Ensemble Calibration from high Temporal resolution data

[pe2ju]

Consider an example below which uses sample data from the ensembleBMA package. This works fine. How can I achieve the same using high temporal resolution data say at 3 hourly? Or is this works only for daily data and not intermediate hours of the day? Please see my codes below;

ensembleMOS: Ensemble Model Output Statistics

rm( list = ls(all.names = TRUE) )  # Clear all 
# Load libraries to use
library("ensembleBMA")
library("ensembleMOS")

# Get sample data 
data("ensBMAtest", package = "ensembleBMA")

# Ensemble names
ensMemNames <- c("gfs","cmcg","eta","gasp","jma","ngps","tcwb","ukmo")

# Prepare input variable & target 
obs <- paste("MAXWSP10","obs", sep = ".")
ens <- paste("MAXWSP10", ensMemNames, sep = ".")

# Prepare an ensemble data object. This uses "ensembleBMA" package
windTestData <- ensembleData( forecasts = ensBMAtest[, ens],
                              dates = ensBMAtest[, "vdate"],
                              observations = ensBMAtest[, obs],
                              station = ensBMAtest[, "station"],
                              forecastHour = 48,
                              initializationTime = "00")

# Train EMOS model
date_target <-  format( seq( ISOdate(2007, 12, 27, 0), by = "1 day", 
                             length.out = 7 ), "%Y%m%d%H" )

windFitLN  <- ensembleMOS( ensembleData = windTestData, 
                           trainingDays = 25,
                           dates = date_target,
                           model = "lognormal")

# Get the  CRPS
windCRPS <- crps( windFitLN, windTestData )
# Get the quantile forecast
windFcst <- quantileForecast( fit = windFitLN, 
                               quantiles = seq(0.1, 0.9, 0.1),
                               ensembleData = windTestData,
                               dates = date_target )

#===============================================================================
#========== Now I modify the dates to 3-hourly high temporal resolution ======== 
# Load libraries to use
library("ensembleBMA")
library("ensembleMOS")

# Get sample data 
data("ensBMAtest", package = "ensembleBMA")

# Ensemble names
ensMemNames <- c("gfs","cmcg","eta","gasp","jma","ngps","tcwb","ukmo")

# Prepare input variable & target 
obs <- paste("MAXWSP10","obs", sep = ".")
ens <- paste("MAXWSP10", ensMemNames, sep = ".")

# Generate new date with high time resolution
new_vdate <-  rep( format( seq( ISOdate(2007, 12, 27, 0), 
                           by = "3 hour", 
                           length.out = nrow(ensBMAtest)*0.5 ), 
                      "%Y%m%d%H" ), each = 2 )
new_ensBMAtest <- dplyr::mutate( ensBMAtest[, -1], vdate = new_vdate )

# new ensemble data object using "ensembleBMA" package
new_windTestData <- ensembleData( forecasts = new_ensBMAtest[, ens],
                                  dates = new_ensBMAtest[, "vdate"],
                                  observations = ensBMAtest[, obs],
                                  station = ensBMAtest[, "station"],
                                  forecastHour = 3, # We change to 3 hour from 48 hour
                                  initializationTime = "00")

# Build the lognormal EMOS model
new_date_target <- new_vdate[49:66]
new_windFitLN  <- ensembleMOS( ensembleData = new_windTestData, 
                               trainingDays = 25,
                               dates = new_date_target,
                               model = "lognormal")
#> Warning in ensembleMOSlognormal(ensembleData = new_windTestData, trainingDays =
#> 25, : valid dates do not have a unique forecast hour
#> Warning in ensembleMOSlognormal(ensembleData = new_windTestData, trainingDays =
#> 25, : dates do not have a unique forecast hour
#> Error in ensembleMOSlognormal(ensembleData = new_windTestData, trainingDays = 25, : some dates precede the first training period