setting the parameter "frequency" for the multi-year average of hourly ozone data

Question

I am trying to decompose a time series which is the monthly multi-year average of hourly ozone data. There are 288 data points (24 hours * 12 months). STL needs ts object to extract the components of time series. And ts has the parameter "frequency". As far as I know, it is the number of observations in one period. For example, it is 12 for monthly averaged temperature data. What is the frequency for my case since If I use 288

data_ts=stl(ts(data,frequency = 288),s.window = "per"))

As expected, it throws the error "series is not periodic or has less than two periods". BTW, I am aware of other methods to extract seasonality, but I also need to check the results with STL.

Best

Answer 1

Assuming you have hourly data, there are 24 periods per day, and 24*365.25 periods per year on average. Months would appear to be irrelevant for a natural phenomenon such as ozone. Similarly, weeks are irrelevant. So you just need seasonal periods of 24 and 24*265.35.

The mstl() function from the forecast package can handle multiple seasonal periods.

library(forecast)
data_ts <- mstl(msts(data, seasonal.periods = c(24, 24*365.25)))

However, if you actually have monthly data, then the frequency is 12.

data_ts <- mstl(ts(data, frequency = 12))

Answer 2

As you can see in the picture ACF, the ACF of your data clearly shows an annual seasonal trend. it peaks at yearly lag at about 12, 24, etc.

If I am behalf on you, I will use freq=12 to decompose my time series data.