Mapping our land more clearly and colourfully
for better analysis and visualisation

Acknowledging Sentinel Missions

ESA: Sentinel Online

ESA: Sentinel-2

 



Sentinel-2 liftoff in June 2015

Earth Observation (EO) is the most effective way to monitor the environment of our changing home planet. The recent significant contributions by the U.S. and the European Union countries are the successful lunch and operation of Landsat-8 and Sentinel-2 satellites. High-quality and rich multispectral data covering very large acquisition areas are freely available and provide enormous opportunities for innovation and applications. They are "YOUR satellites" and highly complementary (even competitive for many applications) to high-resolution imaging.

We wish to help users quickly realise the full potential of the latest Landsat and Sentinel imagery. Our aim is to develop the highest-quality 10m-resolution Sentinel-2 and 15m-resolution Landsat-8 imagery composites (in natural and false colours). Quality matters!

 


A Unique Study: Identifying Flying Airplanes Over a Very Large Region Using 10m-Resolution Sentinel-2 Satellite Imagery

 

Some generous user comments:

"Your software made my job easy." (Comment from a user)

"I like what I've seen of this software so far! I do a ton of image processing and analysis daily, and this makes stacking and sharpening much simpler than in arcgis, envi, or photoshop (best results so far, then I found yours)... Thanks for your time, and the software!" (Email from an industry veteran, March 2016)

"It is a very impressive piece of software and created very crisp and natural looking images." (Comment from an industry veteran)

"Thank you for an excellent product. I have used other pan-sharpeners and yours beats them all." (Email from an industry veteran, January 2016)

"I have found ST for Landsat-8 Imagery to be my most useful tool in working with the OLI images. Thanks for developing these tools!." (Email from a retired industry veteran, January 2016)"

"I am using your Spectral Transformer and all works perfectly. Many thanks to you for this great and accurate tool." (Comments from a user)

"I enjoy your GUI tool. It's like an App on PC and the first button click acts as simple as the 'voice' input in Google search." (Comment from an industry veteran)

"I recently downloaded the Spectral Transformer for Landsat 8 software... thoroughly impressed with the ease of use, speed and quality of the results..." (Email from a senior image processing specialist, January 2016)

"Thanks for making available two software programs that have greatly improved my ability to generate actionable imagery. I am a geologist that works in many environments and I require a great variety of band combinations to optimize information... Thanks to your software I have been able to greatly improve our targeting operations..." (Email from a senior geologist, 2018)

"We are still using your software with great pleasure..."; "I am still incredibly pleased with your programmes..."; "Thank you for your wonderful spectral analysis suites..."; "Many thanks for your usual prompt and efficient reply..." (Emails from users)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Spectral Discovery for Sentinel-2 Imagery

Sentinel-2

 

Advanced software to easily and rapidly make colourful and detail-rich (10m-resolution) Sentinel-2 imagery composite, with 3 simple button clicks.

  • Popular software for geospatial professionals and general users. It is productive and saves time!
  • Functions include band combinations, image stretching, image pan-sharpening, and exploratory image feature extraction. Perfect software tool to get the daily fresh Sentinel-2 imagery into GIS and remote sensing software, and Google Earth Pro.
  • Version 3.1 released in 03/2020
  • Recent features
    • 1 button click for automated NDVI calculation, with 3 types of outputs (demo)
    • 1 button click for automated extraction of surface water areas, with 3 types of outputs (demo)
    • batch processing
    • supporting the new format of Sentinel-2 Level-1C products published after 12/2016 (link)
    • multiple common band combination options included
    • support reading Sentinel-2 Level -2A products (03/2020)


Run software (with 3 button clicks)

  • Button 1: Select metadata file (*.xml) of your downloaded Sentinel-2 granule / tile, and the rest of GUI fields will be automatically populated.
  • Button 2: Convert downloaded imagery data in JPEG2000 format to uncompressed, analysis-ready GeoTIFF format (assume the most popular open-source GIS software QGIS already installed)
  • Button 3: Run three steps altogether (Band Combination, Image Stretching, and Image Pan-sharpening) and output files in GeoTIFF format.

      

 

 

Sentinel-2 Image Sources

ESA's Sentinel Online provides the most comprehensive information about all satellites from the Sentinel missions, including Sentinel-2. It is always useful to read its user guide, handbook, etc. before conducting proper image processing.

Sentinel-2 (2A and 2B satellites) imaging features wide-swath (290 km), multi-spectral (13 spectral bands), high-resolution (4 visible and near-infrared bands at 10 m, 6 red-edge/shortwave-infrared bands at 20 m, and 3 atmospheric correction bands at 60 m), and high revisit frequency (5 days at the Equator).

 

A - Swath width comparisons: Sentinel-2, Landsat-8, and the very high-resolution WorldView-3

 

B - Comparison of spectral bands between Sentinel-2 and Landsat-8 (the "cousins"): (Source: NASA)

 

C - Sentinel-2 spectral bands:

 

Details of 13 bands:

  Band Name Central Wavelength
       
Resolution 10m: B02 Blue 490nm
  B03 Green 560nm
  B04 Red 665nm
  B08 NIR 842nm
       
Resolution 20m: B05 Red Edge 1 705nm
  B06 Red Edge 2 740nm
  B07 Red Edge 3 783nm
  B8a Red Edge 4 865nm
  B11 SWIR 1 1610nm
  B12 SWIR 2 2190nm
       
Resolution 60m: B01 Aerosols 443nm
  B09 Water vapor 940nm
  B10 Cirrus 1375nm
       

Common band combinations for Red/Green/Blue (RGB) imagery composite:

  Sentinel-2 Landsat-8
Natural colour 04 / 03 / 02 4 / 3 / 2
Colour infrared
(good for vegetation identification)
08 / 04 / 03 5 / 4 / 3
Shortwave infrared (SWIR, good for mapping fire scars or burnt areas) 12 / 11 / 8a 7 / 6 / 5
     

 

D: Sentinel-2 Level-1C data product

Features of the Sentinel-2 Level-1C data product:

  • Representing Top of Atmosphere (TOA) reflectance scaled by 10,000
  • Applied radiometric and geometric corrections (including orthorectification and spatial registration or georeference)
  • GIS / mapping ready cartographic coordinates (with UTM / WGS84 projection)
  • A full large scene is divided into a set of granules (also called tiles), each at 100 km x 100 km in size (excluding overlapping edges of 9.8 km). Typical file size of all bands with a granule / tile: ~500 MB (in compressed JPEG2000 format). User can download data for a full scene or an individual granule / tile, up to the download method (introduced below).


E: Identify granule / tile for your Area of Interest (AOI)

Sentinel-2 granule / tile gridding is based on the NATO Military Grid Reference System (MGRS, GIS files). For quick and easy identification, Sentinel Online is providing a Google Earth KML file: Sentinel-2 Level-1C tiling grid (103 MB, uncompressed). Alternatively, a compressed version can be downloaded here (zip file, ~9 MB only).

Open this KML file in Google Earth, and then locate the tile name in the format [UTM code] / latitude band / square / for your AOI. For example, the tile name for Paris is 31 / U / DQ / (as shown in the figure below).

 

F: Methods of downloading Sentinel-2 Level-1C data

After knowing the granule / tile name for an AOI, one can start to download Sentile-2 imagery data right away. For example, the Method 1 introduced below is a very easy option.

Method 1: Sentinel-2 on AWS (public bucket sentinel-s2-l1c)
https://registry.opendata.aws/sentinel-2/

The use of GUI-based tools (e.g. S3 Browser) or AWS Command Line Interface can make data download very quickly. The User Guide below contains some practical tips on how to quickly download data by folders from AWS S3.

Using easy-to-use web-based mapping Apps, e.g. Sentinel Hub’s EO Browser
https://apps.sentinel-hub.com/eo-browser/
one can get file path of the Sentinel-2 imagery stored on AWS S3.


Method 2: Sentinel-2 imagery on Google Earth Engine and Google Cloud Platform:
https://cloud.google.com/storage/docs/public-datasets/sentinel-2

Method 3: ESA's Copernicus Open Access Hub, or directly, map-based imagery search tool (to download large-sized whole scenes, typical file size ~7 GB)

Method 4: USGS EarthExplorer - http://earthexplorer.usgs.gov/ or GloVis - http://glovis.usgs.gov/

Method 5: Copernicus Australasia http://www.copernicus.gov.au/

 

Detailed methods are introduced in the software User Guide (PDF file).

      

      

 

G: Ready to go: If you have already downloaded some Sentinel-2 granules / tiles, and installed a recent version of open-source GIS software QGIS, then using the Spectral Discovery for Sentinel-2 Imagery is very easy. It may take less than 10 seconds to process a granule / tile and make a beautiful GIS/mapping-ready imagery composite!



New Image Analysis Software Tools

Recently, there are cloud-based platforms to process a massive amount of Earth imagery at the global level. But for the majority of professionals, the main approach for detailed image classification and interpretation is still desktop based, along with important field knowledge that is essential for accuracy assessment and outcome validation.

Spectral Discovery for Sentinel-2 Imagery we are contributing to is a dedicated software tool that promises easy and rapid processing of Sentinel-2 imagery for both geospatial professionals and general users. Unique features on image stretching, image fusion, exploratory feature extraction and batch processing are included in a simple workflow. It performs most common and important image analysis steps:

  • Step 1 - Band combination (to make three-band imagery composite)
  • Step 2 - Image histogram stretching (to make colourful composite)
  • Step 3 - Image pan-sharpening (to make colourful and spatially sharper composite)
  • Step 4 - Exploratory image feature extraction with image feature indices (e.g. NDVI and NDWI) and band ratios

 

Step 1: Simple and convenient band combination to make three-band imagery composite

Sentinel2_RGB.exe
<red_band_in> <green_band_in> <blue_band_in> <RGB_out.tif>

e.g. Sentinel2_RGB.exe B04.tif B03.tif B02.tif RGB_B04B03B02_Res10m.tif
(for true colour)

e.g. Sentinel2_RGB.exe B08.tif B04.tif B03.tif CIR_B08B04B03_Res10m.tif
(for colour infrared)
 

 

Step 2: Adaptive linear and non-linear image histogram stretching over the three-band imagery composite

Sentinel2_Stretch.exe
<red_band_order> <green_band_order> <blue_band_order> -s
<image_stretch_band1_left_cut> <image_stretch_band1_right_cut> <image_stretch_band2_left_cut> <image_stretch_band2_right_cut> <image_stretch_band3_left_cut> <image_stretch_band3_right_cut> <nonlinear_stretch_band1>
<nonlinear_stretch_band2>
<nonlinear_stretch_band3>
<RGB_in.tif> <RGB_stretched_out.tif>

e.g. Sentinel2_Stretch.exe 1 2 3 -s 2 0.2 2 0.2 2 0.2 10 10 10
B04B03B02_Res10m.tif B04B03B02_Stretched_Res10m.tif
e.g. Sentinel2_Stretch.exe 1 2 3 -s 0 0 0 0 0 0 0 0 0
B11B8aB05_Res20m.tif B11B8aB05_Stretched_Res20m.tif

 

Step 3: Advanced and fast image pan-sharpening

Sentinel2_Fuse.exe
<pan-sharpening_method> -s
<image_stretch_band1_left_cut> <image_stretch_band1_right_cut> <image_stretch_band2_left_cut> <image_stretch_band2_right_cut> <image_stretch_band3_left_cut> <image_stretch_band3_right_cut> <nonlinear_stretch_band1>
<nonlinear_stretch_band2>

<nonlinear_stretch_band3>
<pan_band_in.tif> <ms_in.tif> <fused_out.tif>

e.g. Sentinel2_Fuse.exe 1 -s 2 0.2 2 0.2 2 0.2 10 10 10
B04.tif B11B8aB05_Res20m.tif B11B8aB05_B04_Fused_Res10m.tif

 

Step 4: Calculating image feature indices and band ratios

Sentinel2_Feature.exe
<calculation_method> <input_band1.tif> <input_band2.tif> <threshold>

e.g. Sentinel2_Feature.exe 1 B08.tif B04.tif 0.3

 

Tips:

  • Make sure the length of file name/path for each output is less than 256 characters, as the software automatically append file name extensions such as "_Res10m_Stretched.tif". You may shorten or remove some very long folder names first, although the software does check the length of each input and output file name/path.
  • Please note that the first input to the software is the metadata file (*.xml) specific to the granule/tile, not the metadata file (also in *.xml) specific to the whole scene.
  • File names for all outputs are automatically populated by the software, so usually there is no need to modify those names.
  • If a scene contains a large proportion of white objects (e.g. cloud and snow), try to increase the right-end cut-off values in image stretching for their exclusion.
  • To improve the brightness of the imagery composite, one may increase non-linear stretching values to 10, 20, 30 and so on.
  • Once comfortable with a set of linear and non-linear stretching options for a single band combination (at 10m-, 20m- or 60m-resolution bands, respectively), the user may start to run batch band combinations at a corresponding resolution level.
  • Output files from image stretching and pan-sharpening, if exist, will be over-written automatically.
  • The same procedure equally works for any three-band combinations. Through iterations, any number of multispectral bands can be fused or pan-sharpened.
  • In pan-sharpening, the higher resolution RED band (Band 4 at 10m resolution) is commonly selected as an ideal "Pan" band in order to achieve sharper results.
  • To process 100s or 1000s of granules / tiles, one may use the DOS-based command-lines directly.
  • The software needs to allocate a max of ~1.2 GB RAM for the pan-sharpening process, and this may cause a specific run to fail IF the machine has limited RAM and/or fragmented RAM. If this occurs, please try to run the software on a more powerful machine with sufficient RAM (e.g. 8 GB or more).

 

Run time:

Spectral Discovery for Sentinel-2 Imagery is very productive. The time to process a full Sentinel-2 granule / tile on an average office computer is far less than 1 minute. Detailed estimates are as follows:

  • Step 1 (band combination): 1-5 seconds
  • Step 2 (image stretching): 5-15 seconds
  • Step 3 (image pan-sharpening): 10-60 seconds


Notes:

It is with great privilege that we all have access to another superior imagery source. We are very motivated and have tried our best to develop the tools that can preserve that high quality as much as possible.

We are confident that anyone even with limited image processing experience can perform the above analysis. It is our hope that more people can take advantage of and benefit from the excellent Sentinel-2 imagery. Thank you, the Copernicus Programme and ESA!

 

Rapid Processing for Innovative Applications

This unique software can be useful for a wide range of environmental studies, including:

  • Monitoring land cover changes (e.g. deforestation and urban expansion)
  • Mining exploration and mineral detection with SWIR bands
  • Precision agriculture and vegetation mapping (e.g. NDVI)
  • Natural hazards (e.g. fire scar mapping, flooding, volcanic eruption)
  • Updating GIS basemaps with the latest fresh imagery

Three more button clicks: To rapidly make many band combinations using Sentinel-2's full spectral bands in a batch mode:

  • 24 = 4x3x2, using 4 multispectral bands at 10m-resolution - Bands 02, 03, 04, 08 (~10 GB disk space required for each granule / tile)
  • 120 = 6x5x4, using 6 multispectral bands at 20m-resolution - Bands 05, 06, 07, 8a, 11, 12 (~12GB disk space required for each granule / tile)
  • 6 = 3x2x1, using 3 multispectral bands at 60m-resolution - Bands 01, 09, 10 (~0.1 GB disk space required for each granule / tile)

A simple yet extremely powerful way of image (visual) analysis: Through rapid, various band combinations, important features/phenomena of interest on Earth land surface can be better highlighted and revealed.

Example 1: Mapping fire scars (burnt areas) with numerous band combinations in an automated procedure. Massive fire scars from the May 2016 Fort McMurray Fire, Alberta, Canada. Left: Natural colour imagery; Right: False colour SWIR imagery. Tile name: 12/V/VH; Date: 2016/5/12.

 

See the Detail: Advanced Image Fusion and Pan-sharpening

(A) Compare spatial resolution from a pan-sharpening process. Left: Natural colour imagery (B04/B03/B02) at original 10m resolution; Right: False-colour imagery (B11/B8A/B05) pan-sharpened at 10m resolution. Tile name: 10/S/EG; Date: 2016/4/15. Location: San Francisco.

(B) Compare images before and after pan-sharpening. Left: False-colour imagery (B11/B8A/B05) at original 20m resolution; Right: False-colour imagery (B11/B8A/B05) pan-sharpened at 10m resolution. Tile name: 10/S/EG; Date: 2016/4/15. Location: San Francisco.

(C) Compare images before and after pan-sharpening. Left: False-colour imagery (B10/B09/B01) at original 60m resolution; Right: False-colour imagery (B10/B09/B01) pan-sharpened at 10m resolution. Tile name: 32/U/LB; Date: 2016/5/8. Location: near Cologne, Germany.

Example 2: Unique pan-sharpening of Sentinel-2 imagery from coarse 20m- or 60m-resolution to very sharp 10m-resolution. In these cases, a higher resolution (10m) band - Band 4 (red band) - is used as the "pan" band for pan-sharpening.

 

 

See the Invisible: Exploring Unique Short-wave Infrared Bands

Like Landsat-8, Sentinel-2 short-wave infrared (SWIR) bands are capable of detecting wildfire hotspots and lava flows/heat through smoke.

Example 3: The May 2016 Fort McMurray Fire, Alberta, Canada. Left: Natural colour imagery showing full smoke; Right: False-colour SWIR imagery. Tile name: 12/V/VH; Date: 2016/5/5.

 

See Features of Interest (FOI): From Pixels to Information

Exploratory image feature extraction with image indices (e.g. Normalised Difference Vegetation Index - NDVI, Normalised Difference Water Index - NDWI, Normalised Difference Urban Index, Normalised Difference Snow Index, Burned Area Index) and various band ratios, for a wide range of environmental studies.

Example 4: Calculating feature indices with three types of outputs automatically from the software. Left: Natural colour imagery; Right: NDVI. Tile name: 10/S/EG; Date: 2016/4/15. Location: San Francisco.

 

See the Big Picture: Using Global Imagery Mosaics

Need a global satellite imagery mosaic? Please take a look at the high-quality global satellite imagery mosaics (30m-resolution, multiple styles) in ELS2000 series.

Automated, exploratory land cover classifications at 10m resolution, at the global scale (ongoing)

Land cover changes

Near real-time / daily monitoring

...

Processed Samples

The imagery data for each Sentinel-2 granule / tile is from AWS. Please make sure you do download input imagery for each granule / tile based on its whole data folder, not one file by one file; otherwise, the download process might be too laborious. All outputs are based on the auto parameterisation in the software.

 

Evaluation

Please evaluate the quality of the processed imagery (stretched and pan-sharpened) in terms of:

- spectral consistency / colour saturation / colour contrast
- spatial precision
/ object detail / spatial sharpness / spatial texture

Be the judge on the performance of the tools. Quality Matters!

Stretched or pan-sharpened result at 10m resolution in GeoTIFF, ready to be used in all GIS and mapping software (e.g. QGIS, ArcGIS, MapInfo), popular image-editing software (e.g. PhotoShop), and Google Earth Pro.

For demo only, output image in JPEG format is included in zip download below:

- JPG (plus *.xml file)


San Francisco

Tile name: 10/S/EG
Date: 2016/4/15

Processed output in JPEG

Rio de Janeiro

Tile name: 23/K/PQ
Date: 2016/4/24

Processed output in JPEG

Paris

Tile name: 31/U/DQ
Date: 2016/1/25

Processed output in JPEG

Cape Town

Tile name: 34/H/BH
Date: 2016/4/6

Processed output in JPEG

Beijing

Tile name: 50/T/MK
Date: 2016/5/3

Processed output in JPEG

Perth

Tile name: 50/H/LK
Date: 2016/5/12

Processed output in JPEG

It only takes ~10 seconds to process each granule / tile on an ordinary computer. All daily scenes can be processed as above and efficiently delivered as soon as they are available in the following formats (for professionals in remote sensing and GIS fields, as well as the general public):

  • Conventional GeoTIFF
  • JPEG2000
  • KMZ (Google Earth)
  • Various web mapping tilesets, e.g. Google Map Tiles, Bing Maps Tiles, OSM Tiles, TMS Tiles

Like Landsat-8, the potential for the broadest use of the new Sentinel-2 imagery is great.






Disclaimer | Privacy Statement | ©2021 GeoSage