Using preprocessed-only birdflow objects as the bf argument for BirdFlowR functions

[slager]

Using preprocessed-only birdflow objects as the bf argument for BirdFlowR functions is currently possible, but isn't explicitly supported and does not always produce expected results. However, it is sometimes very useful to be able to do so. For example, when wanting to extract timestep or CRS info created during the preprocessing.

The overall user experience related to using preprocessed-only objects could be improved with some combination of documentation, tests, warnings, or errors.

The following example from lookup_timestep() will soon be fixed by broader changes being made by @ethanplunkett, but similar situations might arise with other functions.

# goal: use existing birdflow object to lookup timesteps

bf <- BirdFlowR::preprocess_species('paibun', res = 1000, hdf5 = FALSE, skip_quality_checks = TRUE)
timesteps <-
  structure(
    c(1325376000, 1293408000, 1283040000),
    class = c("POSIXct",
              "POSIXt"),
    tzone = "UTC"
  )
tail(bf$dates, 1) #53
# timestep #53 in preprocessed model creates non-ascending `breaks` vector, 
# which causes error in findIntervals
lookup_timestep(timesteps, bf) # error, see above note

BirdFlowR/R/lookup_timestep.R

Line 64 in 43beefa

x <- findInterval(py, vec = breaks, all.inside = TRUE)

bf2 <- BirdFlowModels::amewoo
tail(bf2$dates, 1) #52
lookup_timestep(timesteps, bf2) # no error

[ethanplunkett]

I think just about everything we'd expect to work now works:

• lookup_timestep() now works as expected with date input as of (Preprocess 2022 #148)
• lookup_timetep_sequence() works identically to a standard model.
• Any use of the model as a spatial reference works identical to a full model.

A few things are still different between a preprocessed model relative to a standard model:

• Lacks marginals and transitions
• Has great circle distances
• First distribution and dynamic mask are duplicated
• First $date row is duplicated as last with timestep 53, week 1, and same date as first row.
• Metadata related to fitting and importing is missing.

The implications of those differences:

• n_distr() returns 53.
• n_timesteps() returns 52 (somewhat arbitrary)
• n_transitions() returns 52
• is_cyclical() returns TRUE (somewhat arbitrary)
• lookup_timestep("all") will return 1:53 (somewhat arbitrary but see next item)
• get_distr() and get_dynamic_mask() will return all 53 with default "all" (passed to lookup_timetep())