Methodology

This dataset is based off the ESA Level-1 “Ground Range Detected” product that has gone through a series of calibration, noise removal, radiometric and terrain correction, and speckle filter steps. Once applied, we consider the output a “Radiometric Terrain Corrected” (RTC) product that is suitable for analysis.

If you wish to perform these steps differently, you may be interested in the Level-1 GRD product which is also available on the AWS Registry of Open Data. We download our GRD data from this data source, making this product downstream of that dataset.

We perform all preprocessing steps using the Sentinel-1 Toolbox using version 7.0.2.

To preprocess the GRD scenes to RTC, we use the Sentinel-1 Toolbox to:

Step	Process	Description
1	Apply the orbit file	Refine positioning information. Restituted orbit files are used for products added after launch. The more accurate precise orbit files are used for historical imagery.
2	Apply thermal noise removal	Remove thermal noise from acquisition.
3	Apply GRD product border noise removal	Removes “noisy” pixels along the edges of GRD scenes. These pixels are significantly lower than expected but not labeled as NODATA in the GRD product.
4	Apply product calibration	Calibrate the product image “Digital Numbers” (DNs) to radiometrically calibrated backscatter.
5	Apply the Lee Sigma speckle filter	Parameter choices for the Lee Sigma filter are, Window size: 7x7; Target window size: 5x5; Number of looks: 2; Sigma: 0.9
6	Apply terrain flattening	We use the SRTM 1Sec HGT resampled using bilinear interpolation as the DEM.
7	Apply terrain correction	We use the same SRTM 1Sec HGT DEM. The image is exported in EPSG:4326 with a pixel resolution of ~20m.

Once the correction from GRD to RTC has been performed for all scenes for a given day, they are then regridded according to the MGRS grid used by Sentinel-2 20m products. This step is important for making analysis easier because it organizes the imagery into a predictable grid of known and consistent sizes and locations.

The process of creating grid tiles from Sentinel-1 scenes involves a few steps for a requested grid tile, date, polarization, and orbit direction:

1. Locate the RTC scene images that intersect with the requested grid tile for the requested date, orbit direction (ascending/descending), and polarizations (VV/VH).

2. For each RTC scene, reproject the image and read any observed data for the tile footprint. Areas that are not observed (e.g., no data) are masked.

3. Combine together the data we read from each RTC scene, taking into account where each RTC scene has unmasked data to fill in. For example, two north-south adjacent scenes from the same swath might be combined to create the tile image. If any pixels happen to overlap between two scenes, we take the maximum value.

4. Save the tile image as a Cloud Optimized GeoTIFF using cog_translate from the rio-cogeo package.

   • This function also generates image overviews for each tile using “average” resampling for the overview downscaling.

   • Finally, we run cog_validate from this package to double check that our COG images have been written correctly and according to conventions.

We perform this process of tiling the RTC scenes using “warped VRTs” from GDAL through the rasterio package.

Dataset Tags

To help provide information about the original GRD products that go into our RTC tiles, we write metadata as GeoTIFF image tags which you can access using GDAL.

These tags include,

Tag	Example	Description
ABSOLUTE_ORBIT_NUMBER	33706	Absolute orbit number
MISSION_ID	S1A	Mission satellite name
ORBIT_DIRECTION	ASC	Orbit direction, ascending (ASC) or descending (DSC)
DATE	2020-08-01	Acquisition starting date, formatted as %Y-%m-%d
TILE_ID	14TPN	MGRS 20m tile grid ID
VALID_PIXEL_PERCENT	34.385	Percent of the tile that is “observed”, “unmasked” or “valid”
NUMBER_SCENES	1	The number of Sentinel-1 GRD scenes that were mosaiced into the gridded tile
SCENES	[Sentinel-1 scene title]	A comma-separated list of the GRD scene titles that were mosaiced into the tile
SCENE_[N]_METADATA SCENE_[N]_PRODUCT_INFO	(too long to show)	For each Sentinel-1 GRD scene contained within the tile, we dump various metadata attributes describing the scene as JSON. These attributes include the scene UUID, polygon footprint,and acquistion mode information.