EITrans::EITrans

EITrans::EITrans is the main function for ensemble forecast calibration with the empirical inverse transform function method.

EITrans(
  ens,
  ens_times,
  ens_flts,
  ens_times_train,
  ens_times_test,
  obs,
  deltas,
  infinity_estimators,
  multipliers,
  circular_ens = F,
  pre_sorted = F,
  save_intermediate = F,
  optimize_lead_time = F
)

Arguments

ens	A 4-dimensional array for ensemble forecasts. Dimensions should be [stations, times, lead times, members].
ens_times	A vector for ensemble forecast times.
ens_flts	A vector for ensemble forecast lead times.
ens_times_train	Training times from the ensemble forecast times.
ens_times_test	Testing times from the ensemble forecast times.
obs	A 3-dimensional array for observations that correspond to the ensemble forecasts. Dimensions should be [stations, times, lead times].
deltas	A vector of deltas to experiment. It can also be a list with left and right to explicitly control deltas for edges.
infinity_estimators	A vector of values to experiment for estimating the ensemble spread. Or a list with named members left and right for different left/right infinity estimators.
multipliers	A vector of values to experiment for adjusting the ensemble member offset.
circular_ens	Whether the ensemble forecast variable is circular.
pre_sorted	Whether the ensemble members are presorted.
save_intermediate	Whether to save intermediate data.
optimize_lead_time	Whether to calibrate each forecast lead time respectively. Theoretically, this would lead to better calibration results, but also significantly more computation because calibration needs to be evaluated per lead time.

Value

A list with the calibrated ensemble forecasts and intermediate results.

Details

This function uses foreach parallel mechanism. Parallelization is handled by users creating the parallel backend. Please see examples. It is recommended to use doSNOW for handling the parallel backend.

Examples

if (FALSE) {
# Use doSNOW to launch parallel backend
library(doSNOW)
cl <- snow::makeCluster(4) # Use 4 cores
registerDoSNOW(cl)

# If you know a host file, you can also launch on remote server
nodefile <- Sys.getenv("PBS_NODEFILE")
nodes <- readLines(nodefile)
cl <- makeCluster(nodes, type = "SOCK")
registerDoSNOW(cl)

eitrans_results <- EITrans(
  ens = ens$analogs,
  ens_times = ens$test_times,
  ens_flts = ens$flts,

  ens_times_train = ens$test_times[1:(test_start - 366)],
  ens_times_test = ens$test_times[test_start:test_end],

  obs = ens$obs_aligned,

  deltas = seq(-0.02, 0.044, by = 0.002),
  infinity_estimators = seq(0.1, 0.5, by = 0.1),
  multipliers = seq(0.1, 1.1, by = 0.1))

# If you are using doSNOW
stopCluster(cl)
}