jannefiluren / waterflows.jl Goto Github PK

When restarting the model, the internal time goes nut (gets add on)...

Should this be included in this package, or in external package? Or even something more general? A shiny app?

Currently, parameter values can be written to files using jld/jld2 format. For an example, see:

https://github.com/jmgnve/ForecastControl/blob/master/calibration_operational.jl#L159

Should we include a more user friendly option?

Shouldn´t we copy the model structure from ODM such that we have completly parallell models?

This is quite tedious. Perhaps Bård can make a c# library that we can call?

Yes, but how should this format look like? For now, we can simply write tables with columns representing time, runoff ...

Right now, lakes are not included in the models. The snow routine should not be used for lakes, whereas some of the subsurface routines, like hbv_light, should include lakes. How to solve this???

Running the model currently only results in simulated runoff. Add option to also add simulated state variables, and other outputs such as actual evapotranspiration.

Should the output states be catchment averages, elevation band averages, or should no aggregation be performed?

The calibration using blackboxoptim is rather slow.

The calibration produces bad results for some basins. See these results:

\hdata\fou\jmg\flood_forecasting\model_calib\model_gr4j\figures\2_634_runoff.png

Perhaps this problem is related to #7

perhaps use metaprogramming to create functions for getting the variables of nested structs

name.($int_var).($int_var)

The model internal time sometimes does not seem to be updated, and sometimes it does, and in excess. For example, during calibration, the model internal time grows to a very large number after all simulations performed by blackbox optim.

Check the internal time
Check the init_states! function. Probably time should be an input to this function.

Use a code like the one below to determine the simulation time step from the input time:

t1 = DateTime(2000,1,1)
t2 = DateTime(2000,1,2)
convert(Int64, Dates.value(t2-t1))/3600e3