gfJulendat: Imputation of medium-sized gaps by linear modeling using...

In environmentalinformatics-marburg/GSODTools: R Functions to Select, Download, and Process GSOD Data

Imputation of medium-sized gaps by linear modeling using adjacent GSOD stations

Description

This is a wrapper function encompassing several sub-functions from the Julendat gap-filling routine (processing level "0310", see Julendat). Briefly, a discontinuous measurement series from a distinct GSOD station is being filled by the use of simultaneous measurements from adjacent GSOD stations. Taking the gappy data as response variable, a linear model is fitted to explain the measurements taken by the station under investigation based on surrounding stations.

Usage

gfJulendat(
  files.dep,
  files.indep,
  filepath.coords = NULL,
  quality.levels = NULL,
  gap.limit,
  end.datetime = Sys.Date(),
  units = "days",
  na.limit = 0.5,
  time.window = 365,
  n.plot,
  prm.dep = "TEMP",
  prm.indep = "NA",
  family = stats::gaussian,
  plevel,
  ...
)

Arguments

files.dep	Character. Path leading to the response GSOD station. Note: the file needs to be formatted according to standard KiLi SP1 format, see gsod2ki.
files.indep	Character. Path(s) leading to the predictor GSOD station(s).
filepath.coords	Character, default is NULL. Path to file holding coordinate information. Not required for GSOD data processing.
quality.levels	Character, default is NULL. Not required for GSOD data processing.
gap.limit	Numeric. Maximum length of a measurement gap to be imputed. All gaps exceeding this threshold will not be considered.
end.datetime	Object of class Date, default is Sys.Date(). Not required for GSOD data processing.
units	Character. Measurement interval, typically "days" for GSOD data.
na.limit	Numeric, default is 0.5. Maximum amount of missing data within time.window. If a certain explanatory station exceeds this threshold, it will not be considered for linear modeling.
time.window	Numeric, default is 365. Window width before and after a data gap to extract data for linear modeling.
n.plot	Numeric. Number of explanatory GSOD stations to consider for linear modeling. If not supplied, all stations specified in files.indep will be considered.
prm.dep	Character, default is "TEMP". Determines which parameters to fill.
prm.indep	Character, default is "NA". Not required for GSOD data processing.
family	Object of class family, default is gaussian.
plevel	Character. Determines current processing level. Not required for GSOD data processing.
...	Additional arguments. Currently not in use.

Value

An object of class ki.data.

Author(s)

Florian Detsch

Examples

## Not run: 
library(GSODTools)

# Download data sets for selected GSOD stations
usafs = c(
  "gar" = "637230"   # GARISSA
  , "jom" = "637400" # NAIROBI JKIA
  , "kia" = "637910" # KILIMANJARO INTL
  , "voi" = "637930" # VOI
  , "mom" = "638200" # MOMBASA MOI INTL
  , "mor" = "638660" # MOROGORO (MET)
)

shp_gsod <- 
  gsodstations |> 
  gsodDf2Sp() |> 
  subset(`USAF` %in% usafs) |> 
  gsodDf2Sp()

df_gsod = Map(
  \(usaf, plot_id) {
    
    # Download and extraction
    tmp_df_gsod <- dlGsodStations(usaf = usaf, 
                                  start_year = 1990, end_year = 1995, 
                                  dsn = tempdir(), 
                                  unzip = TRUE,
                                  save_output = FALSE)
    
    # Fahrenheit -> Celsius
    tmp_df_gsod = transform(
      tmp_df_gsod
      , TEMP = toCelsius(TEMP, digits = 1L)
      , MIN = toCelsius(MIN, digits = 1L)
      , MAX = toCelsius(MAX, digits = 1L)
    )
    
    # GSOD -> `ki.data`
    tmp_ki_gsod <- gsod2ki(tmp_df_gsod, 
                           prm_col = c("TEMP", "MIN", "MAX"), 
                           timezone = "eat",
                           aggtime = "diurnal",
                           plot_id = plot_id, 
                           df2ki = TRUE)
    
    # Remove outliers
    for (j in c("TEMP", "MIN", "MAX")) {
      methods::slot(tmp_ki_gsod, "Parameter")[[j]] = outlier2na(
        methods::slot(tmp_ki_gsod, "Parameter")[[j]]
        , lower_quantile = 0.2
        , upper_quantile = 0.8
      )
    }
    
    # Fill small gaps by linear interpolation
    tmp_ki_gsod <- gfLinInt(tmp_ki_gsod, 
                            prm = c("TEMP", "MIN", "MAX"))
    
    # Save data created so far
    tmp_df_gsod <- gfOutputData(tmp_ki_gsod, plevel = "NA")
    write.csv(tmp_df_gsod, sprintf("%s/%s.csv", tempdir(), usaf), 
              row.names = FALSE)
    
    return(tmp_ki_gsod)    
  }
  , methods::slot(shp_gsod, "data")$USAF
  , names(usafs)
)

# Fill medium-sized gaps at Kilimanjaro Intl. Airport by linear modeling
fls_gsod <- list.files(tempdir(), pattern = "^\\d{6}.csv$", 
                       recursive = TRUE, full.names = TRUE)

jul_gsod <- gfJulendat(files.dep = fls_gsod[3],
                       files.indep = fls_gsod[-3],
                       filepath.coords = NULL,
                       quality.levels = NULL,
                       gap.limit = 1825, 
                       na.limit = .9,
                       time.window = 913,
                       n.plot = 10,
                       prm.dep = c("TEMP", "MAX", "MIN"), 
                       prm.indep = c(NA, NA, NA), 
                       plevel = "NA", 
                       end.datetime = Sys.Date(), 
                       units = "days")

plot(jul_gsod$TEMP, col = "red", type = "l")
lines(methods::slot(df_gsod[[3]], "Parameter")$TEMP, col = "grey75")

## End(Not run)

environmentalinformatics-marburg/GSODTools documentation built on Jan. 5, 2024, 12:19 a.m.