Backwards dispersion models don't work

[AlFontal]

Hi!

Thanks for wrapping HYSPLIT with such a nice R package!

I was able to get the last version running and did some test runs on dispersion models. Everything seems to work correctly for the 'forward' models, but I am unable to properly run 'backward' models. Just building on the example dispersion model:

f_dispersion_model <-
  create_dispersion_model() %>%
  add_source(
    name = "particle",
    lat = 49.0, lon = -123.0, height = 50,
    rate = 5, pdiam = 15, density = 1.5, shape_factor = 0.8,
    release_start = lubridate::ymd_hm("2015-07-01 00:00"),
    release_end = lubridate::ymd_hm("2015-07-01 00:00") + lubridate::hours(2)
  ) %>%
  add_dispersion_params(
    start_time = lubridate::ymd_hm("2015-07-01 00:00"),
    end_time = lubridate::ymd_hm("2015-07-01 00:00") + lubridate::hours(6),
    direction = "forward", 
    met_type = "reanalysis",
    met_dir = here::here("hysplit/met"),
    exec_dir = here::here("hysplit/out")
  ) %>%
  run_model()

However, if I now change to a 'backwards' model, and edit the time such that:

...
  add_dispersion_params(
    start_time = lubridate::ymd_hm("2015-07-01 00:00") - lubridate::hours(6),
    end_time = lubridate::ymd_hm("2015-07-01 00:00"),
    direction = "backward", 
...

This basically generates an empty output table.

Then I thought, okay, maybe start and end time definitions are switched in the backwards case, so I test with the following:

...
  add_dispersion_params(
    start_time = lubridate::ymd_hm("2015-07-01 00:00"),
    end_time = lubridate::ymd_hm("2015-07-01 00:00") - lubridate::hours(6),
    direction = "backward", 
...

However, again we obtain an empty output table.

Simply as a last resort, if I test simply changing the direction parameter with the original example (of course, start and end times in the future of the release start-end don't make sense in a backwards scenario, but just for testing purposes):

...
  add_dispersion_params(
    start_time = lubridate::ymd_hm("2015-07-01 00:00"),
    end_time = lubridate::ymd_hm("2015-07-01 00:00") + lubridate::hours(6),
    direction = "backward", 
...

This does give a result, which happens to be exactly the same as when we test with the forward direction.

So, given this, has anyone been able to successfully run a backwards dispersion model with splitr?

[AlFontal]

I think I might have noticed an issue in the run_model.R file, where the direction of the models when passed to run_model() is hardcoded to be 'forward' instead of using model$direction: https://github.com/rich-iannone/splitr/blob/44851122514e1296291f0e9b082326717549715f/R/run_model.R#L68-L89