Tutorials
Under construction(1) Getting started
(2) Namelists
(3) Verification
(4) Input data
(5) New schemes
Configurations
Configurations are defined in ./namelists/default/configurations.nl. A configuration is
composed of the following post-processing components: Selection, correction, discrete and
continuous uncertainty, and calibration. In addition there are a number of other
Options
Each configuration has many options that can be specified. The most important ones are:
selector, correctors, continuous, and
lowerDiscrete (for Precipitation). The defaults for the other other options should at
least get you started.
| Attribute | Type | Default | Description |
|---|---|---|---|
| numDaysParameterSearch | int | 1 | How many days back should parameters be searched for if yesterday's are missing? A value of 1 means use yesterday's only. |
| parameterIo | std::string | Which IO scheme should be used for storing/retrieving parameters? Default is to store them in memory only. | |
| pooler | std::string | Which scheme for pooling observations for estimating parameters should be used? Default to using one set of parameters at each observation location. | |
| spreader | std::string | Which scheme should be used to spread parameters spatially and temporaly? Default to using the scheme that pools observations together. | |
| selector | std::string | Which scheme should be used to select the ensemble? | |
| correctors | std::string[] | Which correction schemes should be used? | |
| averager | std::string | Which scheme should be used to collapse distribution to deterministic forecast? Default to distribution mean | |
| updaters | std::string[] | Which scheme should be used to update the probability distribution using recent observations? | |
| smoothers | std::string[] | Which scheme should be used to smooth the timeseries at the end? | |
| continuous | std::string | Which scheme should be used to create the continuous part of the distribution? | |
| discreteLower | std::string | Which scheme should be used to create discrete probability at the lower boundary? Can be used together with 'continuous' and discreteUpper | |
| discreteUpper | std::string | Which scheme should be used to create discrete probability at the upper boundary? Can be used together with 'continuous' and discreteLower | |
| discrete | std::string | Which scheme should be used to create discrete probability? Only valid for pure discrete variables, and cannot be used together with 'continuous' | |
| calibrators | std::string[] | Which scheme should be used to calibrate the distributions? | |
| numOffsetsSpreadObs | int | 0 | Across how many offsets should observations be allowed to be spread? For some stations, the obs occur less frequent than the output offsets. In these cases Parameters are usually never updated. Allow obs to be taken from neighbouring offsets. |
| downscaler | std::string | Which downscaler should be used? If not specified, the downscaler from the run will be used, and if that is not speicified, the nearest neighbour approach is used. |
Optional.
Example
As an example, consider the following configuration declaration:clim selector=sClim downscaler=dNearest continuous=mm1
Here the clim selection scheme, the dNearest downscaler and the
mm1 continuous probability model has referenced from
./namelists/default/schemes.nl.
Note that this configuration was named clim, after the selector that it uses. It makes
sense to call the configuration after what it mostly tries to represent.