1

I am using Landsat-8-OLI Multispectral images to perform Principal Component Analysis. The aim is to detect noise and reduce dimensionality contained by the spectral bands. The resultant Principal components having new data dimensions with minimum noise will be used in biomass study which involves Multi Regression Analysis. The aim is to ascertain whether there is any potential relationship between biomass samples and PCA components.

The OLI sensor consist of 7 Multi spectral bands including ultra blue (B1) , wavelength spanned between 0.43 - 0.45 micrometer.

https://landsat.usgs.gov/what-are-band-designations-landsat-satellites

Should I exclude Band 1 - Ultra Blue (coastal/aerosol) and run PCA analysis on remaining Six bands (B, G, R, NIR, SWIR 1 & SWIR 2)?

  • Can I ask why do you think you should? – Albert Jul 4 '17 at 12:53
  • Because this band is chiefly meant to analyse aerosol (dust and smoke) in contrast to my scope of work for which rest of the bands contain most of the information. – Ben Jul 5 '17 at 6:14
  • Ok, and when you say biomass you mean biomass from vegetation? – Albert Jul 5 '17 at 6:45
1

Yes. Discard the coastal blue band since that will introduce variation that is not relevant to your study. NIR and Red are probably the most important. Consider investigating the closely related, more physically-based, Tasseled Cap transformation. Since PCA is statistically-based (not physically-based), it rotates through the n-dimensions of your data set, trying to find vectors that maximize variation based on statistical assumptions about the data. Ensure that the data (image extent and spectral composition) you feed to PCA are appropriate since you are hoping that a statistical approach will provide physically relevant results.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.