autoFilter: A multiple-approach filtering procedure

In phenopix: Process Digital Images of a Vegetation Cover

Description

5 different filters can be applied to raw green coordinate values. Filters can be applied alone or in sequence, in this case order matters. Available filters are 'night', 'max', 'spline', 'blue' and 'mad'. See details for further explaination.

Usage

autoFilter(data, dn=c('ri.av', 'gi.av', 'bi.av'), raw.dn = FALSE, 
  brt = 'bri.av', na.fill = TRUE, filter = c("night", "spline", "max"), 
  filter.options = NULL, plot = TRUE, ...)

Arguments

data	A data.frame containing a POSIX vector
dn	The column positions or colnames for red, green and blue digital numbers, in this order. The default is suited to work with a dataframe in output from ExtractVIs
raw.dn	If TRUE you must provide raw digital numbers (range 0-255), for which position or colnames must be provided in dn. In this case gcc is calculated as g/brt where g is raw green digital number and brt is brightness computed as the sum of r + g + b. If FALSE (default) the function expects that in dn you provide colnames or col positions of relative red green and blue, respectively.
brt	The column position for brightness. Used in the filtering procedure. If you provide raw digital numbers and brt=NULL, it will be calculated from raw red green and blue dn.
na.fill	If FALSE, discarded data are filled with NA. If TRUE a call to na.approx from package zoo linearly interpolates between existing records up to a gap of 10 missing values.
filter	Character names for filters. Chose one (or more) between 'night', 'max', 'spline', 'blue' and 'mad'. The order provided in this argument determines the order of application of filters. See details for further information on filters.
filter.options	If NULL filter.options defaults to set values, otherwise they must be specified in a named list. Filter options are designed to work with a wide range of data, the user must change them with caution. See get.options to change default filter options.
plot	If TRUE a dignostic plot of the different filtering effects is returned.
...	Further options, currenlty not used

Details

This function takes a data.frame with raw digital numbers of red green and blue found in dn, converts them in the respective coordinates (example for red: rcc=r/(r+g+b)). If raw.dn=FALSE, the function takes directly color's chromatic coordinates provided in dn, i.e. realtive values. In the dataframe a POSIX vector must be provided, which will subsequently be converted in numeric day of year (doy). Afterwards, the filtering procedure starts. Filters are applied in the order provided in filter. Night filter removes records under a certain gcc value (as specified in filter.options). The default is 0.2. Blue filter is intended to remove bad images and is very aggressive. It is suggested only for very low quality images. The daily mean and standard deviation on bcc is computed and a sd threshold is computed as the quantile of standard deviations with prob = 0.05. An envelope is then computed as daily mean bcc +/- the calculated threshold. Raw data outside this envelope are discarded. The mad filter is applied according to Papale et al 2006 (biogeosciences) created to remove spikes on FLUXNET data. The max filter is based on Sonnentag et al (2012) and computes the 90% of the curve based on a three days moving window. The spline filter is based on Migliavacca et al (2011). Default values in function's arguments are suited to data in output from the function ExtractVIs. Note that computing relative greenness index within the function autoFilter, i.e., providing raw dn and with raw.dn set to TRUE will produce slightly different values than 'gi.av' in output from ExtractVIs. This results from the difference between computing pixel based brightness (and color indices) or ROI average brightness as it happens in autoFilter with raw.dn = TRUE.

Value

A multivariate zoo object with raw data and a column with gcc values after filtering. Colnames for filtered data have the same name as the applied filter.

Author(s)

Gianluca Filippa <[e-mail]>

References

Sonnentag et al.

Migliavacca et al.

Papale et al.

Examples

data(bartlett2009)
## with raw.dn ==TRUE and column position for chromatic coordinates
par(mfrow=c(3,1), oma=c(5,4,4,2), mar=rep(0,4))
filtered.data <- autoFilter(bartlett2009, dn=c(5:7), 
  filter=c('night', 'spline', 'max'),
  filter.options=NULL, raw.dn=TRUE)
## same as before but returning unfilled data 
filtered.data <- autoFilter(bartlett2009, dn=c(5:7), 
  filter=c('night', 'spline', 'max'),
  filter.options=NULL, na.fill=FALSE, raw.dn=TRUE)
## all filters in action (check the plot)
filtered.data <- autoFilter(bartlett2009, dn=c(5:7), 
  filter=c('night','blue', 'mad','spline', 'max'),
  filter.options=NULL, raw.dn=TRUE)
## check filter names
names(filtered.data)

Example output

[1] "rcc"             "gcc"             "bcc"             "brt"            
[5] "night.filtered"  "blue.filtered"   "mad.filtered"    "spline.filtered"
[9] "max.filtered"   

phenopix documentation built on May 2, 2019, 4:50 p.m.