Forecast

Rameau is able to run an ensemble forecast. Forecast can be activated in the TOML configuration file.

[forecast]
forecast = true

Forecast settings

Emission date

The date and time for which to issue a forecast is defined through the emission_date key. If omitted, the default value corresponds to the last date of the meteorological input data.

[forecast]
emission_date = 2009-01-01

Scope

The scope key is the time between the emission_date and the end of the probabilistic forecast.

Default value is one day.

[forecast]
scope = { days = 30 } # Format : { days = 30, minutes = 0, hours = 0, seconds = 0 }

Ensemble forecast members

Rameau uses the past climatology to produce the ensemble forecast members. Each year of the past climatology corresponds to a member. The years to consider to form the forecast ensemble members are provided through the year_members key. If not provided, all years in record are considered.

For example, we consider a 30-day forecast issued on 2009-01-01. Inputs provide meteorological data from 2000-01-01 to 2009-12-31. Therefore, the forecast ensemble will consists in 10 simulations all starting from 2009-01-02 using the following meteorological inputs:

  • 1st member: from 2000-01-02 to 2000-01-31

  • 2nd member: from 2001-01-02 to 2001-01-31

  • …

  • …

  • 10th member: from 2009-01-02 to 2009

We could have also used the year_members key for selecting specific years, for example 2000 and 2001 to have a 2-members ensemble forecast.

[forecast]
year_members = [2000, 2001]

No rain member

An additional member using a rain-free scenario can be added to the ensemble by turning the norain key to true.

Default value is false.

[forecast]
norain = false

Quantiles output

By default, Rameau provides river flow and groundwater level forecast outputs for each member. If the quantiles_output key is set to true, outputs correspond to quantiles defined by the quantiles key.

Default value to false for quantiles_outputs.

Default values for quantiles are 10%, 20%, 50%, 80% and 90%.

[forecast]
quantiles_output = false
quantiles = [10, 20, 50, 80, 90]

Forecast correction method

Two correction methods are available for the ensemble forecast. The key correction can be set to one of the following:

  • “no” does not apply any correction

  • “halflife” applies a correction to the forecast : the forecast values are corrected to match the observed values on the emission date. The initial correction is then propagated through time by applying a decay factor to the difference between the forecasted and observed values.

Default value is “no”.

[forecast]
correction = "no"

Pumping date to consider

Pumping can be applied to the forecast ensemble members. The pumping_date key is used to define the date from which to apply the pumping.

[forecast]
pumping_date = 1999-07-01

Forecast correction parameters

Forecast correction parameters correspond to the halflife times of the output correction method when issuing a forecast. Two keys forecast.river.halflife and forecast.groundwater.halflife for river flows and groundwater levels, respectively, can be set for each watershed of the model. These keys accept float values. A value of zero means no half-life correction.

Default values are 0.

[watershed.1]
forecast.river.halflife = 2
forecast.groundwater.halflife = 1
[watershed.2]
forecast.river.halflife = 0
forecast.groundwater.halflife = 1

forecast.river.halflife and forecast.groundwater.halflife can be automatically estimated based on a simple algorithm that tests a range of values and select the best. Two keys forecast.river.halflife_estimation and forecast.groundwater.halflife_estimation accepting boolean values can be set for activate this estimation when issuing a forecast.

Default values are false.

[watershed.1]
forecast.river.halflife_estimation = true
[watershed.2]
forecast.river.halflife_estimation = true
forecast.groundwater.halflife_estimation = false

Note that these parameters are taken into account only if the key correction is set to halflife in the forecast settings.

A full example

The example below set the forecast settings for a 30-day forecast issued on 2009-01-01 using the years 2000 and 2001 as ensemble members. A rain-free scenario is added to the ensemble. The forecast outputs are quantiles 25%, 50%, and 75%. Halflife output method is applied to groundwater level and river flow for watershed 1, and only to river flow for watershed 2.

[forecast]
forecast = true
emision_date = 2009-01-01
scope = { days = 30 }
year_members = [2000, 2001]
norain = true
quantiles_output = true
quantiles = [25, 50, 75]
correction = "halflife"
pumping_date = 1999-07-01
[watershed.1]
forecast.river.halflife = 1
forecast.groundwater.halflife = 1
[watershed.2]
forecast.river.halflife = 2
forecast.groundwater.halflife = 0