what processing level of WorldView-2 image to request to run Tasseled Cap

[JerryHelfand1]

When ordering WV-2 data, one has the option of ordering raw data that includes some serious haze OR atmospherically-corrected bands (provided directly by DigitalGlobe), which show surface reflectance.  Which version is better for application of the Tasseled Cap tool in ArcGIS Pro?

Note, I don't have atmospheric correction software, and so I'm hoping that surface reflectance is OK.  The instructions for this tool warn to put in raw, unstretched & unresampled data, which is ambiguous because that may or may not include the atmospheric correction.

[DanPatterson_Retired]

From http://desktop.arcgis.com/en/arcmap/latest/manage-data/raster-and-images/tasseled-cap-function.htm

If the data has been preprocessed, this function will not provide the correct results. It is best to use the raw data from the imagery provider. Also, the preprocessed 8-bit IKONOS, QuickBird and WorldView-2 data is not valid since the pixel values have been altered.

If that means anything to you

[JerryHelfand1]

Dan,

I read that document before posting.  See, if I follow the documentation as you suggest, then I would want to use non-atmospherically-corrected image for the Tasseled Cap function.  Maybe if I calculate NDWI and NDVI, the results might be OK but not optimal.  But if I want to do image classification (supervised or object-based) and compare results from different vintage of images, then it's required I use the atmospherically-corrected version.   

As backup for my stance, the section on Atmospheric Correction in Imagery and GIS Best Practices for Extracting Information from Imagery  (by Green, Congalton & Tukman, ESRI Press 2017 pp 153-156) supports this, and also mentions that "availability of [atmospheric absorption data from specific bands within the image],.... has sparked a resurgence of algorithm development to produce better absolute atmospheric correction methods." (p155).  What is not clear is whether the latest capabilities have been programmed into ArcGIS/Pro.

So, might you or anyone have a link to an document that describes the exact parameters used (the formula) in ArcGIS Pro or ArcGIS 10.5, and the methodology followed to derive them?  The existing documentation is still a bit of a black box, and doesn't seem to have been updated other than adding newer sensor names.  For the public health project I'm supporting on a volunteer (zero pay) basis, this is far less than ideal. 

For now, until I get or find more definitive information, I must ignore the Tasseled Cap function because it may give varying results from image collected on different dates because the reflectance will vary along with variations in atmospheric moisture content.   

[DanPatterson_Retired]

Perhaps someone from the Image team could leap in if they are around

[JeffLiedtke]

Thanks for the good question, and subsequent input. 

The Tasseled Cap coefficients are derived from uncorrected imagery, therefore uncorrected imagery is required to obtain valid TC results.  The coefficients are normalized to Top of Atmosphere, so some correction is applied inherently; for more details, see:  Lance D. Yarbrough, Kumar Navulur & Rachana Ravi (2014). Presentation of the Kauth–Thomas transform for WorldView-2 reflectance data, Remote Sensing Letters, 5:2, 131-138.  We have asked DigitalGlobe for information regarding their plans to update TC coefficients for Surface Reflectance products, and will report back when we receive a disposition from them.

[JeffLiedtke]

One additional thought - intuitively you always want to use data corrected to surface reflectance, this makes sense.  However, because Tasseled Cap is a special implementation of Principle Components Analysis (PCA), the first and second components (Brightness and Greenness) account for the most variability in the image, the 3rd component "wetness" is minor, and the 4th component contains most of the noise, i.e., haze, clouds, and other atmospheric and environmental affects that contribute to noise.  So don't use the 4th component, or use it as an indicator of noise (amount and spatial distribution).  Thus, atmospherics have less of an impact on TC analysis than when using other types of analysis.  

When using Landsat 8, WorldView 3-4 and other sensors with more than 4 bands, the lessor components (components 4-8) containing noise are even more pronounced.