Using 'out_meta' 'out_window' and 'out_transform' we can obtain window dimensions
#and position of top left cell of the window => x, y, width, height

#Reading original values using same size window as array produced by clipping with polygon
window_orig = raster.read(1, window=Window(x, y, width, height))

#Swapping values different to NoData with their average value.
#If NoData then use original values
window_avg = (np.where(out_window!=-1, int(round(avg,0)), window_orig))

#Writing back to raster
with rasterio.open(raster_path, "r+", **out_meta) as dest:
     dest.write(window_avg, window=Window(x, y, width, height))