The framework and settings for the framework

Using the .ini file

A number of settings of the framework can be set in the ini file for each model. The settings are explained in the section below.

Settings in the run section

Information for the current run can be given in the run section. Here the start and end-time of the run as well as the timestep can be given. Alternatively a link to a Delft-FEWS runinfo.xml file can be given. An example is shown below.

[run]
#either a runinfo file or a start and end-time are required
#runinfofile=runinfo.xml
starttime= 1995-01-31 00:00:00
endtime= 1995-02-28 00:00:00
# required, base timestep of the model
timestepsecs = 86400
#Indicates the model is running from Delft-FEWS or not

If this section is not present and a runinfo.xml is also not used you will need to specify the number of timesteps using the -T option on the command line (for most models).

Settings in the framework section

The in/output file formats can be specified in the framework section. At present only pcraster mapstacks and netcdf are available fot input. See the supplied pcr2netcdf.py script for information on the layout of the netcdf files. If netcdf files are used the name of the mapstack is used as the standardname in the netcdf file.

[framework]
# outputformat for the *dynamic* mapstacks (not the states and summary maps)
# 1: pcraster
# 2: numpy
# 3: matlab
outputformat=1

# netcdfoutput requires also outputformat = 1 (default) and additionally the name of the file
# read the mapsstacks from a netcdf
netcdfinput= inmaps.nc
#Write to netcdf:
netcdfoutput = outmaps.nc
# Write summary maps to netcdf
netcdfstaticoutput = staticoutmaps.nc
# Write states to netcdf
netcdfstatesoutput = states.nc
#Read states from netcdf
netcdfstatesinput = instates.nc
#netcdfwritebuffer=100

As can be seen from the example above a number of input/ouput streams can be switch on to work with netcdf files. These are:

  • netcdfinput. Time dependant input. This does not work for climatology files at the moment

  • netcdfoutput. Time dependant output.

  • netcdfstaticoutput. Summary output at the end of a run, those that normally end up in the outsum directory

  • netcdfstatesoutput. The model’s state variables at the end of a run.

  • netcdfstatesinput. The model’s input state variables at the start of a run.

To enhance performance when writing netcdf files a netcdfwritebuffer can be set. The number indicates the number of timesteps to keep in memory before flusing the buffer. Setting the buffer to a large value may induce memory problems.

Settings in the API section

In the ini file example below several variables are configured to be available via the API. For most settings this only defines what the API will expose to the outside world. However, if you specify 0 (input) as a role for one of the forcing variables the wf_readmap function will no longer read maps from disk for that variable but will return the contents of that variable via the API.

The API section specifies variables that are exposed via the api. Use the following convention:

variable_name_in_model=variable_role,variable_unit
role: 0 = input (to the model)
      1 = is output (from the model)
      2 = input/output (state information)
      3 = model parameter
unit: 0 = mm/timestep
      1 = m^3/sec
      2 = m
      3 = degree Celcius
      4 = mm
      5 = -

Use role 0 for input maps to the model (forcing data that are normally read from disk) only, role 1 for outputs, role 2 for state variables and role 3 for model parameters. The units may be choose freely and be strings also.

example:

[API]
FreeWater=2,4
SoilMoisture=2,4
UpperZoneStorage=2,4
LowerZoneStorage=2,4
InterceptionStorage=2,4
SurfaceRunoff=2,m^3/sec
WaterLevel=2,2
DrySnow=2,4
Percolation=1,0
ForecQ_qmec=0,1
PERC=3,5
FC=3,4
# Below are the forcing variables. By putting these here you MUST
# supply them via the API, if not these will default to 0.0
#P=0,0
PET=0,0
TEMP=0,3

Settings in the modelparameters section

Most of the time this section is not needed as this will mostly be configured in the python code by the model developer. However, in some case this section can be used alter the model for example force the model to read RootingDepth from an external data source. Not all models support this. You can check if the model you uses support this by looking for the wf_updateparameters() function in de model code.

The format of entries in this section is as follows:

name=stack,type,default,verbose,[lookupmap_1],[lookupmap_2],lookupmap_n]
  • name - Name of the parameter (internal variable, without the self.)

  • stack - Name of the mapstack (representation on disk or in mem) relative to case

  • type - Type of parameter (default = static)

  • default - Default value if map/tbl is not present

  • verbose - If set to 1 (True) the maps a log message will be generated is a default values is used

    instead of a map

  • lookupmap - map(s) [0 to n] to be used in the lookuptable in the case the type is tbl

Possible parameter types (the second option)are:

  • staticmap: Read at startup from map

  • statictbl: [deprecated] Read at startup from tbl, fallback to map (need Landuse, Soil and TopoId (subcatch) maps)!

  • tbl: Read at startup from tbl.

  • tblts: Lookup tables for each timestep, including initial section

  • tblsparse: Lookup tables for each timestep, including initial section. Fills in missing timestep using previous timestep

  • tblmonthlyclim: read a tbl file corresponding to the current month (12 maps in total)

  • timeseries: read map for each timestep

  • monthlyclim: read a map corresponding to the current month (12 maps in total)

  • dailyclim: read a map corresponding to the current day of the year (366 maps in total)

  • hourlyclim: [not implemented yet] read a map corresponding to the current hour of the day (24 maps in total)

  • tss: read a tss file and link to lookupmap (only one allowed) a map using timeinputscalar

Example:

[modelparameters]
RootingDepth=monthlyclim/ROOTS,monthlyclim,75,0
# Force the model to read monthly climatology of P
Precipitation=inmaps/P,monthlyclim,0.0,1

Example:

[modelparameters]
RootingDepth=monthlyclim/ROOT,monthyclim,100,1
Sl=inmaps/clim/LCtoSpecificLeafStorage.tbl,tbl,0.5,1,inmaps/clim/LC.map
Kext=inmaps/clim/LCtoSpecificLeafStorage.tbl,tbl,0.5,1,inmaps/clim/LC.map
Swood=inmaps/clim/LCtoBranchTrunkStorage.tbl,tbl,0.5,1,inmaps/clim/LC.map
LAI=inmaps/clim/LAI,monthlyclim,1.0,1

Settings in the variable_change_timestep/once section

In the two sections “variable_change_timestep” and “variable_change_once” you can set operations on parameters and variable that are executed at the start of each timestep or once in the initialisation of the model respectively. What you specify here should be valid python code and include variable that exists in the model you are using. This only works if the actual model you are using includes the wf_multparameters() function. At the moment wflow_hbv, wflow_sbm, wflow_w3ra and wflow_routing include this. See below for a configuration example. Some models may also support this via the -P and -p command-line options.

[variable_change_timestep]
self.Precipitation = self.Precipitation * 1.2
# Mutiplies the precipitation input times 1.2 every timestep

[variable_change_once]
self.PathFrac = self.PathFrac * 1.1
# Increases the paved area fraction of the model by 10 %

Settings in the [rollingmean] section

The rollingmean section allows you to define a rolling mean for each variable in the model. This variable can be used by other applications (e.g. data assimilation) or you can report it as output. Example:

[rollingmean]
self.Surfacerunoff=12

The above will make a 12 timestep rollingmean and store this in the variable self.Surfacerunoff_mean_12

Settings in the summary_* sections

By adding variable in one or several of these sectiosn the framework will save these variables to disk (using the value at the end, sum, min, max or avg) at the end of a run.

the available sections are:

  • summary - Saves the actual value of the variable

  • summary_avg - Saves the average value over all timesteps of the variable

  • summary_sum - Saves the sum over all timesteps of the variable

  • summary_min - Saves the minimum value over all timesteps of the variable

  • summary_max - Saves the maximum value over all timesteps of the variable

All maps are saved in the outsum directory of the current runid.

Example:

[summary]
self.MaxLeakage=MaxLeakage.map
# Save and average these per LU type

[summary_sum]
self.Precipitation=Sumprecip.map

[summary_max]
self.Precipitation=maxprecip.map

[summary_min]
self.Temperature=mintemp.map

[summary_avg]
self.Precipitation=avgprecip.map

Settings in the outputtss/outputcsv sections

[outputcsv_0-n] [outputtss_0-n]

Number of sections to define output timeseries in csv format. Each section should at least contain one samplemap item and one or more variables to save. The samplemap is the map that determines how the timeseries are averaged/sampled. The function key specifies how the data is sample: average(default), minimum, maximum, total, majority. The timeformat key can either be steps or datetime.

All other items are variable=filename pairs. The filename is given relative to the case directory.

Example:

[outputcsv_0]
samplemap=staticmaps/wflow_subcatch.map
self.SurfaceRunoffMM=Qsubcatch_avg.csv
function=average
# average is the default
timeformat = datetime
# steps is the default

[outputcsv_1]
samplemap=staticmaps/wflow_gauges.map
self.SurfaceRunoffMM=Qgauge.csv
self.WaterLevel=Hgauge.csv

[outputtss_0]
samplemap=staticmaps/wflow_landuse.map
self.SurfaceRunoffMM=Qlu.tss
function=total

In the above example the discharge of this model (self.SurfaceRunoffMM) is saved as an average per subcatchment, a sample at the gauge locations and as an average per landuse.

[run] section: The use of date and time

Available options in the [run] section

The run section can contain information about the model timesteps, the date/time range, how to initialize the model and how interpret the forcing data.

[run]
# either a runinfo file or a start and end-time are required
#runinfo=runinfo.xml
starttime=2016-01-01 00:00:00
endtime=2016-01-04 00:00:00
# Base timestep of the model in seconds, default = 86400
timestepsecs = 86400
#start model with cold state
reinit=1
# Default behaviour: steps
runlengthdetermination=steps

Date and time and timesteps

The original pcraster framework has no notion of date and time, only timesteps that are used to propagate a model forward. However, to be able to support the BMI and netcdf files date and time functionality has been inserted into the framework.

As most of the forcing in hydrological modelling is an accumulation the date/time of an input time series is assumed to represent the total (or average) of that variable one timestep length back in time from the timestamp.

For example, if the forcing data has the following four timestamps the model will run four timesteps. The first timesteps will be to propagate the state from T0 to T1 (in that case the date/time of the state files is assumed to be T0). As such the state going into the model should be valid for the T0, see graph below:

digraph G {
    node [shape=box]

    subgraph cluster_1 {
            node [style=filled];
                " "  -> T1 -> T2 -> T3 -> T4;
                label = "Model steps";
                color=blue
        }

    subgraph cluster_2 {
            node [style=filled];
                "  " ->"1 Jan 2016" -> "2 Jan 2016" -> "3 Jan 2016" -> "4 jan 2016";
                label = "Forcing data";
                color=blue
        }

    subgraph cluster_3 {
            node [style=filled];
                "  " ->" 1 Jan 2016" -> " 2 Jan 2016" -> " 3 Jan 2016" -> " 4 jan 2016";
                label = "Model output data";
                color=blue
        }

    "Model state: T0 31 Dec 2015";


}

Here the first column shows the model steps and the second column the timestamp of the input/output data for that timestep (a empty box means there is nothing for that step). The first empty-row is regarded as the timestamp of the initial conditions. The first forcing data is used to propagate the model from T0 (the state, 31 Dec 2015) to T1 (1 Jan 2016) which will also be the first output the model writes.

The above corresponds to the following date/time in the [run] section:

[run]
# either a runinfo file or a start and end-time are required
#runinfo=runinfo.xml
starttime=2016-01-01 00:00:00
endtime=2016-01-04 00:00:00
# required, base timestep of the model
timestepsecs = 86400
#start model with cold state
reinit=1
# Default behaviour: steps
runlengthdetermination=steps

The above shows the default behaviour of the framework. For each data point in the input forcing a model steps is performed. The ‘runlengthdetermination’ variable in the run section can also be set to ‘intervals’. In that case the number of steps is determined from the number of intervals in the forcing data. Hence, the following run will be performed:

digraph G {
    node [shape=box]

    subgraph cluster_1 {
            node [style=filled];
                " "  -> T1 -> T2 -> T3;
                label = "Model steps";
                color=blue
        }

    subgraph cluster_2 {
            node [style=filled];
                "1 Jan 2016" -> "2 Jan 2016" -> "3 Jan 2016" -> "4 jan 2016";
                label = "Forcing data";
                color=blue
        }

    subgraph cluster_3 {
            node [style=filled];
                " " -> " 2 Jan 2016" -> " 3 Jan 2016" -> " 4 jan 2016";
                label = "Model output data";
                color=blue
        }

    "Model state: T0 01 Jan 2016";


}
[run]
# either a runinfo file or a start and end-time are required
#runinfo=runinfo.xml
starttime=2016-01-01 00:00:00
endtime=2016-01-04 00:00:00
# required, base timestep of the model
timestepsecs = 86400
#start model with cold state
reinit=1
# Default behaviour: steps
runlengthdetermination=intervals

In this case the forcing data has the same input as in the previous case (4 timestamps) but in this case the first timestamp is regarded as the time of the initial conditions. As such, one timestep less (now three in total) is performed and the forcing data from 01 Jan 2016 is NOT used as it is regarded as the initial state. The first output point will be 02 Jan 2016 generated from the forcing marked as 2 Jan 2016.

The same applies when the start and end time of the model run are supplied via the bmi interface.

Settings in the netcdfmetadata section

All items in this section are copied as global attributes into the netcdf output file. Example:

[netcdfmetadata]
license=https://opendatacommons.org/licenses/odbl/
note=Test runs, results are not final

wf_DynamicFramework Module