extra/

Name: Index
Creator: Major TOM

Subsets of global.parquet. Every row exists in global with the same id and the same grid and geoenrich: columns. The files below only add the columns listed for each.

File	Rows	Columns	Size	Contents
`land.parquet`	2,767,104	23	114 MB	Tiles covered by Sentinel-2 or Landsat. About half are coastal water and lakes inside the sensor swaths.
`land_s2.parquet`	1,735,044	34	128 MB	Tiles with a Sentinel-2 image whose patch aligns exactly with the scene: same CRS, origin on the scene's 60 m pixel grid, fully inside the 109.8 km granule, and at least 2 km inside its valid-data footprint.
`land_l8.parquet`	1,802,563	38	113 MB	Tiles with a Landsat 8 image assigned. Same single-MGRS restriction.
`elliot.parquet`	279,166	24	18 MB	ELLIOT sample (250,000 monotemporal + 29,166 temporal). Not restricted to single-MGRS tiles, which would break the designed split.

All files are GeoParquet 1.1.0, ZSTD, sorted by majortom:code_1000km → majortom:code_100km → id.

`land_s2.parquet` added columns

Column	Type	Description
`s2:time`	`timestamp[ms, UTC]`	Sentinel-2 acquisition time (UTC)
`s2:id_gee`	`string`	Google Earth Engine image id for the assigned Sentinel-2 scene
`s2:product_id`	`string`	ESA product identifier
`s2:spacecraft`	`string`	Spacecraft name (Sentinel-2A or Sentinel-2B)
`s2:processing_baseline`	`string`	ESA processing baseline version
`s2:orbit_number`	`uint16`	Relative orbit number
`s2:mean_solar_azimuth`	`float`	Mean solar azimuth angle in degrees, averaged over bands and detectors
`s2:mean_solar_zenith`	`float`

`land_l8.parquet` added columns

Column	Type	Description
`l8:time`	`timestamp[ms, UTC]`	Landsat acquisition time (UTC)
`l8:id_gee`	`string`	Google Earth Engine image id for the assigned Landsat scene
`l8:product_id`	`string`	USGS product identifier
`l8:spacecraft`	`string`	Spacecraft name (Landsat 8 or Landsat 9)
`l8:collection_number`	`uint8`	USGS Collection number
`l8:collection_category`	`string`	Collection category (T1, T2 or RT)
`l8:processing_software`	`string`	USGS processing software version
`l8:wrs_path`	`uint16`	WRS-2 path number

`elliot.parquet` added columns

Column	Type	Description
`split`	`string`	ELLIOT split assignment: monotemporal or temporal

Notes

land_l8.parquet is Landsat 8 only.
l8:cloud_cover is null where USGS reported -1 (not calculated).
The image patch is 1056 × 1056 px at 10 m, defined by majortom:geotransform in majortom:crs.
A land_s2 patch downloads from Earth Engine without resampling: use crs = majortom:crs, crsTransform = [10, 0, x0, 0, -10, y0] from majortom:geotransform, and 1056 × 1056 pixels (528 × 528 at 20 m, 176 × 176 at 60 m).
In about 0.4% of land_s2 tiles one band can have a few masked pixels inside the patch. The worst sampled case was 180 of 1,115,136 pixels.

extra/

Subsets of global.parquet. Every row exists in global with the same id and the same grid and geoenrich: columns. The files below only add the columns listed for each.

File	Rows	Columns	Size	Contents
`land.parquet`	2,767,104	23	114 MB	Tiles covered by Sentinel-2 or Landsat. About half are coastal water and lakes inside the sensor swaths.
`land_s2.parquet`	1,735,044	34	128 MB	Tiles with a Sentinel-2 image whose patch aligns exactly with the scene: same CRS, origin on the scene's 60 m pixel grid, fully inside the 109.8 km granule, and at least 2 km inside its valid-data footprint.
`land_l8.parquet`	1,802,563	38	113 MB	Tiles with a Landsat 8 image assigned. Same single-MGRS restriction.
`elliot.parquet`	279,166	24	18 MB	ELLIOT sample (250,000 monotemporal + 29,166 temporal). Not restricted to single-MGRS tiles, which would break the designed split.

All files are GeoParquet 1.1.0, ZSTD, sorted by majortom:code_1000km → majortom:code_100km → id.

`land_s2.parquet` added columns

Column	Type	Description
`s2:time`	`timestamp[ms, UTC]`	Sentinel-2 acquisition time (UTC)
`s2:id_gee`	`string`	Google Earth Engine image id for the assigned Sentinel-2 scene
`s2:product_id`	`string`	ESA product identifier
`s2:spacecraft`	`string`	Spacecraft name (Sentinel-2A or Sentinel-2B)
`s2:processing_baseline`	`string`	ESA processing baseline version
`s2:orbit_number`	`uint16`	Relative orbit number
`s2:mean_solar_azimuth`	`float`	Mean solar azimuth angle in degrees, averaged over bands and detectors
`s2:mean_solar_zenith`	`float`

`land_l8.parquet` added columns

Column	Type	Description
`l8:time`	`timestamp[ms, UTC]`	Landsat acquisition time (UTC)
`l8:id_gee`	`string`	Google Earth Engine image id for the assigned Landsat scene
`l8:product_id`	`string`	USGS product identifier
`l8:spacecraft`	`string`	Spacecraft name (Landsat 8 or Landsat 9)
`l8:collection_number`	`uint8`	USGS Collection number
`l8:collection_category`	`string`	Collection category (T1, T2 or RT)
`l8:processing_software`	`string`	USGS processing software version
`l8:wrs_path`	`uint16`	WRS-2 path number

`elliot.parquet` added columns

Column	Type	Description
`split`	`string`	ELLIOT split assignment: monotemporal or temporal

Notes

land_l8.parquet is Landsat 8 only.
l8:cloud_cover is null where USGS reported -1 (not calculated).
The image patch is 1056 × 1056 px at 10 m, defined by majortom:geotransform in majortom:crs.
A land_s2 patch downloads from Earth Engine without resampling: use crs = majortom:crs, crsTransform = [10, 0, x0, 0, -10, y0] from majortom:geotransform, and 1056 × 1056 pixels (528 × 528 at 20 m, 176 × 176 at 60 m).
In about 0.4% of land_s2 tiles one band can have a few masked pixels inside the patch. The worst sampled case was 180 of 1,115,136 pixels.

Index

Index

extra/

land_s2.parquet added columns

land_l8.parquet added columns

elliot.parquet added columns

Notes

extra/

land_s2.parquet added columns

land_l8.parquet added columns

elliot.parquet added columns

Notes

`land_s2.parquet` added columns

`land_l8.parquet` added columns

`elliot.parquet` added columns

`land_s2.parquet` added columns

`land_l8.parquet` added columns

`elliot.parquet` added columns