meteoland (version 2.2.2)

interpolation_cross_validation: Calibration and validation of interpolation procedures

Description

Calibration and validation of interpolation procedures

Usage

interpolation_cross_validation(
  interpolator,
  stations = NULL,
  verbose = getOption("meteoland_verbosity", TRUE)
)
interpolator_calibration(
  interpolator,
  stations = NULL,
  update_interpolation_params = FALSE,
  variable = "MinTemperature",
  N_seq = seq(5, 30, by = 5),
  alpha_seq = seq(0.25, 10, by = 0.25),
  verbose = getOption("meteoland_verbosity", TRUE)
)

Value

interpolation_cross_validation returns a list with the following items

• errors: Data frame with each combination of station and date with observed variables, predicated variables and the total error (predicted - observed) calculated for each variable

• station_stats: Data frame with error and bias statistics aggregated by station

• dates_stats: Data frame with error and bias statistics aggregated by date

• r2: correlation indexes between observed and predicted values for each meteorological variable

If update_interpolation_params is FALSE (default), interpolator_calibration returns a list with the following items

• MAE: A numeric matrix with the mean absolute error values, averaged across stations, for each combination of parameters "N" and "alpha"

• minMAE: Minimum MAE value

• N: Value of parameter "N" corresponding to the minimum MAE

• alpha: Value of parameter "alpha" corresponding the the minimum MAE

• observed: matrix with observed values (meteorological measured values)

• predicted: matrix with interpolated values for the optimum parameter combination

If update_interpolation_params is FALSE, interpolator_calibration

returns the interpolator provided with the parameters updated

Arguments

interpolator

A meteoland interpolator object, as created by create_meteo_interpolator

stations

A vector with the stations (numeric for station indexes or character for stations id) to be used to calculate "MAE". All stations with data are included in the training set but predictive "MAE" are calculated for the stations subset indicated in stations param only. If NULL all stations are used in the predictive "MAE" calculation.

verbose

Logical indicating if the function must show messages and info. Default value checks "meteoland_verbosity" option and if not set, defaults to TRUE. It can be turned off for the function with FALSE, or session wide with options(meteoland_verbosity = FALSE)

update_interpolation_params

Logical indicating if the interpolator object must be updated with the calculated parameters. Default to FALSE

variable

A string indicating the meteorological variable for which interpolation parameters "N" and "alpha" will be calibrated. Accepted values are:

• MinTemperature (kernel for minimum temperature)

• MaxTemperature (kernel for maximum temperature)

• DewTemperature (kernel for dew-temperature (i.e. relative humidity))

• Precipitation (to calibrate the same kernel for both precipitation events and regression of precipitation amounts; not recommended)

• PrecipitationAmount (kernel for regression of precipitation amounts)

• PrecipitationEvent (kernel for precipitation events)

N_seq

Numeric vector with "N" values to be tested

alpha_seq

Numeric vector with "alpha"

Functions

• interpolation_cross_validation():

Author

Miquel De Cáceres Ainsa, EMF-CREAF

Victor Granda García, EMF-CREAF

Details

Function interpolator_calibration determines optimal interpolation parameters "N" and "alpha" for a given meteorological variable. Optimization is done by minimizing mean absolute error ("MAE") (Thornton et al. 1997). Function interpolation_cross_validation calculates average mean absolute errors ("MAE") for the prediction period of the interpolator object. In both calibration and cross validation procedures, predictions for each meteorological station are made using a leave-one-out procedure (i.e. after excluding the station from the predictive set).

References

Thornton, P.E., Running, S.W., 1999. An improved algorithm for estimating incident daily solar radiation from measurements of temperature, humidity, and precipitation. Agric. For. Meteorol. 93, 211–228. doi:10.1016/S0168-1923(98)00126-9.

De Caceres M, Martin-StPaul N, Turco M, Cabon A, Granda V (2018) Estimating daily meteorological data and downscaling climate models over landscapes. Environmental Modelling and Software 108: 186-196.

Examples

# \donttest{
# example interpolator
data("meteoland_interpolator_example")

# As the cross validation for all stations can be time consuming, we are
# gonna use only for the first 5 stations of the 198
cv <- interpolation_cross_validation(meteoland_interpolator_example, stations = 1:5)

# Inspect the results
cv$errors
cv$station_stats
cv$dates_stats
cv$r2
# }


# \donttest{
# example interpolator
data("meteoland_interpolator_example")

# As the calibration for all stations can be time consuming, we are gonna
# interpolate only for the first 5 stations of the 198 and only a handful
# of parameter combinations
calibration <- interpolator_calibration(
  meteoland_interpolator_example,
  stations = 1:5,
  variable = "MaxTemperature",
  N_seq = seq(10, 20, by = 5),
  alpha_seq = seq(8, 9, by = 0.25)
)

# we can update the interpolator params directly:
updated_interpolator <- interpolator_calibration(
  meteoland_interpolator_example,
  stations = 1:5,
  update_interpolation_params = TRUE,
  variable = "MaxTemperature",
  N_seq = seq(10, 20, by = 5),
  alpha_seq = seq(8, 9, by = 0.25)
)


# check the new interpolator have the parameters updated
get_interpolation_params(updated_interpolator)$N_MaxTemperature
get_interpolation_params(updated_interpolator)$alpha_MaxTemperature
# }