/EXstat

🌾 R package to provide an efficient and simple solution to aggregate and analyze the stationarity of time series

Primary LanguageRGNU General Public License v3.0GPL-3.0

EXstat

R-CMD-check Lifecycle: stable Contributor Covenant

EXstat is a R package which provide an efficient and simple solution to aggregate and analyze the stationarity of time series.

This project was carried out for National Research Institute for Agriculture, Food and the Environment (Institut National de Recherche pour l’Agriculture, l’Alimentation et l’Environnement, INRAE in french).

Installation

For latest development version

remotes::install_github("super-lou/EXstat")

Documentation

Extraction process

Based on dplyr, input data format is a tibble of at least a column of date, some columns of numeric value and one or more character columns for names of time series in order to identify them uniquely. Thus it is possible to have a tibble with multiple time series which can be grouped by their names.

For example, we can use the following tibble :

library(dplyr)

# Date
Start = as.Date("1972-01-01")
End = as.Date("2020-12-31")
Date = seq.Date(Start, End, by="day")

# Value to analyse
set.seed(100)
X = seq(1, length(Date))/1e4 + runif(length(Date), -100, 100)
X[as.Date("2000-03-01") <= Date & Date <= as.Date("2000-09-30")] = NA

# Creation of tibble
data = tibble(Date=Date, ID="serie A", X=X)

Which looks like that :

> data
# A tibble: 17,898 × 3
   Date       ID           X
   <date>     <chr>    <dbl>
 1 1972-01-01 serie A -38.4 
 2 1972-01-02 serie A -48.5 
 3 1972-01-03 serie A  10.5 
 4 1972-01-04 serie A -88.7 
 5 1972-01-05 serie A  -6.29
 6 1972-01-06 serie A  -3.25
 7 1972-01-07 serie A  62.5 
 8 1972-01-08 serie A -25.9 
 9 1972-01-09 serie A   9.31
10 1972-01-10 serie A -65.9 
# ℹ 17,888 more rows
# ℹ Use `print(n = ...)` to see more rows

The process of variable extraction (for example the yearly mean of time series) is realised with the process_extraction() function. Minimum arguments are :

  • Input data described above
  • The function funct (for example mean) you want to use. Arguments of the chosen function can be passed to this extraction process and the function can be previously defined.

Some of the optional arguments are :

  • period A vector of two dates (or two unambiguous character strings that can be coerced to dates) to restrict the period of analysis. As an example, it can be c("1950-01-01", "2020-12-31") to select data from the 1st January of 1950 to the end of December of 2020. The default option is period=NULL, which considers all available data for each time serie.
  • time_step A character string specifying the time step of the variable extraction process. Possible values are :
    • "year" for a value per year
    • "month" for a value for each month of the year (so 12 values if at least a full year is given)
    • "year-month" for a value for each month of each year (so 12 times the number of given year values at the end)
    • "season" for a value for each season of th year (so by default 4 values)
    • "year-season" for a value for each season of each year (so by default 4 times the number of given year values at the end)
    • "yearday" for one value per day of the year (so 365 values at the end if at least a full year is given... but more than one year seems obviously more interesting)
    • "none" if you want to extract a unique value for the whole time serie
  • sampling_period A character string or a vector of two character strings that will indicate how to sample the data for each time step defined by time_step. Hence, the choice of this argument needs to be link with the choice of the time step. For example, for a yearly extraction so if time_step is set to "year", sampling_period needs to be formated as %m-%d (a month - a day of the year) in order to indicate the start of the sampling of data for the current year. More precisly, if time_step="year" and sampling_period="03-19", funct will be apply on every data from the 3rd march of each year to the 2nd march of the following one. In this way, it is possible to create a sub-year sampling with a vector of two character strings as sampling_period=c("02-01", "07-31") in order to process data only if the date is between the 1st february and the 31th jully of each year.

More parameters are available, for example, to :

  • handle missing values,
  • use suffixes to simplify expressions, and
  • manage variables related to seasonality.

In this way

dataEX = process_extraction(data=data,
                            funct=max,
                            funct_args=list("X", na.rm=TRUE),
                            time_step="year",
                            sampling_period=c("05-01",
                                              "11-30"),
                            period=c(as.Date("1990-01-01"),
                                     as.Date("2020-12-31")))

will perform a yearly extraction of the maximum value between may and november, from the 1th march of 1990 to the 31th october of 2020, ignoring NA values.

The output is also a tibble with a column of date, of character for the name of time series and a numerical column with the extracted variable from the time series.

> dataEX
# A tibble: 31 × 3
   ID      Date           X
   <chr>   <date>     <dbl>
 1 serie A 1990-05-01 100. 
 2 serie A 1991-05-01 101. 
 3 serie A 1992-05-01 100. 
 4 serie A 1993-05-01  99.9
 5 serie A 1994-05-01  99.0
 6 serie A 1995-05-01 100. 
 7 serie A 1996-05-01 100. 
 8 serie A 1997-05-01 101. 
 9 serie A 1998-05-01  99.6
10 serie A 1999-05-01 101. 
# ℹ 21 more rows
# ℹ Use `print(n = ...)` to see more rows

Other examples of more complex cases are available in the package documentation. Try starting with

library(EXstat)
?EXstat

Extraction process with CARD

For a more user-friendly interaction, this package has been developed in symbiosis with predefined parameterisation files called CARD.

So you don't have to define complex parameters yourself to extract hydroclimatological variables. What's more, if the CARD you want doesn't exist, it's easy to create one based on the others.

To do this, you need to download the CARD archive, extract it and place it wherever you like (as if it were data). Then you can create a new subdirectory within this main CARD directory, which you can call for example "analyse_1", and copy and paste the CARD "__all__/Hautes_Eaux/QJXA.R" into it.

In this way, you can carry out "analyse_1" by running:

CARD_extraction(data %>% rename(Q=X),
                CARD_path="path/to/CARD",
                CARD_dir="analyse_1")

Thus, you can place several CARDs in your "analyse_1" sub-directory for multiple analyses.

However, this copy/pasting action can be quite cumbersome and repetitive for large analyses. Therefore, with CARD_management(), it is possible to automate the CARD copy/pasting from the CARD directory to an external temporary directory, like this:

CARD_management(CARD_path="path/to/CARD",
                CARD_tmp="path/to/copy/CARD",
                CARD_dir="analyse_1",
                CARD_name=c("QA", "QJXA"),
                overwrite=TRUE,
                verbose=TRUE)

As a result, to run the analysis, use:

CARD_extraction(data %>% rename(Q=X),
                CARD_path="path/to/CARD",
                CARD_tmp="path/to/copy/CARD",
                CARD_dir="analyse_1")

Take a look at the CARD repository to better understand the CARD formatting.

Trend analyse

The stationarity analyse is computed with the process_trend() function on the extracted data dataEX. The statistical test used here is the Mann-Kendall test12.

Hence, the following expression

trendEX = process_trend(data=dataEX)

produces the result below

# A tibble: 1 × 12
  ID      variable_en level H          p      a     b period_trend
  <chr>   <chr>       <dbl> <lgl>  <dbl>  <dbl> <dbl> <list>      
1 serie A X             0.1 TRUE  0.0958 0.0260  99.3 <date [2]>

  mean_period_trend a_normalise a_normalise_min a_normalise_max
  <lgl>                   <dbl>           <dbl>           <dbl>
1 NA                     0.0260          0.0260          0.0260

It is a tibble which precises, among other information, the name of the time serie the p value and a the Theil-Sen's slope34, for each row.

Finaly, as the p value is below 0.1, the previous time serie shows an increasing linear trend which can be represented by the equation Y = 0.0260*X + b with a type I error of 10 % or a trust of 90 %.

FAQ

I have a question.

  • Solution: Search existing issue list and if no one has a similar question create a new issue.

I found a bug.

  • Good Solution: Search existing issue list and if no one has reported it create a new issue.
  • Better Solution: Along with issue submission provide a minimal reproducible example of the bug.
  • Best Solution: Fix the issue and submit a pull request. This is the fastest way to get a bug fixed.

Code of Conduct

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

Footnotes

  1. Mann, H. B. (1945). Nonparametric tests against trend. Econometrica: Journal of the econometric society, 245-259. ↩

  2. Kendall, M.G. (1975) Rank Correlation Methods. 4th Edition, Charles Grifin, London. ↩

  3. Theil, H. (1950). A rank-invariant method of linear and polynomial regression analysis. Indagationes mathematicae, 12(85), 173. ↩

  4. Sen, P. K. (1968). Estimates of the regression coefficient based on Kendall's tau. Journal of the American statistical association, 63(324), 1379-1389. ↩