Example of using grouped linear interpolation

linear-interp-spectra.R

library(tidyverse)
library(reshape2)
library(pbapply)

n_samples <- 100
n_wavelength <- 400

wavelengths <- matrix(nrow = n_samples, ncol = n_wavelength)
reflectances <- matrix(nrow = n_samples, ncol = n_wavelength)
for (i in 1:n_samples) {
  wavelengths[i, ] <- sort(runif(n_wavelength, min = 1, max = 1000))
  reflectances[i, ] <- rnorm(1, sd = .2) + 
    sin(.005 * wavelengths[i, ]) + 
    rnorm(n_wavelength, sd = .03)
}

wv_df <- reshape2::melt(wavelengths, varnames = c("sample", "wavelength")) %>%
  as_tibble %>%
  rename(band_index = wavelength, 
         wavelength = value)

rf_df <- reshape2::melt(reflectances, varnames = c("sample", "wavelength")) %>%
  as_tibble %>%
  rename(band_index = wavelength, 
         reflectance = value)

d <- full_join(wv_df, rf_df)

d

d %>%
  ggplot(aes(x = wavelength, y = reflectance, group = sample)) + 
  geom_path()

wavelength_vals <- seq(min(wavelengths), max(wavelengths),
                       length.out = n_wavelength)

interpolations <- d %>%
  split(.$sample) %>%
  pblapply(function(df) {
    interpolated_reflectance <- approx(x= df$wavelength, 
                                       y = df$reflectance, 
                                       xout = wavelength_vals)
    tibble(sample = unique(df$sample), 
           band_index = seq_along(wavelength_vals), 
           wavelength = interpolated_reflectance$x, 
           reflectance = interpolated_reflectance$y)
  }) %>%
  bind_rows

interpolations %>%
  ggplot(aes(x = wavelength, y = reflectance, group = sample)) + 
  geom_path()