raster and oceancolor L2 netcdf data

You can't georeference these data without remapping the data, essentially
treating the pixels as points. They have no natural regular grid form,
except possibly a unique satellite-perspective one. The data are in 2D
array form, but they have explicit "geolocation arrays", i.e. a longitude
and latitude for every cell and not based on a regular mapping.

R does not have tools for this directly from these data, though it can be
treated as a resampling or modelling problem.
You can use raster to get at the values of the locations easily enough,
here's a couple of plotting options in case it's useful:

u <- "
https://github.com/dmwarn/Tethys/blob/master/A2016244185500.L2_LAC_OC.x.nc?raw=true
"
f <- basename(f)
download.file(u, f, mode = "wb")

library(raster)
## use raster to treat as raw point data, on long-lat locations
rrs <- raster(f, varname = "geophysical_data/Rrs_412")
longitude <- raster(f, varname = "navigation_data/longitude")
latitude <- raster(f, varname = "navigation_data/latitude")

## plot in grid space, and add the geolocation space as a graticule
plot(rrs)
contour(longitude, add = TRUE)
contour(latitude, add = TRUE)

## raw scaling against rrs values
scl <- function(x) (x - min(x, na.rm = TRUE))/diff(range(x, na.rm = TRUE))
plot(values(longitude), values(latitude), pch = ".", col =
topo.colors(56)[scl(values(rrs)) * 55 + 1])

## ggplot
library(ggplot2)
d <- data.frame(x = values(longitude), y = values(latitude), rrs =
values(rrs))
ggplot(d, aes(x = x, y = y, colour = rrs)) + geom_point(pch = ".")

## might as well discard the missing values (depends on the other vars in
the file)
d <- d[!is.na(d$rrs), ]
ggplot(d, aes(x = x, y = y, colour = rrs)) + geom_point(pch = 19, cex = 0.1)

There are some MODIS and GDAL based packages that might be of use, but I
haven't yet seen any R tool that does this remapping task at scale. (I
believe the MODIS tools and the best warping tools in GDAL use thin-plate
spline techniques).

 Some applications would use the observations as points (i.e. the ocean
colour L3 bins as a daily aggregate from L2) and others use a direct
remapping of the data as an array, for say high-resolution sea ice imagery.

You might not need to do anything particularly complicated though, what's
the goal?

Cheers, Mike.

raster and oceancolor L2 netcdf data

Thread (8 messages)