ndvi-timeserieslisted

1. Inputs: parcel asset id, date range, index (NDVI default), step (5 days). 2. In GEE: cloud-masked S2 → index image per date → `reduceRegions(mean + count, scale=10, tileScale=8)` per date with `date` property → flatten → `Export.table.toDrive` (CSV) or `toAsset`. Drop observations with `count < 50 %` of parcel pixels. 3. Locally: pivot to parcel × date; interpolate gaps ≤ 20 days (linear), then Savitzky–Golay (window 5, order 2) or Whittaker smoothing; keep raw and smoothed columns. 4. Output `ndvi_ts_<aoi>_<range>.parquet` + a quick plot of 5 random parcels for sanity.