Composite Multi-Product Imagery

Composite imagery from two data sources and display as a single image.

from descarteslabs.catalog import Image, properties as p
from descarteslabs.utils import display
import numpy as np

# Define a bounding box around Taos in a GeoJSON

taos = {
    "type": "Polygon",
    "coordinates": [
        [
            [-105.71868896484375, 36.33725319397006],
            [-105.2105712890625, 36.33725319397006],
            [-105.2105712890625, 36.73668306473141],
            [-105.71868896484375, 36.73668306473141],
            [-105.71868896484375, 36.33725319397006],
        ]
    ],
}

# Create an ImageCollection
search = (
    Image.search()
    .intersects(taos)
    .filter(p.product_id.any_of(["landsat:LC08:01:RT:TOAR", "sentinel-2:L1C"]))
    .filter("2018-05-01" <= p.acquired < "2018-06-01")
    .filter(p.cloud_fraction < 0.2)
    .sort("acquired")
    .limit(15)
)
images = search.collect()

See which images we have, and how many per product:

print(images)

Out:

ImageCollection of 4 images
  * Dates: May 14, 2018 to May 29, 2018
  * Products: sentinel-2:L1C: 3, landsat:LC08:01:RT:TOAR: 1

And if you’re curious, which image IDs:

print(images.each.id)

Out:

'sentinel-2:L1C:2018-05-14_13SDA_99_S2A_v1'
'landsat:LC08:01:RT:TOAR:meta_LC08_L1TP_033035_20180519_20180520_01_RT_v1'
'sentinel-2:L1C:2018-05-24_13SDA_99_S2A_v1'
'sentinel-2:L1C:2018-05-29_13SDA_99_S2B_v1'

Make a median composite of the images

# Request a stack of all the images using the same GeoContext with lower resolution
arr_stack = images.stack("red green blue", resolution=60, data_type="Float64")

# Composite the images based on the median pixel value
composite = np.ma.median(arr_stack, axis=0)
display(composite, title="Taos Composite", size=2)