check_input: check_input
In kongdd/whittaker: Whittaker smoother

Check input data, interpolate NA values in y, remove spike values, and set weights for NA in y and w.

1
2
3

check_input(t, y, w, QC_flag, nptperyear, south = FALSE, Tn = NULL,
  perc_wc = 0.4, wmin = 0.2, ymin, missval, maxgap, alpha = 0.02,
  ...)

`t`	Numeric vector, `Date` variable
`y`	Numeric vector, vegetation index time-series
`w`	(optional) Numeric vector, weights of `y`. If not specified, weights of all `NA` values will be `wmin`, the others will be 1.0.
`QC_flag`	Factor (optional) returned by `qcFUN`, levels should be in the range of `c("snow", "cloud", "shadow", "aerosol", "marginal", "good")`, others will be categoried into `others`. `QC_flag` is used for visualization in `get_pheno()` and `plot_phenofit()`.
`nptperyear`	Integer, number of images per year.
`south`	Boolean. In south hemisphere, growing year is 1 July to the following year 31 June; In north hemisphere, growing year is 1 Jan to 31 Dec.
`Tn`	Numeric vector, night temperature, default is null. If provided, Tn is used to help divide ungrowing period, and then get background value in ungrowing season (see details in `phenofit::backval()`).
`perc_wc`	critical percentage of good- and marginal- quality points for `wc`.
`wmin`	Double, minimum weight of bad points, which could be smaller the weight of snow, ice and cloud.
`ymin`	If specified, `ylu[1]` is constrained greater than ymin. This value is critical for bare, snow/ice land, where vegetation amplitude is quite small. Generally, you can set ymin=0.08 for NDVI, ymin=0.05 for EVI, ymin=0.5 gC m-2 s-1 for GPP.
`missval`	Double, which is used to replace NA values in y. If missing, the default vlaue is `ylu[1]`.
`maxgap`	Integer, nptperyear/4 will be a suitable value. If continuous missing value numbers less than maxgap, then interpolate those NA values by zoo::na.approx; If false, then replace those NA values with a constant value `ylu[1]`. Replacing NA values with a constant missing value (e.g. background value ymin) is inappropriate for middle growing season points. Interpolating all values by na.approx, it is unsuitable for large number continous missing segments, e.g. in the start or end of growing season.
`alpha`	Double value in `[0,1]`, quantile prob of ylu_min.
`...`	Others will be ignored.

A list object returned

t Numeric vector
y0 Numeric vector, original vegetation time-series.
y Numeric vector, checked vegetation time-series, NA values are interpolated.
w Numeric vector
Tn Numeric vector
ylu = [ymin, ymax]. w_critical is used to filter not too bad values.

If the percentage good values (w=1) is greater than 30%, then w_critical=1.

The else, if the percentage of w >= 0.5 points is greater than 10%, then w_critical=0.5. In boreal regions, even if the percentage of w >= 0.5 points is only 10%, we still can't set w_critical=wmin.

We can't rely on points with the wmin weights. Then,
y_good = y[w >= w_critical ],
ymin = pmax( quantile(y_good, alpha/2), 0)
ymax = max(y_good).