Conditional replacement e.g. where(condition, true, false)

openEO Platform
1

Hi,

I have two datacubes and I want to evaluate where both are valid (not masked), if the condition is true I want to return one value, if false another.

What is the best way to do conditional replacement?

2

Hi @dadr,

I’ve created an example for your use case.
In the example I compute the median NDVI for two months, then I check where the NDVI is greater than a value for both months and at the end I replace 0 and 1 with other values.

import openeo
from openeo.processes import eq, not_
import xarray as xr

openeoHost = 'https://openeo.cloud'
conn = openeo.connect(openeoHost).authenticate_oidc('egi')

im = conn.load_collection(
        'SENTINEL2_L2A_SENTINELHUB',
        spatial_extent={'west': 11.304588, 'east': 11.377716, 'south': 46.465311, 'north': 46.516839},
        temporal_extent=['2021-06-01', '2021-08-01'],
        bands = ['B04','B08'],
        properties={
            'eo:cloud_cover': lambda x: lte(x, 50)
        })
median_ndvi_june = im.filter_temporal(['2021-06-01', '2021-07-01']).ndvi().reduce_dimension(dimension='t',reducer='median')
median_ndvi_july = im.filter_temporal(['2021-07-01', '2021-08-01']).ndvi().reduce_dimension(dimension='t',reducer='median')

# Check weather the median NDVI is above 0.8 in both months. If yes set the area to 100, else to 50

value_0 = 100
value_1 = 50

# Firstly create a boolean mask for each month
median_ndvi_june_mask = median_ndvi_june > 0.8
median_ndvi_july_mask = median_ndvi_july > 0.8

# Merge the masks with a boolean AND operation
mask_june_and_july_gt_0_8 = median_ndvi_june_mask.merge_cubes(median_ndvi_july_mask,overlap_resolver='and')

# Create the inverse of it
mask_june_and_july_lte_0_8 = mask_june_and_july_gt_0_8.apply(lambda x: x.eq(0))

# Use the mask operator twice (with the mask and the inverse of it) to set two different values to replace 0 and 1
ndvi_jun_jul_gt_0_8 = mask_june_and_july.mask(mask_june_and_july_gt_0_8,replacement=value_0).mask(mask_june_and_july_lte_0_8,replacement=value_1)

# Alternative approach using multiply + mask
# ndvi_jun_jul_gt_0_8 = (mask_june_and_july_gt_0_8 * value_0).mask(mask_june_and_july_lte_0_8,replacement=value_1)


ndvi_jun_jul_gt_0_8.download('ndvi_jun_jul_gt_0_8.nc')

result:

image

3

Thanks, looks close to what I want. However, what if i want the replacement value to be a datacube rather than a constant? In the documentation it doesn’t look like that’s possible.

4

You can do that, but not only with mask, you would need to combine it with merge_cubes.
Look at the modified example:

import openeo
from openeo.processes import eq, not_
import xarray as xr

openeoHost = 'https://openeo.cloud'
conn = openeo.connect(openeoHost).authenticate_oidc('egi')

im = conn.load_collection(
        'SENTINEL2_L2A_SENTINELHUB',
        spatial_extent={'west': 11.304588, 'east': 11.377716, 'south': 46.465311, 'north': 46.516839},
        temporal_extent=['2021-06-01', '2021-08-01'],
        bands = ['B04','B08'],
        properties={
            'eo:cloud_cover': lambda x: lte(x, 50)
        })
median_ndvi_june = im.filter_temporal(['2021-06-01', '2021-07-01']).ndvi().reduce_dimension(dimension='t',reducer='median')
median_ndvi_july = im.filter_temporal(['2021-07-01', '2021-08-01']).ndvi().reduce_dimension(dimension='t',reducer='median')

# Replace values of june NDVI with july NDVI when NDVI in June is > 0.8

# Firstly create a boolean mask for each month
median_ndvi_june_mask = median_ndvi_june > 0.8
median_ndvi_june_mask_inverse = median_ndvi_june <= 0.8

# Set to zero the values in June where NDVI < 0.8
median_ndvi_june_masked = median_ndvi_june.mask(median_ndvi_june_mask,replacement=0)

# Set to zero the values in July where June NDVI <= 0.8
median_ndvi_july_masked = median_ndvi_july.mask(median_ndvi_june_mask_inverse,replacement=0)

# Merge the results
ndvi_june_and_july = median_ndvi_june_masked.merge_cubes(median_ndvi_july_masked,overlap_resolver='sum')

ndvi_june_and_july.download('ndvi_june_and_july.nc')
1 Like
5

Thanks Michele! Your nice examples helped me get the solution I wanted :slight_smile: