pantierra/efast-phenocam-validation
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

looked at preparation.

Browse source
This commit is contained in:
Felix Delattre 2026-03-04 14:37:51 +01:00
parent 585b2d31b0
commit 585919d130
2 changed files with 47 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

28
preparation.py
View file

@ -72,6 +72,7 @@ def _reproject_raster_to_target(
def prepare_s2(season, site_position, site_name, cleaning_strategy="aggressive", date_range=None):
lat, lon = site_position
s2_dir = Path(f"data/{site_name}/{season}/raw/s2/")
s3_dir = Path(f"data/{site_name}/{season}/raw/s3/")
s2_output_dir = _get_base_dir(season, site_name, cleaning_strategy) / "s2"
@ -79,6 +80,8 @@ def prepare_s2(season, site_position, site_name, cleaning_strategy="aggressive",
clouds = _load_excluded(season, site_name, cleaning_strategy)
s2_output_dir.mkdir(parents=True, exist_ok=True)
print(f"[S2-PREP] Starting preparation: {site_name} ({lat:.6f}, {lon:.6f}), {season}, strategy={cleaning_strategy}")
s3_files = [f for f in s3_dir.glob("*.geotiff") if f.name not in clouds["s3"]]
if not s3_files:
raise ValueError("No non-cloud S3 files found for reference bounds")
@ -89,14 +92,17 @@ def prepare_s2(season, site_position, site_name, cleaning_strategy="aggressive",
s2_width = s3_ref.width * RESOLUTION_RATIO
s2_height = s3_ref.height * RESOLUTION_RATIO
for s2_file in s2_dir.glob("*.geotiff"):
for s2_file in sorted(s2_dir.glob("*.geotiff")):
if s2_file.name in clouds["s2"]:
print(f"[S2-PREP] Skipping {s2_file.name} (excluded by {cleaning_strategy})")
continue
date_str = s2_file.name.split("_")[0]
refl_dst = s2_output_dir / f"S2A_MSIL2A_{date_str}_REFL.tif"
if refl_dst.exists():
print(f"[S2-PREP] Skipping {s2_file.name} (exists)")
continue
print(f"[S2-PREP] Processing {s2_file.name}...")
temp_normalized = s2_output_dir / f"temp_{s2_file.name}"
with rasterio.open(s2_file) as src:
data = src.read().astype("float32") / 10000.0
@ -109,12 +115,16 @@ def prepare_s2(season, site_position, site_name, cleaning_strategy="aggressive",
temp_normalized, refl_dst, target_bounds, target_crs, s2_width, s2_height
)
temp_normalized.unlink()
print(f"[S2-PREP] Saved: {refl_dst}")
print(f"[S2-PREP] Computing distance-to-clouds...")
distance_to_clouds = _import_distance_to_clouds()
distance_to_clouds(s2_output_dir, ratio=RESOLUTION_RATIO)
print("[S2-PREP] Completed")
def prepare_s3(season, site_position, site_name, cleaning_strategy="aggressive", date_range=None):
lat, lon = site_position
s3_dir = Path(f"data/{site_name}/{season}/raw/s3/")
base_dir = _get_base_dir(season, site_name, cleaning_strategy)
s2_prepared_dir = base_dir / "s2"
@ -123,20 +133,27 @@ def prepare_s3(season, site_position, site_name, cleaning_strategy="aggressive",
clouds = _load_excluded(season, site_name, cleaning_strategy)
s3_preprocessed_dir.mkdir(parents=True, exist_ok=True)
print(f"[S3-PREP] Starting preparation: {site_name} ({lat:.6f}, {lon:.6f}), {season}, strategy={cleaning_strategy}")
s3_by_date = defaultdict(list)
for s3_file in s3_dir.glob("*.geotiff"):
if s3_file.name not in clouds["s3"]:
s3_by_date[s3_file.name.split("_")[0]].append(s3_file)
else:
print(f"[S3-PREP] Skipping {s3_file.name} (excluded by {cleaning_strategy})")
print(f"[S3-PREP] Found {sum(len(v) for v in s3_by_date.values())} acquisitions across {len(s3_by_date)} dates")
temp_composite_dir = s3_preprocessed_dir / "temp_composites"
if temp_composite_dir.exists():
shutil.rmtree(temp_composite_dir)
temp_composite_dir.mkdir()
for date_str, s3_files in s3_by_date.items():
for date_str, s3_files in sorted(s3_by_date.items()):
composite_path = temp_composite_dir / f"composite_{date_str}.tif"
if len(s3_files) == 1:
shutil.copy(s3_files[0], composite_path)
print(f"[S3-PREP] Composite {date_str}: 1 acquisition")
else:
s3_stack = []
for s3_file in s3_files:
@ -150,6 +167,7 @@ def prepare_s3(season, site_position, site_name, cleaning_strategy="aggressive",
profile.update({"count": composite.shape[0], "dtype": "float32"})
with rasterio.open(composite_path, "w", **profile) as dst:
dst.write(composite)
print(f"[S3-PREP] Composite {date_str}: {len(s3_files)} acquisitions merged")
# Reproject S3 to match S2 REFL bounds (full coverage) instead of DIST_CLOUD bounds
# This ensures fusion covers the same area as S2 and dimensions match
@ -174,8 +192,10 @@ def prepare_s3(season, site_position, site_name, cleaning_strategy="aggressive",
height,
)
print(f"[S3-PREP] Reprojecting {len(list(temp_composite_dir.glob('*.tif')))} composites to S2 grid ({width}×{height} px)...")
# Reproject each S3 composite to match S2 REFL bounds
sen3_paths = list(temp_composite_dir.glob("*.tif"))
sen3_paths = sorted(temp_composite_dir.glob("*.tif"))
for sen3_path in sen3_paths:
vrt_options = {
"transform": s3_transform,
@ -191,5 +211,7 @@ def prepare_s3(season, site_position, site_name, cleaning_strategy="aggressive",
profile = vrt.profile.copy()
profile.update({"dtype": "float32", "nodata": 0, "driver": "GTiff"})
rio_shutil.copy(vrt, outfile, **profile)
print(f"[S3-PREP] Saved: {outfile}")
shutil.rmtree(temp_composite_dir)
print("[S3-PREP] Completed")