Unclear message when Qupstream is NA

When we have NA in upstream discharge (observations), the error message is unclear.

For a run (I modified the following in the example InputsModel$Qupstream[5000] <- NA) :

OutputsModel <- RunModel_Lag(InputsModel = InputsModel,
+                              RunOptions = RunOptions, Param = Lag)
Error in if (any(OutputsModel$Qsim < 0)) { : 
  valeur manquante là où TRUE / FALSE est requis

For a calibration (according to @cyril.thebault):

Grid-Screening in progress(0%Error in FUN_MOD(InputsModel, RunOptions = RunOptions, Param = Param[iFirstParamRunOffModel:length(Param)]):
NA/NaN/Inf in foreign function call (arg2)

@olivier.delaigue and @david.dorchies, I think we need to better check NAs. In my opinion, NAs should be forbidden, such as are NAs for P and T.

added SUGGESTION label

@charles.perrin suggests putting as NAs the downstream simulated discharge over the two time steps corresponding to the lagged upstream NA discharge value. That would allow evaluating and calibrating downstream discharge (as in .Error_Crit we ignore time steps for which obs or sim are NAs, through the TS_ignore <- !is.finite(VarObs) | !is.finite(VarSim) command line).

To discuss!

Calculation of error criteria and plotting work properly when Qsim contain NAs.

Based on that and on the interest of allowing NAs in Qupstream, I'm in favour of modifying the code accordingly.

Update to first message:

The error message in Calibration was due to something else.

Regarding the run, if we do things properly (see below), we get an explicit error message before running the model:

> Qamont[[1]]$Qsim[1034] <- NA
> InputsModel2 <- CreateInputsModel(FUN_MOD = RunModel_GR5H, DatesR = temp2$Date, 
+                                  Precip = temp2$pmm, PotEvap = temp2$ETP,
+                                  Qupstream = matrix(c(Qamont[[1]]$Qsim,Qamont[[2]]$Qsim),ncol=2), 
+                                  LengthHydro = LengthHydro,
+                                  BasinAreas = BasinAreas)
Error in CreateInputsModel(FUN_MOD = RunModel_GR5H, DatesR = temp2$Date,  : 
  'Qupstream' cannot contain any NA value

The confusion comes from that it's forbidden to have NAs in Qupstream for the creation of the InputsModel object but if you correctly set the InputsModel object without any NA and after change the value of InputsModel$Qupstream then RunModel_Lag will crash.

I agree with the solution proposed by @charles.perrin which is:

• to accept NAs at the creation of the InputsModel object with a warning: "the parameter Qupstream contains NA values, this will lead to NA values in the ouputs of RunModel"
• to have a robust RunModel_Lag that returns NA values for the 2 time steps concerned by these inputs.

Good morning everyone, what is the branch of airGR where the NA are allowed in Qupstream (if the modifications are already done) ?

Hi @cyril.thebault,

I didn't do it already. As said in !31 (comment 35873) , I'm presently waiting that the merge request !31 (merged) is solved to start a new branch with this implementation.

@olivier.delaigue can you quickly merge !31 (merged) ?

I finally solved this ticket in !31 (comment 35873). It is merged in dev branch.

CreateInputsModel produces a warning. And also RunModel_Lag except when use in calibration.

changed milestone to %v1.6.11

mentioned in merge request !31 (merged)

mentioned in commit 68ce1825

mentioned in commit 0035e63c

mentioned in commit a9edee4b

closed via merge request !31 (merged)