Time series and stl in R: only one-dimensional series are allowed

Question

I am doing an analysis of the hourly rainfall in a file that is disorganized. However, I managed to clear it and save it in a data frame (called CA1), which takes the form as follows:

Station_ID Guage_Type Lat Long Date Time_Zone Time_Frame H0 H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16 H17 H18 H19 H20 H21 H22 H23 1 4457700 HI 41.52 124.03 1948-07-01 8 LST 0 0 0 0 0 0 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0 0 0 0 0 0 0 0 0 0 0 0 2 4457700 HI 41.52 124.03 1948-07-05 8 LST 0 1 1 1 1 1 2.0000000 2.0000000 2.0000000 4.0000000 5.0000000 5.0000000 4 7 1 1 0 0 10 13 5 1 1 3 3 4457700 HI 41.52 124.03 1948-07-06 8 LST 1 1 1 0 1 1 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0 0 0 0 0 0 0 0 0 0 0 0 4 4457700 HI 41.52 124.03 1948-07-27 8 LST 3 0 0 0 0 0 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0 0 0 0 0 0 0 0 0 0 0 0 5 4457700 HI 41.52 124.03 1948-08-01 8 LST 0 0 0 0 0 0 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0.0000000 0 0 0 0 0 0 0 0 0 0 0 0 6 4457700 HI 41.52 124.03 1948-08-17 8 LST 0 0 0 0 0 0 0.3888889 0.3888889 0.3888889 0.3888889 0.3888889 0.3888889 6 1 0 0 0 0 0 0 0 0 0 0 

Where H0 - H23 - 24 hours a day (line)

Using only CA1 (the data frame above), I take every day (row) of 24 points and transpose it vertically and combine the remaining days (rows) with one variable, which I call dat1:

  > dat1[1:48,] H0 H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16 H17 H18 H19 H20 H21 H22 H23 H0 H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13 H14 H15 H16 H17 H18 H19 H20 H21 H22 H23 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 2 2 2 4 5 5 4 7 1 1 0 0 10 13 5 1 1 3 

Using the variable dat1, I enter it as an argument to get the time series data:

 > rainCA1 <- ts(dat1, start = c(1900+as.POSIXlt(CA1[1,5])$year, 1+as.POSIXlt(CA1[1,5])$mon), frequency = 24) 

A few notes:

 >dim(CA1) [1] 5636 31 >length(dat1) [1] 135264 

Thus, 5636 * 24 (common data points [24] per row) = 135264 points. The length (rainCA1) is consistent with the above points. However, if I put an end to the ts function, for example

 >rainCA1 <- ts(dat1, start = c(1900+as.POSIXlt(CA1[1,5])$year, 1+as.POSIXlt(CA1[1,5])$mon), end = c(1900+as.POSIXlt(CA1[5636,5])$year, 1+as.POSIXlt(CA1[5636,5])$mon), frequency = 24) 

I get 1134 of the total length of the points, where I miss a lot of data. I assume this is due to the fact that the dates are not consecutive, and since I only use the month and year as an argument to the starting point.

Continuing, in my opinion, the correct path, using the first ts calculation without end argument, I will put it as input for stl:

 >rainCA1_2 <-stl(rainCA1, "periodic") 

Sorry, I get an error:

 Error in stl(rainCA1, "periodic") : only univariate series are allowed 

What I don’t understand and how to do it. However, if I go back to the ts function and provide the final argument, stl works fine without any errors.

I have researched in many forums, but not one (or as far as I know) gives a good solution for getting hourly data attribute data. If anyone can help me, I will really appreciate it. Thanks!

Answer 1

This error is the result of the form of your data. Try > dim(rainCA1) ; I suspect that he gave something like > [1] 135264 1 . Replace rainCA1 <- ts(dat1 ... with rainCA1 <- ts(dat1[[1]] ... and it should work.

Whether this is correct is interesting ... It seems to me that your first order is to receive your data in a consistent format. Make sure ts() gets the correct input. Check out the exact ts specification.

ts() does not interpret date and time formats. ts() requires consecutive data points with a fixed interval. It uses the main counter and the lowest counter (of which frequency fits into one main counter). For example, if your data is hourly, and you expect seasonality at a daily level, frequency is 24. start and end , therefore, primarily cosmetic: start simply indicates t (0) for the main counter, while end means t (end).

Answer 2

I tried to explain the recording method with a very simple example in order to avoid similar errors in another question related here:

stl () decomposition will not accept one-dimensional object ts?

Answer 3

One solution found is time_series_var <- ts(data[, c("var_of_interest")]) , and then time_series_var <- ts(as.vector(time_series_var)) , and then the error related to the one-dimensional disappears, because the dimensions are now correct .