A band reducer for a linear combination of bands.

Returns an anvl reducer fn(x, dims) for reduce_over(cube, fn, over = "band"): it centres each band (optional) and forms the weighted sum sum_b weights[b] * (band_b - center[b]) per pixel – a linear projection of the band vector. This is the "reduce over bands" primitive behind spectral indices, linear/logistic prediction, and PCA. For multiple outputs (e.g. the first k principal components) build one reducer per weight column and stack:

Usage

band_project(weights, center = NULL)

Arguments

weights: Per-band coefficients (length = number of bands).
center: Optional per-band centre subtracted before weighting (e.g. a PCA's column means); length must match weights.

Value

A function fn(x, dims) suitable for reduce_over() over = "band".

Details

pc <- lapply(1:3, \(i) reduce_over(cube, band_project(rot[, i], centre),
                                   over = "band"))
collect(lazy_stack(pc, along = "band"))            # (3, y, x)