No support for *.op.gz files in reformat_GSOD()?
taraskaduk opened this issue · 13 comments
I may be confusing something here, but I think that a while back, reformat_GSOD() worked with *.op.gz files obtained from NOAA ftp.
It seems like this is no longer the case. Is something changed?
This is not an issue per se, it's a confusion. I couldn't rerun my code from a year ago, and I'm trying to see what changes do I need to make to the code to make it work.
You're not crazy. I'm sorry, it did. When NCEI changed the formats of the data that they served I had to update GSODR a few months ago and reformat_GSOD() uses the same code behind the scenes that get_GSOD() uses. So it no longer supports the .op.gz format because NCEI is serving .gz files of .csv files.
I figured as much after poking around this repo. Thanks. I repointed my code to load data from the other source and not from ftp, but ran into a different issue, and now I'm just trying to simply use get_GSOD() for consistency purposes (my previous workflow avoided get_GSOD() and downloaded full year archives instead, as get_GSOD() was erroring out and I couldn't figure out why)
I remember my problem with get_GSOD() now. It is very long to run for many stations (global scale) for many years (e.g. 10). I found that curl_download() of the entire year archive of all the data is a much faster route.
In the old process, when I downloaded data from ftp, each .op.gz file was titled as stationid-year.op.gz. In the current source that I assume you point to, the .csv files are named after a station, no year. I now just need to figure out the renaming process so that files from different years don't overwrite each other.
get_GSOD() would be a much more straight-forward way to pull the data, it's a shame it doesn't work well for me.
Ah, interesting. It's faster to download the entire .gz and sort it out after? How many stations are you fetching at once?
I hadn't considered this case. I though either download all or just a few selected not many selected.
My case is exactly in between. I pull 10 years worth of data for every station within a 25 km radius of every major city in the world.
What I do instead is download full year archives, then delete the files of the stations that I don't need, and then process the remaining station files with reformat_GSOD
Here is the function that does it all for me:
get_weather <- function(yrs = seq(year(today()) - 11, year(today())),
stns = stations_v) {
for (yr in yrs) {
file <- paste0(yr, '.tar.gz')
destfile <- paste0('data/gsod/', file)
if (!file.exists(destfile)) {
link <- paste0('https://www.ncei.noaa.gov/data/global-summary-of-the-day/archive/', file)
curl::curl_download(link, destfile)
}
untar(destfile, exdir = paste(tempdir(), yr, sep = "/"))
}
# Go through all unpacked files, decide what to remove and what to keep
# based on the stations of interest
files_all <- list.files(path = tempdir(), pattern = "^.*\\.csv$", recursive = TRUE, full.names = FALSE)
# Get a cartesian join of all stations of interest and all years.
files_stations <-
purrr::cross(list(x1 = paste0(yrs, "/"), x2 = paste0(stns, ".csv"))) %>%
purrr::map(purrr::lift(paste0)) %>%
as_vector()
files_keep <- subset(files_all, files_all %in% files_stations)
# Transform weather data ----------------------------------------------------
out <- GSODR::reformat_GSOD(file_list = paste(tempdir(), files_keep, sep = "/"))
unlink(tempdir(), force = TRUE, recursive = TRUE)
out
}
Here is the repo, the code snippet above is from functions.R
Awesome, thanks! I'll have a look and see if I can improve the package.
Sweet! Let me know if I can contribute in any way.
I've updated the internal functionality to check how many requests are being made. If the number of stations is greater than 10, GSODR will download the entire global annual file and sort out the needed files locally. If there are less than 10 stations, then it will download each requested station individually.
I did a few tests to check how many individual requests were faster vs downloading the whole. The number is not exact due to things I can't control (Internet), but this should help in most cases to make it faster to request large numbers of stations that are not ALL stations or just a few.
Line 155 in b1b6a5d
Sweet! I'll make sure to chuck my extra piece of code to pull full archives for my analysis on my next update.
Thank you for letting me know about the bottleneck. You OK if I add you as a contributor to the package for the ideas/input?
I would be honored! To be honest, I'd do a pull request on this item (and a couple of others), but I've never done a PR before, and was afraid to mess things up 🤷♂️