How to rearrange time series data in 10 days time window in R?-CodePudding

I am currently working on a timeseries streamflow data range from 2002-10-04 to 2012-10-28, what I need to do is to develop regression models using 10-day time window data. To be more specific, I need to use Oct-04 to Oct 13 across all years from 2002 to 2012 to build a regression model. Then I need to use Oct-05 to Oct 14 across all years from 2002 to 2012 to build another regression model, then use Oct-06 to Oct 15 across all years to build next model and so on repeatedly till the end.

This is how my data look like.

> head(CFbasin)
        Data      Qkl     Qllt     Qhaw      Qdp     Qlit
1 2002-10-04 19.25546 8.353470 4.502379 2.217209 1.985011
2 2002-10-05 19.56694 8.126935 4.615646 1.622555 1.628219
3 2002-10-06 19.73684 7.560598 4.389111 1.251605 1.492298
4 2002-10-07 18.12278 7.079212 3.992675 1.158159 1.413011
5 2002-10-08 18.12278 6.824360 3.794457 1.070377 1.393189
6 2002-10-09 17.83961 6.739409 3.369705 1.073208 1.353545

> tail(CFbasin)
           Data      Qkl     Qllt      Qhaw      Qdp     Qlit
3673 2012-10-23 24.89051 16.67862  8.608321 2.477724 1.432832
3674 2012-10-24 25.00378 16.48040 11.638224 2.820358 1.393189
3675 2012-10-25 25.37189 16.99011  7.758816 3.001586 1.322397
3676 2012-10-26 26.07982 16.87684  6.484558 2.814695 1.279921
3677 2012-10-27 27.41071 17.07506  4.813864 3.086536 1.228951
3678 2012-10-28 28.88318 17.16001  5.635052 3.114853 1.220456

I only tried once and this is my code:

CFbasin %>% filter(month(Date) == 10 & day(Date) >= 4 & day(Date) <= 14)

this allows me to get all the data within Oct 4 to Oct13 from 2002 to 2012, and then conduct linear regression. But I am not sure how to have it work on the whole dataset and then conduct linear regressions, I am considering for loop and function rollapply(), but really unclear about how to arrange my dataset. Any suggestion and recommendation will be really appreciated, thank you in advance!

CodePudding user response：

Here is an example using the data in the Note at the end and a width of 5.

library(zoo)
coefs <- function(x) coef(lm.fit(cbind(1, x[, -1]), x[, 1]))
rollapplyr(CFbasin[, -1], 5, coefs, by.column = FALSE, fill = NA)

Added

This uses NA for the first output 4 rows and then the rows that correspond to rows 1:5 of all years to form the next regression, rows 2:6 of all years to form the regression after that and so on. Dec 31st is not used in leap years.

yday <- as.POSIXlt(CFbasin$Data)$yday
coefs <- function(ix) {
  x <- CFbasin[yday %in% ix, -1]
  if (NROW(x) == 0) NA else coef(lm(Qkl ~., x))
}
rollapplyr(0:364, 5, coefs, fill = NA)

Note

Lines <- "
        Data      Qkl     Qllt     Qhaw      Qdp     Qlit
1 2002-10-04 19.25546 8.353470 4.502379 2.217209 1.985011
2 2002-10-05 19.56694 8.126935 4.615646 1.622555 1.628219
3 2002-10-06 19.73684 7.560598 4.389111 1.251605 1.492298
4 2002-10-07 18.12278 7.079212 3.992675 1.158159 1.413011
5 2002-10-08 18.12278 6.824360 3.794457 1.070377 1.393189
6 2002-10-09 17.83961 6.739409 3.369705 1.073208 1.353545"
CFbasin <- read.table(text = Lines)