RasterFlow Datasets

Public Preview

RasterFlow built-in datasets

RasterFlow includes popular open datasets for common geospatial use cases.

Sentinel-2

The Sentinel-2 dataset provides worldwide coverage, low resolution (10m to 30m, depending on the band) and 5-day revisit rates. Sentinel-2 is suitable for use cases that do not require high-resolution imagery and benefit from its frequent revisit rates for change detection or analyzing seasonal trends. RasterFlow has six built-in Sentinel-2 datasets. Pass one or more to the datasets input of the build_mosaics task:

`datasets` value	What it builds	Bands
`s2_median`	Per-pixel median over `[start, end)` at 10 m	B02, B03, B04, B08
`s2_median_all_bands`	The same median, with every band resampled to 10 m	B01–B09, B8A, B11, B12
`s2_closest_valid`	One composite at `target_date`, using the closest valid scenes in `[start, end]`	B02, B03, B04, B08
`s2_closest_valid_before_after`	Two closest-valid composites, at `start` and at `end`	B02, B03, B04, B08
`s2med_planting`	Planting-season median for the AOI’s location, one per full calendar year	B02, B03, B04, B08, N_VALID_PIXELS
`s2med_harvest`	Harvest-season median for the AOI’s location, one per full calendar year	B02, B03, B04, B08, N_VALID_PIXELS

You can combine Sentinel-2 datasets in one run, but not with NAIP. For each dataset’s full time rules, call Get a task for build_mosaics and read its description. For an example of a workflow that uses Sentinel-2 imagery, see the Fields of the World solution notebook. Additionally, for an example of how to build a Sentinel-2 mosaic using RasterFlow, see the Sentinel-2 Mosaic example notebook.

Cloud and quality filtering

RasterFlow automatically applies cloud and quality filtering when building Sentinel-2 mosaics. This ensures that the resulting composites contain only high-quality, cloud-free pixels.

Scene-level filtering

Scenes with 75% or more cloud cover are excluded from the mosaic. Only scenes with cloud_cover < 75% are considered as candidates for contributing pixels to the composite.

Pixel-level quality masking

Individual pixels are masked out based on the Sentinel-2 Scene Classification Layer (SCL) band. The following SCL values are masked:

SCL Value	Classification	Description
1	Saturated or defective	Sensor anomalies
3	Cloud shadow	Shadows cast by clouds
7	Unclassified	Unclassified pixels
8	Cloud medium probability	Likely cloud pixels
9	Cloud high probability	Highly likely cloud pixels
10	Thin cirrus	High-altitude ice clouds

The s2_closest_valid, s2_closest_valid_before_after, s2_median, and s2_median_all_bands datasets also mask out snow or ice (SCL value 11). s2med_planting and s2med_harvest keep snow and ice pixels.

After masking, RasterFlow computes a pixel-wise median (or selects the best scene, depending on the dataset) across the remaining valid observations to produce the final composite.

Seasonal date filtering

The planting and harvest season datasets (s2med_planting and s2med_harvest) apply latitude-based heuristics to determine the optimal day-of-year (DOY) ranges for imagery selection. This ensures that composites represent the appropriate growing stage for agricultural analysis at a given location. They only produce full calendar years, so set start and end to year boundaries: for 2023, use 2023-01-01T00:00:00Z through 2024-01-01T00:00:00Z. The FTW and PRUE recipes build these composites, so they follow the same rule.

Build a Sentinel-2 mosaic

This example builds 2023 planting- and harvest-season composites for Haskell County, Kansas:

The examples on this page run RasterFlow through the Tasks API, in two tabs. Python uses the Wherobots Python SDK, and API calls the REST API with curl. Outside a Wherobots notebook or job, both need a Wherobots API key in WHEROBOTS_API_KEY. Each example starts a task run; to wait for it and read its results, see Get started and Run RasterFlow Tasks.

# pip install wherobots-python-sdk
import wherobots

wb = wherobots.Wherobots()  # outside Wherobots, reads WHEROBOTS_API_KEY

run = wb.tasks.run("build_mosaics", {
    "datasets": ["s2med_planting", "s2med_harvest"],
    "aoi": "s3://wherobots-examples/rasterflow/aois/haskell.parquet",
    "start": "2023-01-01T00:00:00Z",
    "end": "2024-01-01T00:00:00Z",  # seasonal datasets need full calendar years
    "target_crs": "EPSG:32614",
}, region="aws-us-west-2")
# When the run succeeds, results["mosaic_index_uri"] lists the mosaic stores.

curl -X POST "https://api.cloud.wherobots.com/tasks/build_mosaics/runs" \
  -H "X-API-Key: $WHEROBOTS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "region": "aws-us-west-2",
    "name": "Haskell County seasonal mosaics",
    "inputs": {
      "datasets": ["s2med_planting", "s2med_harvest"],
      "aoi": "s3://wherobots-examples/rasterflow/aois/haskell.parquet",
      "start": "2023-01-01T00:00:00Z",
      "end": "2024-01-01T00:00:00Z",
      "target_crs": "EPSG:32614"
    }
  }'

# When the run succeeds, results.mosaic_index_uri lists the mosaic stores.

National Agriculture Imagery Program (NAIP)

The National Agriculture Imagery Program (NAIP) captures high-resolution imagery (30cm to 1m) across the continental United States. The imagery is captured regionally on a state by state basis and revisits states every two years. NAIP is available at different resolutions from 30cm to 1m, depending on the year of capture and the specific state. To ensure imagery in your Area of Interest (AOI) and time range are available in sufficient resolution, please refer to the USDA’s NAIP coverage map, a multi-layered PDF covering 2002 through 2025 that you download from that page.

Built-in NAIP configurations

RasterFlow includes four NAIP datasets. Pass one to the datasets input of the build_mosaics task:

`datasets` value	Resolution	Bands
`naip_30cm`	30 cm	Red, green, blue, near-infrared
`naip_60cm`	60 cm	Red, green, blue, near-infrared
`naip_rgb_30cm`	30 cm	Red, green, blue (visualization COGs with lossy JPEG compression)
`naip_rgb_60cm`	60 cm	Red, green, blue (visualization COGs with lossy JPEG compression)

A built-in mosaic labels its bands <dataset>:<band>, so a naip_30cm mosaic has the bands naip_30cm:red, naip_30cm:green, naip_30cm:blue, and naip_30cm:nir. The dataset name selects scenes at that native resolution. The optional resolution input sets the output pixel size; it does not change which scenes the dataset selects. For native 1-meter NAIP, use the chesapeakersc recipe or filter a tile index with build_gti_mosaics.

Which datasets the pre-configured models use

RasterFlow comes pre-configured with five open-source models. Each recipe selects its imagery source and resolution and passes the bands listed below to the model:

Model	`model_recipe` value	Imagery the recipe ingests	Bands used by the model
Meta SAM3	`sam3-text-bbox`, `sam3-text-geometry`	30 cm NAIP	RGB
Tile2Net	`tile2net`	30 cm NAIP	RGB
Meta CHM v1	`meta-chm-v1`	60 cm NAIP	RGB
Meta CHM v2	`meta-chm-v2`	60 cm NAIP	RGB
ChesapeakeRSC	`chesapeakersc`	1 m NAIP	RGB + NIR
Fields of the World	`ftw`	Sentinel-2 planting- and harvest-season composites	B02, B03, B04, and B08 from each season
PRUE	`prue`	Sentinel-2 planting- and harvest-season composites	B02, B03, B04, and B08 from each season

The SAM3 recipe builds a four-band NAIP mosaic, then uses only its RGB bands for inference.

A recipe ingests its own imagery, so you do not pass datasets when you run one. The datasets above are for building a mosaic yourself with the build_mosaics task. See RasterFlow Models for what each model returns.

Bring your own rasters

You are not limited to the built-in datasets. Query a STAC catalog to discover imagery for your Area of Interest, create a GDAL Raster Tile Index (GTI) from the results, and build a seamless mosaic from your own tiles with the build_gti_mosaics task. See the Bring Your Own Rasters (BYOR) with RasterFlow notebook for a complete walkthrough using NAIP imagery.

API reference

Run RasterFlow Tasks: run a task, run states and each task’s output
Tasks API reference: the endpoints for running tasks and reading their inputs

​RasterFlow built-in datasets

​Sentinel-2

​Cloud and quality filtering

Scene-level filtering

Pixel-level quality masking

​Seasonal date filtering

​Build a Sentinel-2 mosaic

​National Agriculture Imagery Program (NAIP)

​Built-in NAIP configurations

​Which datasets the pre-configured models use

​Bring your own rasters

​API reference

RasterFlow built-in datasets

Sentinel-2

Cloud and quality filtering

Seasonal date filtering

Build a Sentinel-2 mosaic

National Agriculture Imagery Program (NAIP)

Built-in NAIP configurations

Which datasets the pre-configured models use

Bring your own rasters

API reference