The structure of the GRP model is based on stores and has three modules:
- A snow accumulation and melt module: useful for snow-influenced catchments, this module (called CemaNeige) is placed on top of the GRP model. It splits liquid and solid precipitations and manages snow accumulation and melt over a few ranges of altitudes, using air temperature.
- The production module, which includes three functions:
- The interception of precipitation (P) by potential evapotranspiration (E),
- A production (soil-moisture accounting) store with fixed capacity (CAP = 275 mm), which determines excess rainfall and produces a percolation Perc, which combines into effective rainfall Pr,
- A volumetric modification of the effective rainfall by an adjustment factor, CORR.
At every time step, rainfall P is first intercepted by potential evapotranspiration E. The part of P that is not intercepted by E is used to feed the production store and/or to produce excess rainfall. The repartition of P into these two parts Ps and Pn-Ps is directly determined based on the filling rate of the production store. Net evapotranspiration En is used to calculate actual evapotranspiration from the production store. Perc and Pn-Ps are added to form effective rainfall Pr, which is multiplied by the adjustment factor CORR to take into account possible water exchanges with the underground, which gives the corrected effective rainfall P’r.
- The transfer module, which follows the production module, spread P’r in time. It consists of:
- a unit hydrograph (HU) (with time base TB), which accounts for the lag between rainfall and streamflow
- a quadratic routing store of capacity ROUT, which temporally smoothes effective rainfall.
The model has two main inputs, precipitation and potential evapotranspiration, calculated at the catchment scale. Temperature is also used in case the snow module is applied. Streamflow is the only output of the model.
Running GRP in forecasting mode
The GRP model can be run in classic simulation mode, just considering meteorological inputs (rainfall, potential evapotranspiration, temperature). However, this application mode can yield large discrepancy between observed and simulated streamflow, which are not compatible with operational requirements.
In a forecast context, better results are obtained using also the information given by the last observed streamflow available at the time of issuing forecast time. We call this use of information data assimilation. In GRP, it is done in two consecutive parts:
- The update of the routing store to a level that produces a forecast streamflow exactly equal to the observed streamflow. GRP has a single flow component, producing an unequivocal relationship between the simulated streamflow and the routing store level.
- A correction of the streamflow Q from the routing store that can be done in two ways: using either an autoregressive-type correction (called Tangara method), or the Artificial Neural Networks method. The first method calculates the relative error between observed and the previous forecast streamflow put at power 0.45 and applies this coefficient to every future forecasts. The second method takes as input observed streamflow at the forecast time, and additive errors between observed and forecast streamflow at t-1 and t, where t is the time step just before the forecast time.
This data assimilation is considered as an integral part of the structure of the model for forecasting objectives. In forecasting mode, observed streamflow is an input to the model in addition to meteorological inputs. GRP also requires one (or multiple) scenario(s) of future precipitation over the catchment.
Parameters and calibration
The model has three parameters to optimize:
- CORR (-): the adjustment factor of effective rainfall which contributes to find good water balance (by accounting for possible water exchange with underground)
- TB (h): the unit hydrogram (UH) time base used to account for the time lag between rainfall and streamflow
- ROUT (mm): the capacity of the routing store, which temporally smoothes effective rainfall.
The parameters are calibrated in forecasting mode, i.e. with the application of the updating procedure. Thus parameter calibration is made for a specific lead time (horizon).
To know more: check our publications.