data_ingestion/sentinel2/preprocess_s2_ensemble_masks
Downloads and preprocesses Sentinel-2 imagery that covers the input geometry and time range, and computes improved cloud masks using an ensemble of cloud and shadow segmentation models. This workflow selects a minimum set of tiles that covers the input geometry, downloads Sentinel-2 imagery for the selected time range, and preprocesses it by generating a single multi-band raster at 10m resolution. It then improves cloud masks by merging the product mask with cloud and shadow masks computed using an ensemble of cloud and shadow segmentation models.
Sources
user_input: Time range and geometry of interest.
Sinks
raster: Sentinel-2 L2A rasters with all bands resampled to 10m resolution.
mask: Cloud masks at 10m resolution.
Parameters
min_tile_cover: Minimum RoI coverage to consider a set of tiles sufficient.
max_tiles_per_time: Maximum number of tiles used to cover the RoI in each date.
cloud_thr: Confidence threshold to assign a pixel as cloud.
shadow_thr: Confidence threshold to assign a pixel as shadow.
pc_key: Optional Planetary Computer API key.
Tasks
s2: Downloads and preprocesses Sentinel-2 imagery that covers the input geometry and time range.
cloud: Improves cloud masks by merging the product cloud mask with cloud and shadow masks computed by an ensemble of machine learning segmentation models.
Workflow Yaml
name: preprocess_s2_ensemble_masks
sources:
user_input:
- s2.user_input
sinks:
raster: s2.raster
mask: cloud.mask
parameters:
min_tile_cover: null
max_tiles_per_time: null
cloud_thr: null
shadow_thr: null
pc_key: null
tasks:
s2:
workflow: data_ingestion/sentinel2/preprocess_s2
parameters:
min_tile_cover: '@from(min_tile_cover)'
max_tiles_per_time: '@from(max_tiles_per_time)'
pc_key: '@from(pc_key)'
cloud:
workflow: data_ingestion/sentinel2/improve_cloud_mask_ensemble
parameters:
cloud_thr: '@from(cloud_thr)'
shadow_thr: '@from(shadow_thr)'
edges:
- origin: s2.raster
destination:
- cloud.s2_raster
- origin: s2.mask
destination:
- cloud.product_mask
description:
short_description: Downloads and preprocesses Sentinel-2 imagery that covers the
input geometry and time range, and computes improved cloud masks using an ensemble
of cloud and shadow segmentation models.
long_description: This workflow selects a minimum set of tiles that covers the input
geometry, downloads Sentinel-2 imagery for the selected time range, and preprocesses
it by generating a single multi-band raster at 10m resolution. It then improves
cloud masks by merging the product mask with cloud and shadow masks computed using
an ensemble of cloud and shadow segmentation models.
sources:
user_input: Time range and geometry of interest.
sinks:
raster: Sentinel-2 L2A rasters with all bands resampled to 10m resolution.
mask: Cloud masks at 10m resolution.