feat: add functionality to skip empty tiles in TIFF processing and clean up orphaned files

This commit is contained in:
Timon 2026-03-24 11:23:59 +01:00
parent 32cbf5c0db
commit 506af5079f
5 changed files with 194 additions and 20 deletions

View file

@ -0,0 +1,154 @@
"""
Clean empty field-tile TIFFs and orphaned RDS files
====================================================
Scans field_tiles/ and/or field_tiles_CI/ directories and identifies TIF files
where ALL pixels have RGBNIR == 0 (no satellite data collected).
Partially-covered tiles (some valid pixels present) are kept.
When deleting from field_tiles_CI/, also deletes the paired
daily_ci_vals/{FIELD}/{DATE}.rds file if it exists.
USAGE:
# Dry run — list empty files (default, scans both dirs):
& "C:\\Users\\timon\\anaconda3\\envs\\pytorch_gpu\\python.exe" python_app/clean_empty_tiles.py
# Actually delete:
& "...\\python.exe" python_app/clean_empty_tiles.py --delete
# Only one directory type:
& "...\\python.exe" python_app/clean_empty_tiles.py --dirs field_tiles
& "...\\python.exe" python_app/clean_empty_tiles.py --dirs field_tiles_CI
# Specific projects or fields:
& "...\\python.exe" python_app/clean_empty_tiles.py --projects angata aura --delete
& "...\\python.exe" python_app/clean_empty_tiles.py --fields 544 301 --delete
"""
import argparse
from pathlib import Path
import numpy as np
import rasterio
ROOT = Path(__file__).resolve().parent.parent
DEFAULT_DIRS = ["field_tiles", "field_tiles_CI"]
def is_empty_tif(path: Path) -> bool:
"""Return True if ALL pixels in RGBNIR bands are 0 or NaN (no satellite data).
Cloud-masked pixels are stored as 0 in uint16 (NaN is not representable).
A tile is considered empty only when every pixel across bands 1-4 is 0 or NaN,
meaning no valid satellite data was captured for that field on that date.
Partially-covered tiles (some pixels valid) return False and are left alone.
"""
try:
with rasterio.open(path) as src:
if src.count < 4:
return False # unexpected band count — leave it alone
rgbnir = src.read([1, 2, 3, 4]).astype(np.float32)
except Exception as e:
print(f" WARNING: could not open {path.name}: {e}")
return False
return bool(np.all((rgbnir == 0) | np.isnan(rgbnir)))
def scan_directory(storage_root: Path, dir_name: str, delete: bool, fields: list = None) -> dict:
"""Scan one tile directory within a project storage root.
When dir_name == 'field_tiles_CI' and delete=True, also removes the paired
daily_ci_vals/{FIELD}/{DATE}.rds file for each deleted TIF.
Returns:
dict mapping field_id -> list of empty Path objects
"""
tiff_root = storage_root / dir_name
# Paired RDS files only exist for field_tiles_CI output
rds_root = storage_root / "daily_ci_vals" if dir_name == "field_tiles_CI" else None
if not tiff_root.exists():
print(f" [{dir_name}] Directory not found: {tiff_root}")
return {}
field_dirs = sorted(d for d in tiff_root.iterdir() if d.is_dir())
if fields:
field_dirs = [d for d in field_dirs if d.name in fields]
print(f"\n [{dir_name}] Scanning {len(field_dirs)} fields ...")
results = {}
for field_dir in field_dirs:
tif_files = sorted(field_dir.glob("*.tif"))
empty = [f for f in tif_files if is_empty_tif(f)]
if empty:
results[field_dir.name] = empty
print(f" Field {field_dir.name:>6}: {len(empty)}/{len(tif_files)} empty"
f" ({', '.join(f.stem for f in empty)})")
total_empty = sum(len(v) for v in results.values())
total_tifs = sum(len(list(d.glob("*.tif"))) for d in field_dirs)
print(f"\n [{dir_name}] Summary: {total_empty} empty / {total_tifs} total TIFs"
f" across {len(results)} fields")
if delete and total_empty > 0:
print(f"\n [{dir_name}] Deleting {total_empty} empty TIFs ...")
rds_deleted = 0
for field_id, files in results.items():
for f in files:
f.unlink()
print(f" Deleted TIF: {f.relative_to(ROOT)}")
# Also remove the paired RDS from daily_ci_vals/ (Script 20 output)
if rds_root is not None:
paired_rds = rds_root / field_id / f"{f.stem}.rds"
if paired_rds.exists():
paired_rds.unlink()
print(f" Deleted RDS: {paired_rds.relative_to(ROOT)}")
rds_deleted += 1
print(f" [{dir_name}] Done. ({rds_deleted} paired RDS files also removed)")
elif not delete and total_empty > 0:
print(f"\n [{dir_name}] Dry run — pass --delete to remove these files.")
return results
def scan_project(project: str, delete: bool, fields: list = None, dirs: list = None) -> None:
storage_root = ROOT / "laravel_app" / "storage" / "app" / project
if not storage_root.exists():
print(f"[{project}] Project directory not found: {storage_root}")
return
print(f"\n[{project}] ========================================")
for dir_name in (dirs or DEFAULT_DIRS):
scan_directory(storage_root, dir_name, delete, fields)
def main():
parser = argparse.ArgumentParser(
description="Remove empty field-tile TIFFs and paired RDS files"
)
parser.add_argument(
"--delete", action="store_true",
help="Actually delete empty files (default: dry run)"
)
parser.add_argument(
"--projects", nargs="+", default=["angata"],
help="Project names to scan (default: angata)"
)
parser.add_argument(
"--fields", nargs="+", default=None,
help="Limit to specific field IDs, e.g. --fields 544 301"
)
parser.add_argument(
"--dirs", nargs="+", default=None, choices=DEFAULT_DIRS,
help=f"Which subdirs to scan (default: both {DEFAULT_DIRS})"
)
args = parser.parse_args()
print("=== Mode: DELETE ===" if args.delete else "=== Mode: DRY RUN (use --delete to remove) ===")
for project in args.projects:
scan_project(project, args.delete, args.fields, args.dirs)
if __name__ == "__main__":
main()

View file

@ -129,8 +129,18 @@ crop_tiff_to_fields <- function(tif_path, tif_date, fields, output_base_dir) {
# Crop raster to field boundary
tryCatch({
field_rast <- crop(rast, field_geom)
writeRaster(field_rast, output_path, overwrite = TRUE)
created <- created + 1
# Skip empty tiles: cloud-masked pixels are stored as 0 in uint16
# (NaN cannot be represented in that format). A band sum of 0 means
# no satellite data was captured for this field on this date.
band_sums <- terra::global(field_rast, fun = "sum", na.rm = TRUE)
if (sum(band_sums$sum, na.rm = TRUE) == 0) {
safe_log(paste("SKIP (no data):", field_name, tif_date), "WARNING")
skipped <- skipped + 1
} else {
writeRaster(field_rast, output_path, overwrite = TRUE)
created <- created + 1
}
}, error = function(e) {
safe_log(paste("ERROR cropping field", field_name, ":", e$message), "ERROR")
errors <<- errors + 1

View file

@ -200,22 +200,32 @@ main <- function() {
# Crop 5-band TIFF to field boundary
field_geom <- field_boundaries_sf %>% filter(field == !!field)
five_band_cropped <- terra::crop(five_band, field_geom, mask = TRUE)
# Save 5-band field TIFF
terra::writeRaster(five_band_cropped, output_tif_path, overwrite = TRUE)
# Extract CI statistics by sub_field (from cropped CI raster)
ci_cropped <- five_band_cropped[[5]] # 5th band is CI
ci_stats <- extract_ci_by_subfield(ci_cropped, field_boundaries_sf, field)
# Save RDS
if (!is.null(ci_stats) && nrow(ci_stats) > 0) {
saveRDS(ci_stats, output_rds_path)
# Skip empty tiles: cloud-masked pixels are stored as 0 in uint16
# (NaN cannot be represented in that format). Sum of RGBNIR bands == 0
# means no valid satellite data was captured for this field on this date.
rgbnir_sum <- sum(
terra::global(five_band_cropped[[1:4]], fun = "sum", na.rm = TRUE)$sum,
na.rm = TRUE
)
if (rgbnir_sum == 0) {
safe_log(sprintf(" SKIP (no data): field %s on %s", field, date_str), "WARNING")
} else {
# Save 5-band field TIFF
terra::writeRaster(five_band_cropped, output_tif_path, overwrite = TRUE)
# Extract CI statistics by sub_field (from cropped CI raster)
ci_cropped <- five_band_cropped[[5]] # 5th band is CI
ci_stats <- extract_ci_by_subfield(ci_cropped, field_boundaries_sf, field)
# Save RDS
if (!is.null(ci_stats) && nrow(ci_stats) > 0) {
saveRDS(ci_stats, output_rds_path)
}
fields_processed_this_date <- fields_processed_this_date + 1
raster_processed_this_date <- raster_processed_this_date + 1
}
fields_processed_this_date <- fields_processed_this_date + 1
raster_processed_this_date <- raster_processed_this_date + 1
}, error = function(e) {
# Error in individual field, continue to next
safe_log(sprintf(" Error processing field %s: %s", field, e$message), "WARNING")

View file

@ -438,8 +438,8 @@
# rmarkdown::render(
rmarkdown::render(
"r_app/90_CI_report_with_kpis_agronomic_support.Rmd",
params = list(data_dir = "aura", report_date = as.Date("2026-02-18"), language = "en" ),
output_file = "SmartCane_Report_agronomic_support_aura_2026-02-18_en_test.docx",
params = list(data_dir = "aura", report_date = as.Date("2026-03-23"), language = "en" ),
output_file = "SmartCane_Report_agronomic_support_aura_2026-03-23_en.docx",
output_dir = "laravel_app/storage/app/aura/reports"
)

View file

@ -57,7 +57,7 @@ AREA_UNIT_PREFERENCE <- tolower(Sys.getenv("AREA_UNIT", unset = "hectare"))
# Validate area unit value
if (!AREA_UNIT_PREFERENCE %in% c("hectare", "acre")) {
warning(paste0("Invalid AREA_UNIT env var: '", AREA_UNIT_PREFERENCE, "'. Using 'hectare'."))
AREA_UNIT_PREFERENCE <- "hectare"
AREA_UNIT_PREFERENCE <- "acre"
}
#' Get area unit label for display