Open bart1 opened 3 weeks ago
We tend to refer to the volume radar data in 5 minutes intervals but I think sometimes there is a bit of ambiguity to what time the radar data refers. The time in the data for in cz
is for example 4 minutes after the reference time. Although the scans start pretty much directly at the reference time. I would need to look a bit more into this
require(getRad)
#> Loading required package: getRad
pv<-get_pvol("czska",t<- as.POSIXct(Sys.Date())+lubridate::hours(3))
t
#> [1] "2024-10-15 03:00:00 UTC"
pv$datetime
#> [1] "2024-10-15 03:04:12 UTC"
bioRad::attribute_table(pv, select=c("where.elangle","what.endtime","what.starttime"))[,-4]
#> what.endtime what.starttime where.elangle
#> 030412 030345 0.1
#> 1 030343 030316 0.5
#> 2 030314 030251 0.9
#> 3 030250 030227 1.3
#> 4 030225 030202 1.7
#> 5 030201 030138 2.2
#> 6 030136 030114 3.2
#> 7 030112 030100 4.5
#> 8 030058 030045 6.3
#> 9 030041 030030 8.7
#> 10 030027 030016 13.7
#> 11 030013 030003 21.6
Now time can be a vector rounded to the nearest 5 minutes and is checked, parallelization / request optimization I will leave for later
currently time should be a POSIXct rounded to 5 minutes we might consider also allowing for intervals and multiple timestamps. The individual functions now only download one file at a time and do not parallelize requests.