Running a forecast

As any RHEAS simulation, we begin by creating a configuration file. In your favorite text editor create a new file, let’s assume that its name is forecast.conf. The first section contains the simulation options and has a forecast header:

[forecast]

Let’s assume that we are requesting both a hydrologic and agricultural simulation. In that case, we need to set both models (VIC and DSSAT) as an option

[forecast]
model: vic, dssat

We then set the starting and ending dates of the forecast

[forecast]
model: vic, dssat
startdate: 2001-4-1
enddate: 2001-6-30

Let’s also assume that we have a shapefile of the study domain (the sample data contain such a file in the data/tests directory). Moreover, let’s set the name of the simulation (that name will be useful to retrieve the output from the database).

[forecast]
model: vic, dssat
startdate: 2001-4-1
enddate: 2001-6-30
basin: data/tests/basin.shp
name: basin

The model spatial resolution is next, here set as 0.25 degrees (~25 km).

[forecast]
model: vic, dssat
startdate: 2001-4-1
enddate: 2001-6-30
basin: data/tests/basin.shp
name: basin
resolution: 0.25

In order to capture the uncertainty in the forecast, we perform an ensemble simulation and set the ensemble size to 10. RHEAS will run the VIC ensemble in parallel (using multiple threads), thus reducing computation time. The ensemble simulation also requires a method to generate the meteorological forcings ensemble (currently, model uncertainty due to parameter errors is not implemented). Available methods are the Ensemble Streamflow Prediction (ESP, resampling the climatology); resampling from the IRI meteorological forecasts; bias-correcting and downscaling the CFSv2 meteorological forecasts.

[forecast]
model: vic, dssat
startdate: 2001-4-1
enddate: 2001-6-30
basin: data/tests/basin.shp
name: basin
resolution: 0.25
ensemble size: 10
method: esp

These are all the necessary options to define our forecast simulation. Then we need to define the parameters for the two models (VIC and DSSAT).

We begin with the vic section

[vic]

The generation of the meteorological forecast ensemble requires a base dataset for each of the required variables. Therefore, we need to define the source for precipitation, temperature and wind speed

[vic]
precip: chirps
temperature: ncep
wind: ncep

We can opt to initialize the ensemble of VIC models from a a state file saved in the database, by randomly selecting a model initial condition from climatology, or by starting a 1-year model run. All these options do not need to be set by the user apart from

[vic]
precip: chirps
temperature: ncep
wind: ncep
initialize: yes

If RHEAS cannot find previous model states in the database, it will by default run a 1-year simulation for each ensemble member.

We choose the model output, specifically net shortwave radiation and soil moisture, to the database

[vic]
precip: chirps
temperature: ncep
wind: ncep
initialize: yes
save to: db
save: net_short, soil_moist

The DSSAT section dssat requires fewer options than VIC, with the domain shapefile being the first

[dssat]
shapefile: data/tests/basin.shp

and the ensemble size being another parameter that can optionally be set. The forecasts from VIC are randomly paired with each of the DSSAT ensemble members in order to capture uncertainty both in the DSSAT model parameters and the hydroclimatological forcings. Similarly to a nowcast, assimilation cna be enabled/disabled with an option.

[dssat]
shapefile: data/tests/basin.shp
ensemble size: 50
assimilate: no

Finally, let’s run the system (inside the rheas directory)

./bin/rheas forecast.conf