In airGRteaching: Teaching Hydrological Modelling with the GR Rainfall-Runoff Models ('Shiny' Interface Included)

formatGR      <- '<strong><font color="#0BA6AA">%s</font></strong>'
GR            <- sprintf(formatGR, "GR")
airGR         <- sprintf(formatGR, "airGR")
airGRteaching <- sprintf(formatGR, "airGRteaching")

knitr::opts_chunk$set(echo = TRUE)
library(airGRteaching)

Hydrological modelling in three steps

This part explains how to run the r airGR hydrological models in only three simple steps with r airGRteaching.

Preparation of input data

A data.frame of daily hydrometeorological observations time series at the catchment scale is needed. The required fields are:

DatesR: dates in the POSIXt format
P: average precipitation [mm/time step]
T: catchment average air temperature [℃] [OPTIONAL]
E: catchment average potential evapotranspiration [mm/time step]
Qmm: outlet discharge [mm/time step]

data(L0123001, package = "airGR")
BasinObs <- BasinObs[, c("DatesR", "P", "E", "Qmm", "T")]
head(BasinObs)

Before running a model, r airGRteaching functions require data and options with specific formats.

For this step, you just have to use the PrepGR() function. You have to define:

ObsDF: data.frame of hydrometeorological observations time series
HydroModel: the name of the hydrological model you want to run (GR1A, GR2M, GR4J, GR5J, GR6J, GR4H or GR5H)
CemaNeige: if you want or not to use the snowmelt and accumulation model

If you want to use CemaNeige, you also have to define:

catchment average air temperature in ObsDF or in TempMean
HypsoData: a vector of 101 reals: min, quantiles (1 % to 99 %) and max of catchment elevation distribution [m]; if not defined a single elevation layer is used for CemaNeige
NLayers: the number of elevation layers requested [-]

PREP <- PrepGR(ObsDF = BasinObs, HydroModel = "GR5J", CemaNeige = FALSE)

Calibration step

To calibrate a model, you just have to use the CalGR() function. By default, the objective function used is the Nash–Sutcliffe criterion ("NSE"), and the warm-up period is automatically set (depends on model). You just have to define:

PrepGR: the object returned by the PrepGR() function
CalPer: a vector of 2 dates to define the calibration period

You can obviously define another objective function or warm-up period:

CalCrit: name of the objective function ("NSE", "KGE", "KGE2", "RMSE")
WupPer: a vector of 2 dates to define the warm-up period

The calibration algorithm has been developed by Claude Michel (Calibration_Michel() function in the r airGR package) .

CAL <- CalGR(PrepGR = PREP, CalCrit = "KGE2",
             WupPer = NULL, CalPer = c("1990-01-01", "1993-12-31"))

Simulation step

To run a model, please use the SimGR() function. The PrepGR and WupPer arguments of SimGR() are similar to the ones of the CalGR() function. Here, EffCrit is used to calculate the performance of the model over the simulation period SimPer and Param is the object returned by the CalGR() function.

SIM <- SimGR(PrepGR = PREP, Param = CAL, EffCrit = "KGE2",
             WupPer = NULL, SimPer = c("1994-01-01", "1998-12-31"))

Formating outputs

The call of the as.data.frame() function with PrepGR, CalGR or SimGR objects allows to coerce the outputs to a data frame.

head(as.data.frame(PREP))
head(as.data.frame(CAL))
head(as.data.frame(SIM))

Pre-defined graphical plots

Static plots

The call of the plot() function with a PrepGR object draws the observed precipitation and discharge time series.

par(cex.lab = 0.6, cex.axis = 0.6)
plot(PREP, main = "Observation")

By default (with the argument which = "synth"), the call of the plot() function with a CalGR object draws the classical r airGR plot diagnostics (observed and simulated time series together with diagnostic plot)

plot(CAL, which = "synth")

plot(CAL, which = "synth", cex.lab = 0.7, cex.axis = 0.7)

With the CalGR object, if the argument which is set to "iter", the plot() function draws the evolution of the parameters and the values of the objective function during the second step of the calibration (steepest descent local search algorithm):

plot(CAL, which = "iter")

With the CalGR object, if the argument which is set to "ts", the plot() function simply draws the time series of the observed precipitation, and the observed and simulated flows:

par(cex.lab = 0.7, cex.axis = 0.7)
plot(CAL, which = "ts", main = "Calibration")

The call of the plot() function with a SimGR object displays the classical r airGR plot diagnostics.

plot(SIM)

Dynamic plots

Dynamic plots, using the dygraphs JavaScript charting library, can be displayed by the package.

The dyplot() function can be applied on PrepGR, CalGR and SimGR objects and draws the time series of the observed precipitation, and the observed and simulated (except with PrepGR objects) flows.

The user can zoom on the plot device and can read the exact values.

With this function, users can easily explore the data time series and also explore and interpret the possible problems of the calibration or simulation steps.

dyplot(SIM, main = "Simulation")

Graphical user interface

The r airGRteaching package also provides the ShinyGR() function, which allows to launch a graphical user interface using the shiny package.

The ShinyGR() function needs at least:

ObsDF: a (list of) data.frame (or independant vector instead, see ?ShinyGR)
SimPer: a (list of) vector(s) of 2 dates to define the simulation period(s)

ShinyGR(ObsDF = BasinObs, SimPer = c("1994-01-01", "1998-12-31"))

Only the monthly model (GR2M) and the daily models (GR4J, GR5J, GR6J + CemaNeige) are currently available.
If you want to use CemaNeige, you also have to define the same arguments desribed above for the PrepGR() function.