Compute kNDVI using R

Question

I am performing a small test to calculate the RBF-kernel NDVI (kNDVI). kNDVI is a nonlinear generalization of the commonly used Normalized Difference Vegetation Index (NDVI). It is a novel spectral index proposed in Camps-Valls et al. 2021.

I am using Sentinel-2 L2A bands 4 (red) and 8 (NIR) without applying any scaling factor. I am trying to calculate the kNDVI using R with the code provided by the authors from their Github.

library(raster)

s2_red <- raster("C:/test_kNDVI/T31SBD_20220808T104631_B04_10m.jp2")
s2_nir <- raster("C:/test_kNDVI/T31SBD_20220808T104631_B08_10m.jp2")

sigma <- 0.15
knr <- exp(-(s2_nir-s2_red)^2/(2*sigma^2))
kndvi <- (1-knr) / (1+knr)

However, when I run the code with (sigma=0.15) I get a strange kNDVI image.

par(mfrow=c(1,2))
plot(kndvi, main="kNDVI")
plotRGB(raster::stack("C:/test_kNDVI/T31SBD_20220808T104631_TCI_10m.jp2"))

What is happening here?

Answer 1

It seems you are using Digital Numbers instead of reflectances as specified by the paper. Just divide band values by 10000 before calculating knr.

# get reflectance values
s2_red<-s2_red/10000
s2_nir<-s2_nir/10000

sigma <- 0.15
knr <- exp(-(s2_nir-s2_red)^2/(2*sigma^2))
kndvi <- (1-knr) / (1+knr)
ndvi <- (s2_nir-s2_red)/(s2_nir+s2_red)