Create Regular time series from an irregular (as.Date) time series with frequency=23

Question

I have the following problem in R. I would like to create a ts() object (i.e. a regular time series) from a irregular time series (i.e. a list of dates and data values).

You can reproduce the problem with the following data set and R script:

# dput(dd) result    
dd <- structure(list(NDVI = structure(c(14L, 4L, 11L, 12L, 20L, 17L, 
    5L, 7L, 21L, 23L, 25L, 19L, 15L, 9L, 3L, 24L, 2L, 6L, 22L, 16L, 
    13L, 18L, 10L, 8L, 1L), .Names = c("1", "2", "3", "4", "5", "6", 
    "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", "17", 
    "18", "19", "20", "21", "22", "23", "24", "25"), .Label = c("0.4186", 
    "0.5452", "0.5915", "0.5956", "0.6010", "0.6860", "0.6966", "0.7159", 
    "0.7161", "0.7264", "0.7281", "0.7523", "0.7542", "0.7701", "0.7751", 
    "0.7810", "0.7933", "0.8075", "0.8113", "0.8148", "0.8207", "0.8302", 
    "0.8305", "0.8369", "0.9877"), class = "factor"), DATUM = structure(c(11005, 
    11021, 11037, 11085, 11101, 11117, 11133, 11149, 11165, 11181, 
    11197, 11213, 11229, 11245, 11261, 11277, 11293, 11309, 11323, 
    11339, 11355, 11371, 11387, 11403, 11419), class = "Date")), .Names = c("NDVI", 
    "DATUM"), row.names = c("1", "2", "3", "4", "5", "6", "7", "8", 
    "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", 
    "20", "21", "22", "23", "24", "25"), class = "data.frame")

require(zoo)
dd$DATUM <- as.Date(dd$DATUM,"A%Y%j") # Ayear,julianday
z <- zoo(dd$NDVI,dd$DATUM,frequency=23)
z  # this is a regular time series with a frequency=23 and start=c(2000,1)
# there are 5 measurements in 2000 (2 jan, 1 feb, 2 apr) for which no data is available 
# this should be marked as an NA is the final regular time series
ts.z <- as.ts(z,start=c(2000,1),frequency=23)

But this does not work, as I obtain a very long regular time series containing daily time steps. I would like to obtain a ts object with a frequency=23 correctly indicating the position for which data is not available as NA.

I have been trying everything based on the example listed here for yearly data Convert a irregular time series to a regular time series

but it does not work for data with a frequency of 23 (i.e. 23 values a year). I think I could solve it by avoiding to set dd$DATUM as.Date() but as an zoo object that can be ordered as a time series with 23 values a year.

Any ideas?

Thanks for your help

Answer 1

23 does not evenly divide into the number of days in a year so you will have to synthesize your own time scale such that each year is divided into 23 equal pieces. Convert dd (the version that has "Date" class times) to zoo and create a new series based on a new scale made up of the year plus a fraction. Finally convert that to a ts series:

library(zoo)
z <- zoo(as.numeric(as.character(dd[[1]])), dd[[2]]) 
lt <- unclass(as.POSIXlt(time(z)))
yr <- lt$year + 1900
jul <- lt$yday
delta <- min(unlist(tapply(jul, yr, diff))) # 16
zz <- aggregate(z, yr + jul / delta / 23)

as.ts(zz)

giving:

Time Series:
Start = c(2000, 4) 
End = c(2001, 7) 
Frequency = 23 
 [1] 0.7701 0.5956 0.7281     NA     NA 0.7523 0.8148 0.7933 0.6010 0.6966
[11] 0.8207 0.8305 0.9877 0.8113 0.7751 0.7161 0.5915 0.8369 0.5452 0.6860
[21] 0.8302 0.7810 0.7542 0.8075 0.7264 0.7159 0.4186