Nan in NumPy array to NoData in ArcGIS raster

For smaller rasters, NumPy is a much faster alternative to some of the ArcGIS geoprocessing functions. One issue with doing this is that every raster must have exactly the same dimensions (rows, columns, cellsize, extent, etc.) since NumPy has no spatial component. Later on I will post about some of the techniques we use to make sure they are all identical.

There is one issue with converting NumPy arrays back to rasters though that I want to outline. The arcpy function NumPyArrayToRaster does not correctly convert NaN values. The values will display as NoData in the identity window, but they do not act as NoData in subsequent calculations and the 'Display NoData as' option in the Symbology tab has no impact.

To get around this, convert all NaN values to some unused value, and then set that value as the value_to_nodata parameter. For example:

test_array[np.isnan(test_array)] = -9999
test_obj = arcpy.NumPyArrayToRaster(test_array, pnt, cellsize, cellsize, -9999)
test_obj.save(test_raster)

This is Bug NIM063028 and is not fixed in any current or future version.