added code to extract ensemble forecasts #25
Conversation
for each station
|
|
dlebauer
force-pushed
the
ensemble_forecasts
branch
from
5ce96b9
to
f334f5f
Compare
|
Thanks David! I'll review this and then either implement it as part of the |
| data(station_info, package = 'azmetr') | ||
|
|
||
| library(gdalcubes) | ||
| # remotes::install_github("neon4cast/gefs4cast") |
There was a problem hiding this comment.
This one doesn't seem to exist, so I assume it's the eco4cast/gefs4cast one
| # remotes::install_github("neon4cast/gefs4cast") |
| ```{r} | ||
| library(dplyr) | ||
| ## set path for gdal binary | ||
| Sys.setenv("GDAL_BIN"="/usr/bin/") |
There was a problem hiding this comment.
My gdal was in usr/local/bin so I didn't have to set this. I think usr/local/bin is the default homebrew install location. Ironically, my problem was with my location of bash! This package assumes it's at /usr/bin/bash and mine is /bin/bash. I submitted an issue (eco4cast/gefs4cast#7) and I'll try again once it's addressed.
|
gefs4cast is buggy, can't get running for now. Opened issue: eco4cast/gefs4cast#7 |
There was a problem hiding this comment.
I had a lot of issues with gdal and geos system libraries and eventually gave up getting it to work on my laptop and instead tried out rocker/geospatial. Everything except the animated plot works with that image, so if we do implement GEFS-based QA we can use that as a base image for deployment.
|
|
||
| library(gdalcubes) | ||
| # remotes::install_github("neon4cast/gefs4cast") | ||
| #remotes::install_github('eco4cast/gefs4cast') |
There was a problem hiding this comment.
The renv way to do this:
| #remotes::install_github('eco4cast/gefs4cast') | |
| #renv::install('eco4cast/gefs4cast') |
| ```{r} | ||
| library(dplyr) | ||
| ## set path for gdal binary | ||
| Sys.setenv("GDAL_BIN"="/usr/bin/") |
There was a problem hiding this comment.
New default in dev version of gefs4cast is to respect PATH so this probably shouldn't be set.
| # may need to adjust for time zones? | ||
| # yesterday's forecast starts today? | ||
|
|
||
| gefsdir <- fs::dir_create(file.path("gefs", date)) |
There was a problem hiding this comment.
gefs_cog() will create the dir.
| gefsdir <- fs::dir_create(file.path("gefs", date)) | |
| gefsdir <- file.path("gefs", date) |
| gefs_cog(gefsdir, | ||
| ens_avg = FALSE, | ||
| max_horizon = 48, # get next 48 h. might only need 24h? | ||
| date = date) |> | ||
| system.time() |
There was a problem hiding this comment.
these are big files, even with compression. I think we would quickly blow through our GitHub Actions allocation with this and would need to figure out another way to deploy this.
| ```{r} | ||
| ## create statistical summaries by station, variable, and time | ||
|
|
||
| ens_stats <- ens_long %>% |
There was a problem hiding this comment.
What is ens_long? Looks like a pivot_longer() step is missing. I think probably:
| ens_stats <- ens_long %>% | |
| library(tidyr) | |
| ens_long <- | |
| ens_forecast %>% | |
| select(-geometry) %>% | |
| pivot_longer(APCP:VGRD, names_to = "var") | |
| ens_stats <- ens_long %>% |
|
|
||
| # above plot but saved as one file per station, using a for loop | ||
| plots <- list() | ||
| for(stn in unique(ens_stats$meta_station_id)){ |
There was a problem hiding this comment.
| for(stn in unique(ens_stats$meta_station_id)){ | |
| for(stn in unique(ens_stats$meta_station_name)){ |
| ucl_99 = quantile(value, 0.995), | ||
| lcl_99 = quantile(value, 0.005)) %>% | ||
| ungroup() %>% | ||
| mutate(id = paste0(meta_station_name, var)) |
There was a problem hiding this comment.
I needed to add a mutate to parse time to get the plots to work.
| mutate(id = paste0(meta_station_name, var)) | |
| mutate(id = paste0(meta_station_name, var), time = lubridate::ymd_hms(time)) |
| raster_cube(gefs_cube, v) |> | ||
| gdalcubes::select_bands("TMP") |> | ||
| animate( col = viridisLite::viridis, nbreaks=100, | ||
| fps=10, save_as = "temp.gif") |
There was a problem hiding this comment.
Requires packge gifski which requires some system library that isn't in rocker/geospatial. Cool idea, but probably won't implement in the qa/qc report.
This code