The run period is reduced in the example of the Imax() function in order to run faster.
The ‘hydroPSO’ package is no longer suggested (but the code linked to its use and is always present in the ‘param_optim’ vignette).
citation(“airGR”) command.Imax(). The default value of the TestedValues argument was wrong due to a mistyped argument name in the seq() function.plot.Outputsmodel() now allows to draw actual evapotranspiration when which = “ActuEvap” or which = “All” (overlaid to potential evapotranspiration if already drawn).RunModel_GR5H() and RunModel_CemaNeigeGR5H() functions to run the hourly model GR5H (with or without the CemaNeige module). These models present an optional additionnal interception store.Imax() which allows to estimate the maximum capacity of the GR5H interception store.TransfoParam_GR1A(). The number of model parameters was wrong (2 instead of 1) which caused an error during the GR1A model calibration.plot.OutputsModel(). The function does not return any error message when log_scale = TRUE, Qobs = NULL and user want to draw flows time series.RunModel_*GR*(). The functions do not return any error message anymore due to slightly negative values returned by GR4H, GR4J, GR5J or GR6J Fortran codes (the message was returned by CreateIniStates() when the final states were created). The RunModel_*GR*() functions now return zero instead of these slightly negative values, except for the ExpStore where negatives values are allowed..ErrorCrit() function. The Box-Cox transformation formula is now corrected when the ErrorCrit*() functions are used.RunModel_GR4H() function (Pn, Ps, AExch1, AExch2).RunModel_GR2M() documentation.https://gitlab.irstea.fr/HYCAR-Hydro/airgr) and issues (https://gitlab.irstea.fr/HYCAR-Hydro/airgr/issues).RunModel_CemaNeige(). The function now runs correctly when IndPeriod_WarmUp = 0L in CreateRunOptions() in order to completely disable the warm-up period (e.g. to perform a forecast form a given initial state).CreateIniStates(). The function now returns the right number of end states when CemaNeige is used whithout hysteresis.RunModel_CemaNeige*() functions. G and Gthr end states are no more inverted in the output values.NA into the BasinObs time series of the L0123001 dataset.CreateInputsCrit() now allows power (as a numeric or as a character) and the Box-Cox transformations in the transfo argument.
Added RunModel_CemaNeigeGR4H() function to run the hourly model GR4H with the CemaNeige module.
Added PE_Oudin() function to compute Oudin’s potential evapotranspiration for hourly or daily time steps.
The PEdaily_Oudin() function is deprecated and his use has been replaced by the use of PE_Oudin().
plot.OutputsModel() now presents a LayoutMat argument (and additionnal related argument: LayoutWidths, LayoutHeights) to specify complex plot arrangements.
plot.OutputsModel(). The function now runs correctly when the which argument contains the “CorQQ” value without “CumFreq”.plot.OutputsModel() can now draw PE or error time series if the which argument is set to “all” or “PotEvap” or “Error”.
plot.OutputsModel() now allows new values for the which argument: “all” corresponds to all graphs, “synth” corresponds to the main graphs (default value; corresponding to “all” in the previous versions of the package) (i.e. c(“Precip”, “Temp”, “SnowPack”, “Flows”, “Regime”, “CumFreq”, “CorQQ”)), “ts” corresponds to the time series graphs (i.e. c(“Precip”, “PotEvap”, “Temp”, “SnowPack”, “Flows”)) and “perf” corresponds to the performance graphs (i.e. c(“Error”, “Regime”, “CumFreq”, “CorQQ”)).
.ErrorCrit() private function added to check inputs into ErrorCrit_*() functions. The ErrorCrit_*() functions were simplified accordingly.
CreateInputsCrit() now returns FUN_CRIT as a character string.
An example is addeed to illustred the use of the plot.OutputsModel() function.
CreateInputsCrit() now presents a VarObs argument in order to allow to prepare an InputsCrit object in order to run a criterion on other variables than observed discharges with the ErrorCrit() function (at the moment SCA and SWE).
CreateInputsCrit() can now prepare an InputsCrit object in order to compute a single criterion (Single class), multiple criteria (Multi class) or a composite criterion (Compo class) with the ErrorCrit() function.
CreateInputsCrit() now presents a Weights argument in order to allow to prepare an InputsCrit object in order to compute a composite criterion (Compo class) with ErrorCrit() or Calibration_Michel().
CreateInputsCrit() now returns a idLayer element to indicate which layer(s) to use for SCA or SWE aggregation.
CreateInputsCrit() now presents a warnings argument to replace the verbose action (the verbose argument is kept to print messages).
In CreateInputsCrit(), it is now possible to set the following arguments as atomic (as before) or as list: FUN_CRIT, VarObs, Obs, BoolCrit, transfo, Weights. If the list format is chosen, all the lists must have the same length.
CreateRunOptions(), CreateIniStates() and CreateCalibOptions() now present a IsHyst argument to give the possibility to use the Linear Hysteresis with CemaNeige. The objects returned present an hysteresis class.
CreateRunOptions() now presents a warnings argument to replace the verbose action (the verbose argument is kept to print messages).
Added TransfoParam_CemaNeigeHyst() function in order to take into account transformation of the parameters of the CemaNeige module when the Linear Hysteresis is used.
Added the X0310010 dataset to run the examples using the Linear Hysteresis with CemaNeige (it contains necessary SCA data).
Added the ‘cemaneige_hysteresis’ vignette to explain how to manage the use of the Linear Hysteresis with CemaNeige.
The Qobs argument is now deprecated in CreateInputsCrit() and has been renamed Obs.
The FUN_CRIT argument is now deprecated in ErrorCrit(). This function now gets this information from the InputsCrit argument.
The FUN_CRIT argument is now deprecated in Calibration_Michel(). This function now gets this information from the InputsCrit argument.
The plot_OutputsModel() had been deprecated in airGR 1.0.4 (it had been replaced by the use of plot.OutputsModel() or plot()) and is defunct now.
CreateInputsCrit() now return a list of InputsCrit (each element is of the Single class) in the cases of multiple or a composite criteria.
ErrorCrit_*() functions now return an error message if the InputsCrit object is of class Multi or Compo.
ErrorCrit() function can now run on a multiple or a composite InputsCrit. In these cases, it returns a list of ErrorCrit.
ErrorCrit() and ErrorCrit_*() functions can now assess Q, SCA or SWE simulations.
Calibration_Michel() function can now run on a composite InputsCrit. It returns a composite value of error and the formula used to calculate it.
Model diagrams added in documentations of RunModel_GR4J(), RunModel_GR5J() and RunModel_GR6J() functions.
It is now possible to be redirected to the plot.OutputsModel() documentation with ?plot.
It is now possible to use a character vector for all FUN_* arguments (in addition to function objects) in the following functions: Calibration(), Calibration_Michel(), CreateCalibOptions(), CreateIniStates(), CreateIniStates(), CreateInputsCrit(), CreateInputsModel(), CreateRunOptions(), ErrorCrit(), RunModel() and TransfoParam().
ErrorCrit_*() functions now return objects of class ErrorCrit and NSE, KGE, KGE2 or RMSE.
.FortranOutputs() private function added to manage Fortran outputs.
Outputs of frun_GR2M Fortran subroutine were reordered.
DataAltiExtrapolation_Valery() now returns named elements of lists relative to elevation layer.
Calibration() function now returns an error message if FUN_CALIB is not a function.
Inputs of PEdaily_Oudin() are now checked.
PEdaily_Oudin() example corrected (the Julian day was one day too early).
plot.OutputsModel() does not return a warning message anymore when Qobs = NULL.
Inputs of TransfoParam*() functions are now checked.
The order of authors has been updated in the DESCRIPTION and the CITATION files.
CreateRunOptions(). The function now accounts correctly for leap years when no warm-up period is defined.CreateRunOptions() was cleant, with no effect on its outputs.vignetteParam*.rda datasets moved to the inst directory. It contains different objects needed for ‘param_optim’ and ‘param_mcmc’ vignettes.PEdaily_Oudin() now presents a LatUnit argument which allows to choose the unit of the latitude (radians and degrees).The LatRad argument is now deprecated in PEdaily_Oudin() and replaced by the Lat argument.
The unused Ind_zeroes argument of the CreateInputsCrit() function is now deprecated.
Calibration_Michel() is now faster during the grid-screening step when a parameter is set using FixedParam in CreateCalibOptions().
CreateCalibOptions() now returns an error when all the parameters are set in the FixedParam argument and a warning message when all the parameters are free (NA) in the FixedParam argument.
CreateInputsCrit() now returns an error when epsilon is not positive.
CreateInputsCrit() now returns a warning message in the following case: there are zeroes values in Qobs, epsilon = NULL and transfo = log or inv.
ErrorCrit_*() functions now return a warning message in the following case: there are zeroes values in Qobs or Qsim, epsilon = NULL and transfo = log or inv.
Several functions of the package were cleant or slightly modified, with no effect on their outputs.
Dubious Qls and Qmm values set to NA values between 1997-01-05 and 1997-01-21 in the L0123001 dataset.
ORCID numbers are now joined to the names of the authors of the package.
Function name changed in a vignettes to avoid error during the check on Debian distribution
As recomanded by CRAN managers, the NEWS file is now at the text format and is no more just a link to the airGR Website
Added the Vignette_Param. datasets in order to reduce runtime during the re-building of vignettes. It contains different objects needed for param_optim and param_mcmc vignettes.
RunModel_GR2M(). The function now returns the total precipitation (P) instead of the net rainfall (P1).RunModel_GR2M() now returns more explicit precipitation outputs names.
CreateInputsCrit() now returns a warning message when the KGE (or KGE’) is used with a log transformation on flows.
The article reference is corrected.
An article describing the ‘airGR’ package has been published. Its reference has been added and will be displayed with citation(“airGR”).
Added CreateIniStates() function in order to help user to format the IniStates argument for CreateRunOptions().
Added the Param_Sets_GR4J dataset. It contains generalist parameter sets for the GR4J model.
Three vignettes have been added. They are relative to different calibration methods (including the generalist parameters sets of the GR4J model).
RunSnowModule argument is now deprecated in CreateRunOptions().Fixed bug in RunModel_GR4H(): in frun_GR4H Fortran subroutine, St(2) is now set to 0 (instead of St(1)) when St(2) < 0.
Fixed bug in plot.OutputsModel() for the regime plot when the period is less than 1 year.
Fixed bug in plot.OutputsModel() when there is no common data to plot the cumulative frequency or the correlation QQ.
Fixed bug in plot.OutputsModel() for the y-axis labelling of flows time series when log_scale = TRUE and BasinArea is used.
RunModel_GR4J(), RunModel_GR5J() and RunModel_GR6J() (and CemaNeige_GR*J()) now return Ps, Pn and actual exchanges. See the model Fortran codes for more details about the calculation of these variables.
CreateInputsModel() now returns an error when DatesR contains duplicated values.
RunModel_GR5J now returns StateEnd in the same order as the other models.
plot.OutputsModel() now returns a warning message when the length of Qobs is different from the length of Qsim.
The X1 parameter from GR4H, GR4J, GR2M, GR5J and GR6J, the X3 parameter from GR4H, GR4J, GR5J and GR6J and the X6 parameter from GR6J are now set to 1e-2 when they are fixed to lower values. RunModel_*() functions now return a warning message in this case. RunModel_*() functions now return a warning when X4 < 0.5 and its value is set to 0.5.
The commands ?L0123001, ?L0123002 and ?L0123003 now return the documentation page related to BasinObs.
Many functions of the package were cleant or slightly modified, with no effect on their outputs.
The documentation and help of several functions were improved.
DataAltiExtrapolation_Valery() and CreateInputsModel() now present a PrecipScale argument which allows rescaling precipitation when it is interpolated on the elevation layers when CemaNeige is used.DataAltiExtrapolation_Valery(). The elevation gradients for air temperature returned by CreateInputsModel() are improved.DataAltiExtrapolation_Valery() has been improved. DataAltiExtrapolation_Valery() now runs faster (and by consequence CreateInputsModel() too, when CemaNeige is used).RunModel_CemaNeige(), RunModel_CemaNeigeGR4J(), RunModel_CemaNeigeGR5J() and RunModel_CemaNeigeGR6J() now return air temperature for each elevation layer.S3 plot method defined for OutputsModel objects. It means that the plot_OutputsModel() function is deprecated and his use has been replaced by the use of plot.OutputsModel() or plot().
In plot.OutputsModel() the PlotChoice argument is deprecated and has been renamed which.
plot.OutputsModel() displays air temperature time series for each layer when CemaNeige is used (argument which = “Temp” or “all”).ErrorCrit_*() functions gain a warnings argument to replace the verbose action and the verbose argument now prints the criterion value(s).CreateCalibOptions() when StartParamList or StartParamDistrib arguments are used.CreateInputsModel() now returns an error if NLayers <= 0 when CemaNeige is used.
plot_OutputsModel() now displays raw values on the y-axis when the discharge time series is represented with log scale (formerly, log values of discharges were displayed on the y-axis).
SeriesAggreg() gains a TimeLag argument that corresponds to a numeric value indicating a time lag (in seconds) for the time series aggregation (useful to aggregate hourly time series to the daily time step for instance). In addition, the function now accepts input dates in both POSIXt formats (POSIXct and POSIXlt). The output is in POSIXct format.
plot_OutputsModel() gains a log_scale argument in order to plot the flow with a log scale.
A tutorial is available online on the following link: https://hydrogr.github.io/airGR/. It can also be displayed with the vignette(“airGR”) command.
CreateCalibOptions() loses the OptimParam argument that was redundant with the FixedParam argument. The Calibration_Michel() was modified to take into account this change by using directly FixedParam, but this is transparent to the user.
CreateCalibOptions() loses the StartParam argument that was not used.
sort for the transfo argument of CreateInputsCrit() was not taken into account. It is now fixed.RunModel_GR6J() and RunModel_CemaNeigeGR6J() models were modified back to versions previous to 1.0.1 to prevent from unwanted efficiency criteria deterioration related to the calibration with Calibration_Michel(). The actual model codes were not modified but the TransfoParam_GR6J() and CreateCalibOptions() functions were modified regarding the X5 parameter. It is strongly advised to use airGR 1.0.2 for the RunModel_GR6J() and RunModel_CemaNeigeGR6J() functions if you are using Calibration_Michel(), as they are much more efficient. In case you were using your own calibration algorithm, you will not notice any difference.CreateInputsModel() and DataAltiExtrapolation_Valery() functions now allow both POSIXt formats (POSIXct and POSIXlt).The Calibration_HBAN() and DataAltiExtrapolation_HBAN() functions have respectively been renamed as Calibration_Michel() and DataAltiExtrapolation_Valery() after the names of their creators.
The Calibration_optim() function has been removed from the package.
The silent mode is now defined by the verbose = TRUE argument (formerly quiet = FALSE) in the following functions: Calibration(), Calibration_Michel(), CreateInputsModel(), CreateRunOptions(), DataAltiExtrapolation_Valery(), ErrorCrit(), ErrorCrit_KGE(), ErrorCrit_KGE2(), ErrorCrit_NSE(), ErrorCrit_RMSE(), plot_OutputsModel(), SeriesAggreg().
The GR5J model has been modified: previously, two unit hydrographs were used, now only one is remaining. As a consequence, simulations from the GR5J (RunModel_GR5J() function) and CemaNeige (RunModel_CemaNeigeGR5J() function) models will be different.
An important proportion of the transformations of the parameters have been modified (TransfoParam_*() functions). Since this modifies the local search, calibration results will be different .
The quantiles of the parameters have been recalculated with the new transformations (CreateCalibOptions() function). Since these quantiles constitute the starting point of the calibration algorithm, calibration results will be different.
The Fortran model core codes have been modified:
The list of the contributors and authors is now full.
The references of the package has been updated; they are returned by the following R-command citation(“airGR”).
Fixed bug in CreateInputsModel() that was related to the handling of missing values.
Fixed bug in CreateRunOptions() that prevented the correct use of the IniResLevels argument (to manually set the filling rate of the production and routing stores).
IndPeriod_WarmUp = 0.Three new hydrological models: RunModel_GR4H() function for GR4H (hourly), RunModel_GR2M() function for GR2M (monthly) and RunModel_GR1A() function for GR1A (yearly).
New function SeriesAggreg() to easily aggreg timesteps.
ErrorCrit_RMSE() which led to incorrect calibration (the criterion was maximised instead of minimised).Update of the functions CreateRunOptions(), CreateCalibOptions() and plot_OutputsModel() to handle the new models.
Modification of CemaNeige Fortran code to add an update of Gratio after the SnowPack update (no impact on snow simulation).
Improvement of the plot_OutputsModel() function to allow a selection among available plots.
Minor update in ErrorCrit_KGE() and ErrorCrit_KGE2() to handle case when only one values in not NA.
Update of the scripts in airGR-advanced-example to match the structures of the BasinData objects.
Correction of formatting issue in airGR-advanced-example regarding the “List_HypsoData.txt” file.
quiet = TRUE or FALSE) to choose if the warnings should be suppressed or not.CalibrationAlgo_*() functions were renamed into Calibration_*().Fixed bug in CreateCalibOptions() to handle models with only one parameter.
Fixed bug in Calibration_HBAN(). The function was not working properly with models having only one parameter.
CemaNeige users must now specify one MeanAnSolidPrecip for each elevation layer. The CreateRunOptions() function is impacted.
CemaNeige users can now specify the mean elevation of the input series (before it was always considered equal to the catchment median elevation). The impacted functions are CreateInputsModel() and DataAltiExtrapolation_HBAN().
New architecture with better format verification procedure (using classes) and simpler setting of default configuration.
New architecture where the model, calibration and error functions are in the arguments of the functions (the exotic use of “generic function” created by the users has been removed).
Improved documentation and examples.
Improvements allowing the arrival of new models.
Improvements of the argument verifications in CreateInputsModel(), CreateRunOptions(), CreateInputsCrit(), CreateCalibOptions().
Improvements of all the ErrorCrit() functions to better account for the cases with constant flow values or local zeros.
Improvement of the plot_OutputsModel function (to handle 0 in Qobs and Qsim).
Improved documentation.
Additional functions for results plotting (the {zoo} package is required for some of them).
Add multi-objective calibration using nsga2() (the ‘mco’ package is required).
The field Multiplier has been added in the ErrorCrit() outputs, to indicate whether the criterion is an error (to minimise) or and efficiency (to maximise). This allows to provide real efficiency values in the outputs e.g. NSE[Q] instead of (-1) × NSE[Q].
EfficiencyCrit() have been replaced by ErrorCrit() to avoid misunderstanding (by default, the algorithms minimise the error criterion).RC11 bug correction: the automatic selection of the warm-up period was not working properly when no data was available from warm-up (i.e. when the user had set the run to start at the very first index).
RC10 bug correction: the CalibrationAlgo_HBAN() function was not working in the very rare case when the diagonal search was activated and lead to a set outside the authorised range.
RC9 bug correction: the CalibrationAlgo_HBAN() function was not working properly with models having only one parameter.
RC8 bug correction of the ModelDefaultIniOptions() function (this bug was introduced in the RC7 and caused an error when IndPeriod_WarmUp = NULL.
RC7 bug correction of the ModelDefaultIniOptions() function (the automatic selection of one year for warm-up was not handling properly missing data).
RC6 correction of the help files (the description of CemaNeige parameters were inverted).
RC5 differs from previous releases in the way the data are read and stored (in a list instead of individual vectors). The package is similar, only the examples of Main and the files in MyScriptBlocks have changed. All basin data are now stored inside a list named BasinData. This will greatly ease the future use of Rdata files (instead of txt files) as storage format for the time series of observation.
DefineFunctions_Model() or DefineFunctions_ErrorCrit()).Code improvements to reduce the computation time.
Clearer instructions for the adding and modification of a model.
Improvements of the documentation.
The SelectPer arguments are replaced by IndPeriod to ease understanding.
The PE arguments are replaced by PotEvap() to ease understanding.
The Fsol arguments are replaced by FracSolidPrecip to ease understanding.
R 2.15 in not supported by default.
The check that SelectPer_Run() is continuous is now made in the CheckArg() functions.
Check of the model functioning time step.
Name of the calibration criterion provided in OutputsAlgo().
EfficiencyCrit_NSE_sqrtQ() function to compute NSE criterion on sqrt flows.Incorrect arguments in the call to RunModelAndCrit from CalibrationAlgo_optim_stats and CalibrationAlgo_nlminb_stats.
CalibrationAlgo_nlminb_stats argument was wrongly defined in DefineFunctions_CalibrationAlgo() (optim instead of nlminb).
Format checking for RunOptions was incorrectly made in CheckArg() function.