Fill satellite gaps with the model equivalent — fill_satellite

Satellite ocean colour is missing wherever cloud, ice, or low sun angle blocked the view, and those gaps are not random — they cluster in exactly the seasons and latitudes that are often most interesting. This substitutes the model value at each missing cell, giving a gap-free series without discarding the observed values where they exist.

Usage

fill_satellite_gaps(satellite, model, vars, rescale = FALSE)

Arguments

satellite: an sf POINT object of satellite data from accessEnvDat()
model: an sf POINT object of the model equivalent, over the same period. It need not be on the same grid: values are taken from the nearest model cell within each time step.
vars: named character vector mapping satellite columns to model columns, e.g. c(CHL = "CHL_MODEL"). When unnamed, the same name is assumed in both.
rescale: match the model's mean and variance to the satellite's before filling, computed per time step over the cells where both exist

Value

satellite with gaps filled, plus a <var>_source factor column per variable recording "satellite" or "model"

What this does and does not do

The two sources are different measurements of the same quantity, not interchangeable ones. Satellite chlorophyll is an optical retrieval at 4 km; model chlorophyll is simulated at 0.25 degrees. Splicing them produces a series whose statistical properties change wherever the source changes — usually a step in both variance and bias at the seam.

So this is offered with two safeguards rather than as a silent default:

Nothing is rescaled by default. The model values go in as they are, so a discontinuity at the seam is visible rather than smoothed over. Set rescale = TRUE to match the model's mean and variance to the satellite's over the cells where both exist, which reduces the step at the cost of changing the model values.
A companion <var>_source column records which source each value came from, so a model can include it, and an analysis can check whether a result rests on observed or simulated cells.

Examples

if (FALSE) { # \dontrun{
chl_sat <- accessEnvDat(vars = "CHL", years = 2010, months = 1:12, bounding_box = bb)
chl_mod <- accessEnvDat(vars = "CHL_MODEL", years = 2010, months = 1:12,
                        bounding_box = bb)

filled <- fill_satellite_gaps(chl_sat, chl_mod, c(CHL = "CHL_MODEL"))
table(filled$CHL_source)
} # }