Hydrological modeling packages in

• topmodel
• dynatopmodel
• TUWmodel
• Ecohydmod
• hydromad (not on CRAN)
• etc.

… and of course airGR & airGRteaching

The models included are simple…

they can be coded from the publications so many versions in many languages exist.

Prone to code mistakes

Objectives of the packages:

• the GR models development team controls the "official" version
• modular functionalities
• Fortran still used for the model cores (faster)
• all other functions in R

GR hydrological models

Rainfall-runoff lumped models developed since the 90's at IRSTEA

• annual: GR1A
• month: GR2M
• day: GR4J, GR5J, GR6J
• hour: GR4H

Snow model

• CemaNeige

Error criteria

• RMSE
• Nash-Sutcliffe
• KGE
• KGE'

Calibration algorithm

• Housemade global then local optimisation

data.frame of 2 basins from 1985 to 2005

summary(dfObs)

##       Date                  P                  E         
##  Min.   :1985-01-01   Min.   : 0.00000   Min.   :0.0000  
##  1st Qu.:1990-01-01   1st Qu.: 0.04297   1st Qu.:0.4354  
##  Median :1995-01-01   Median : 0.90565   Median :1.0481  
##  Mean   :1995-01-01   Mean   : 3.28253   Mean   :1.2721  
##  3rd Qu.:2000-01-01   3rd Qu.: 4.19317   3rd Qu.:1.9221  
##  Max.   :2004-12-31   Max.   :63.58900   Max.   :5.7799  
##                                                          
##       Qmm                 T             Basin     
##  Min.   : 0.03988   Min.   :-13.668   12001:7305  
##  1st Qu.: 0.62357   1st Qu.:  4.254   50002:7305  
##  Median : 1.42465   Median :  7.986               
##  Mean   : 2.25734   Mean   :  7.814               
##  3rd Qu.: 2.76480   3rd Qu.: 11.480               
##  Max.   :38.97774   Max.   : 24.174               
##  NA's   :1

Coding can be a hard task for some students

• slows down the understanding of the concepts presented during courses
• limited time to teach programming

airGRteaching contains only very basic functions

B12001 <- dfObs[dfObs$Basin == "12001", 
                c("Date", "P", "E", "Qmm", "T")]
PREP <- PrepGR(ObsDF = B12001, 
               HydroModel = "GR4J", CemaNeige = TRUE)

dyplot(): dynamic chart to explore data

dyplot(PREP)

Model calibration on a given period, with the KGE' criterion

CAL <- CalGR(PrepGR = PREP, CalCrit = "KGE2",
             WupPer = c("1989-01-01", "1989-12-31"), 
             CalPer = c("1990-01-01", "1993-12-31"))

## Grid-Screening in progress (0% 20% 40% 60% 80% 100%)
##   Screening completed (729 runs)
##       Param =  432.681 ,   -0.020 ,   20.697 ,    1.417 ,    0.002 ,    4.961
##       Crit KGE'[Q]      = 0.8025
## Steepest-descent local search in progress
##   Calibration completed (36 iterations, 1138 runs)
##       Param =  320.368 ,    0.405 ,   35.589 ,    1.191 ,    0.002 ,    4.589
##       Crit KGE'[Q]      = 0.8518

Plotting calibration process

plot(CAL, which = "iter")

Plotting performances

plot(CAL, which = "perf")

To check if the calibration is correct, we can perform an evaluation on an independent period (Klemes split-sample test)

SIM <- SimGR(PrepGR = PREP, CalGR = CAL, EffCrit = "KGE2",
             WupPer = c("1993-01-01", "1993-12-31"),
             SimPer = c("1994-01-01", "1998-12-31"))

## Crit. KGE'[Q] = 0.8679

##  SubCrit. KGE'[Q] cor(sim, obs, "pearson") = 0.8941 
##  SubCrit. KGE'[Q] cv(sim)/cv(obs)          = 1.0469 
##  SubCrit. KGE'[Q] mean(sim)/mean(obs)      = 0.9365

Plotting the resulting discharge time series

dyplot(SIM)

List of 2 basins data.frame

colSelect <- c("Date", "P", "E", "Qmm", "T")
listObs <- list(B12001 = dfObs[dfObs$Basin == "12001", colSelect],
                B50002 = dfObs[dfObs$Basin == "50002", colSelect])

Launching the Shiny interface for 2 basins

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

Only daily models available

Possibility to put the GUI on the Web with a Shiny server

airGR allows for more advanced or systematic tests

Use of:

• bootstrap technique for calibration
• additional calibration algorithms

Possibility to set:

• some model parameters
• initial reservoir levels
• internal states
• etc.

Tutorial

https://webgr.irstea.fr/airGR-website

Use of optimisation algorithms from other R packages

See the "V02.1_param_optim" vignette for implementation

optDE <- DEoptim::DEoptim(fn = OptimGR4J,
                          lower = lowerGR4J, upper = upperGR4J)
optPSO <- hydroPSO::hydroPSO(fn = OptimGR4J,
                             lower = lowerGR4J, upper = upperGR4J)

airGRteaching can be used to:

• play with the models parameters or understand the models components
• help someone very bad at coding to make hydrological modeling

airGR can be used to:

• make more advanced hydrological modeling
• plug the rainfall-runoff models with your own tools

License: GPL-2

Coron, L., Thirel, G., Delaigue, O., Perrin, C., Andréassian, V., 2017: The Suite of Lumped GR Hydrological Models in an R package, Environmental Modelling & Software, 94, 166-171, DOI: 10.1016/j.envsoft.2017.05.002.

Contact: airGR@irstea.fr