如何分析R中的不规则时间序列

Question

I have a zoo time series in R:

d <- structure(c(50912, 50912, 50912, 50912, 50913, 50913, 50914, 
50914, 50914, 50915, 50915, 50915, 50916, 50916, 50916, 50917, 
50917, 50917, 50918, 50918, 2293.8, 2302.64, 2310.5, 2324.02, 
2312.25, 2323.93, 2323.83, 2338.67, 2323.1, 2320.77, 2329.73, 
2319.63, 2330.86, 2323.38, 2322.92, 2317.71, 2322.76, 2286.64, 
2294.83, 2305.06, 55.9, 62.8, 66.4, 71.9, 59.8, 65.7, 61.9, 67.9, 
38.5, 36.7, 43.2, 30.3, 42.4, 33.5, 48.8, 52.7, 61.2, 30, 41.7, 
50, 8.6, 9.7, 10.3, 11.1, 9.2, 10.1, 9.6, 10.4, 5.9, 5.6, 6.6, 
4.7, 6.5, 5.2, 7.5, 8.1, 9.5, 4.6, 6.4, 7.7, 9.29591864400155, 
10.6585128174944, 10.4386464748912, 11.5738448647708, 10.9486074772952, 
10.9546547052814, 10.3733963771546, 9.15627378048238, 8.22993822910891, 
5.69045896511178, 6.95269658370746, 7.78781665368086, 7.20089569039135, 
4.9759716583555, 8.99378907920762, 10.0924594632635, 10.3909638115674, 
6.28203685114275, 9.16021859457356, 7.56829801052175, 0.695918644001553, 
0.9585128174944, 0.138646474891241, 0.473844864770827, 1.74860747729523, 
0.854654705281426, 0.773396377154565, -1.24372621951762, 2.32993822910891, 
0.0904589651117833, 0.352696583707458, 3.08781665368086, 0.700895690391349, 
-0.224028341644497, 1.49378907920762, 1.99245946326349, 0.890963811567351, 
1.68203685114275, 2.76021859457356, -0.131701989478247), .Dim = c(20L, 
6L), .Dimnames = list(NULL, c("station_id", "ztd", "zwd", "iwv", 
"radiosonde", "error")), index = structure(c(892094400, 892116000, 
892137600, 892159200, 892180800, 892245600, 892267200, 892288800, 
892332000, 892353600, 892375200, 892418400, 892440000, 892461600, 
892504800, 892526400, 892548000, 892591200, 892612800, 892634400
), class = c("POSIXct", "POSIXt")), class = "zoo")

I want to perform some of the analyses that the ts package allows me to do, such as decomposing the time-series into the trend and seasonality, and looking at the auto-correlation function. However, trying to do any of these gives an error of: Error in na.fail.default(as.ts(x)) : missing values in object .

Looking into this in more depth, it seems that all of these functions work on ts objects that have, by definition, regularly-spaced observations. My observations aren't, so I end up with a lot of NA s and everything fails.

Is there a way to analyse the irregular time-series in R? Or do I need to convert them to be regular somehow? If so, is there a simple way to do this?

Answer 1

I have analysed such irregular data in the past using an additive model to "decompose" the seasonal and trend components. As this is a regression-based approach you need to model the residuals as a time series process to account for lack of independence in the residuals.

I used the mgcv package for these analysis. Essentially the model fitted is:

require(mgcv)
require(nlme)
mod <- gamm(response ~ s(dayOfYear, bs = "cc") + s(timeOfSampling), data = foo,
            correlation = corCAR1(form = ~ timeOfSampling))

Which fits a cyclic spline in the day of the year variable dayOfYear for the seasonal term and the trend is represented by timeOfSampling which is a numeric variable. The residuals are modelled here as a continuous-time AR(1) using the timeOfSampling variable as the time component of the CAR(1). This assumes that with increasing temporal separation, the correlation between residuals drops off exponentially.

I have written some blog posts on some of these ideas:

1. Smoothing temporally correlated data
2. Additive modelling and the HadCRUT3v global mean temperature series

which contain additional R code for you to follow.