aemet_temp: AEMET daily temperatures during 1974-2013

In goffda: Goodness-of-Fit Tests for Functional Data

AEMET daily temperatures during 1974–2013

Description

Series of daily temperatures of 73 Spanish weather stations during the 40-year period 1974–2013.

Usage

aemet_temp

Format

A list with the following entries:

temp

an fdata with 2892 temperature (in Celsius degrees) curves, discretized on 365 equispaced grid points (days) on [0.5, 364.5]. Each curve corresponds to the yearly records of a weather station.

df

a dataframe with metadata for each curve:

• ind: identifier of the weather station.

• name: name of the weather station.

• year: year of the observation.

Details

For consistency with the fda.usc-package's aemet dataset, the names and identifiers of the 73 weather stations are the same as in that dataset. Only a minor fix has been applied to the "A CORUÑA/ALVEDRO" station, whose identifier was the same as the "A CORUÑA" station, "1387". The former was set to "1387A".

Information about the province, altitude, longitude, and latitude of each weather station can be retrieved in df from the fda.usc-package's aemet dataset.

The dataset is a curated version of a larger database of 115 stations. It excludes stations with inconsistent records or that were relocated, closed, or opened during the 40-year period. There are 9 stations with missing years. The total of missing years is 28.

In leap years, the daily-average temperature is computed as the average of February 28th and 29th.

Author(s)

Original data processing scripts by Manuel Febrero-Bande and Manuel Oviedo de la Fuente. Adaptations by Eduardo García-Portugués.

Source

The data was retrieved from the FTP of the Meteorological State Agency of Spain (AEMET) in 2014 using a processing script by the authors of the fda.usc-package.

References

Febrero-Bande, M. and Oviedo de la Fuente, M. (2012). Statistical Computing in Functional Data Analysis: The R Package fda.usc. Journal of Statistical Software, 51(4):1–28. \Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v051.i04")}

Examples

## Data splitting

# Load raw data
data("aemet_temp")

# Partition the dataset in the first and last 20 years
with(aemet_temp, {
  ind_pred <- which((1974 <= df$year) & (df$year <= 1993))
  ind_resp <- which((1994 <= df$year) & (df$year <= 2013))
  aemet_temp_pred <<- list("df" = df[ind_pred, ], "temp" = temp[ind_pred])
  aemet_temp_resp <<- list("df" = df[ind_resp, ], "temp" = temp[ind_resp])
})

# Average the temperature on each period
mean_aemet <- function(x) {
  m <- tapply(X = 1:nrow(x$temp$data), INDEX = x$df$ind,
              FUN = function(i) colMeans(x$temp$data[i, , drop = FALSE],
                                         na.rm = TRUE))
 x$temp$data <- do.call(rbind, m)
 return(x$temp)
}

# Build predictor and response fdatas
aemet_temp_pred <- mean_aemet(aemet_temp_pred)
aemet_temp_resp <- mean_aemet(aemet_temp_resp)

# Plot
old_par <- par(mfrow = c(1, 2))
plot(aemet_temp_pred)
plot(aemet_temp_resp)
par(old_par)

# Average daily temperatures
day_avg_pred <- func_mean(aemet_temp_pred)
day_avg_resp <- func_mean(aemet_temp_resp)

# Average yearly temperatures
avg_year_pred <- rowMeans(aemet_temp_pred$data)
avg_year_resp <- rowMeans(aemet_temp_resp$data)

## Test the linear model with functional response and predictor

(comp_flmfr <- flm_test(X = aemet_temp_pred, Y = aemet_temp_resp,
                        est_method = "fpcr_l1s"))
beta0 <- diag(rep(1, length(aemet_temp_pred$argvals)))
(simp_flmfr <- flm_test(X = aemet_temp_pred, Y = aemet_temp_resp,
                        beta0 = beta0, est_method = "fpcr_l1s"))

# Visualize estimation
filled.contour(x = aemet_temp_pred$argvals, y = aemet_temp_resp$argvals,
               z = comp_flmfr$fit_flm$Beta_hat,
               color.palette = viridisLite::viridis, nlevels = 20)

## Test the linear model with scalar response and functional predictor

(comp_flmsr <- flm_test(X = aemet_temp_pred, Y = avg_year_resp,
                        est_method = "fpcr_l1s"))
(simp_flmsr <- flm_test(X = aemet_temp_pred, Y = avg_year_resp,
                        beta0 = 1 / 365, est_method = "fpcr_l1s"))

# Visualize estimation
plot(aemet_temp_pred$argvals, comp_flmsr$fit_flm$Beta_hat, type = "l",
     ylim = c(0, 30 / 365))
abline(h = 1 / 365, col = 2)

## Test the linear model with functional response and scalar predictor

(comp_frsp <- flm_test(X = avg_year_pred, Y = aemet_temp_resp))
(simp_frsp <- flm_test(X = avg_year_pred, Y = aemet_temp_resp, beta0 = 1))

## Test the linear model with scalar response and predictor

(comp_srsp <- flm_test(X = avg_year_pred, Y = avg_year_resp))
(simp_srsp <- flm_test(X = avg_year_pred, Y = avg_year_resp, beta0 = 1))

goffda documentation built on Oct. 14, 2023, 5:08 p.m.