GR4J¶

This page is the technical reference for the GR4J rainfall-runoff model used by the GR4J node. For configuration syntax, properties, and result outputs, see the node page. This page covers the underlying model structure, equations, parameter behaviour, and calibration guidance.

Model overview¶

GR4J ("Génie Rural à 4 paramètres Journalier") is a lumped, conceptual, daily rainfall-runoff model developed at INRAE (formerly Cemagref). It transforms catchment-average rainfall and potential evapotranspiration (PE) into daily streamflow using four calibrated parameters and a fixed structure of two non-linear stores and two unit hydrographs.

GR4J is widely adopted for its parsimony, computational efficiency, and good performance across a range of climates. It was developed and refined through testing on hundreds of catchments; the structure described here corresponds to the version published in Perrin et al. (2003).

Conceptual structure¶

GR4J represents the catchment as a chain of conceptual processes:

Net inputs. Rainfall and PE are netted off against each other, so only one is non-zero on any timestep.
Production store (capacity X1) handles soil-moisture accounting, partitioning rainfall between catchment storage and percolation.
Percolation drains the production store via a non-linear flux.
Split. Effective rainfall is split 90% / 10% between two parallel routing paths.
Unit hydrographs. UH1 (base time X4) on the 90% path; UH2 (base time 2·X4) on the 10% path.
Routing store (reference capacity X3) on the 90% path, also non-linear.
Groundwater exchange (controlled by X2) adds or removes water from both routing paths, representing intercatchment groundwater transfer.

Mathematical formulation¶

In the equations below, P is daily rainfall [mm], E is daily PE [mm], and S, R are the current states of the production and routing stores [mm].

Net inputs¶

Only one of these is non-zero on any timestep:

If P ≥ E: Pn = P − E, En = 0
If P < E: Pn = 0, En = E − P

Production store¶

Fraction of Pn entering the production store:

Ps = X1 · (1 − (S/X1)²) · tanh(Pn/X1) / (1 + (S/X1) · tanh(Pn/X1))

Fraction of En drawn from the production store:

Es = S · (2 − S/X1) · tanh(En/X1) / (1 + (1 − S/X1) · tanh(En/X1))

Update S: S ← S − Es + Ps

Percolation:

Perc = S · { 1 − [ 1 + (4S / 9X1)⁴ ]^(−1/4) }

Update S: S ← S − Perc

Effective rainfall and split¶

Pr = Perc + (Pn − Ps)

90% of Pr enters UH1; 10% enters UH2.

Unit hydrographs¶

UH1 has base time X4 days; UH2 has base time 2·X4 days. Both are derived from a power-law S-curve and convolved with their respective inputs to produce delayed outputs Q9 (from UH1) and Q1 (from UH2).

Groundwater exchange¶

F = X2 · (R/X3)^(7/2)

F can be positive (water gained) or negative (water lost), depending on the sign of X2.

Routing store¶

Update R: R ← max(0, R + Q9 + F)

Outflow from the routing store:

Qr = R · { 1 − [ 1 + (R/X3)⁴ ]^(−1/4) }

Update R: R ← R − Qr

Direct flow¶

Qd = max(0, Q1 + F)

Total streamflow¶

Q = Qr + Qd

Q is the model's daily runoff depth [mm]. In Kalix, this is reported as runoff_depth and converted to runoff_volume [ML] using the catchment area.

Parameters¶

The four parameters are passed to the GR4J node in this order: X1, X2, X3, X4.

X1 — Production store capacity [mm]¶

Maximum capacity of the soil-moisture (production) store. Larger X1 means more buffering of rainfall and slower drying of the catchment.

Typical bounds: 10 to 6000 mm
Sensitivity: usually the most important parameter for water balance. Low values produce flashy systems with little soil-moisture memory.

X2 — Groundwater exchange coefficient [mm/day]¶

Magnitude (and sign) of the intercatchment exchange flux. Positive X2 means water is gained by the system; negative X2 means water is lost.

Typical bounds: −20 to +10 mm/day
Note: X2 is conceptual rather than directly observable. It compensates for water-balance closure errors and sub-surface losses or gains the model otherwise cannot represent.

X3 — Routing store reference capacity [mm]¶

Reference (one-day-ahead maximum) capacity of the routing store. Larger X3 means slower routing response.

Typical bounds: 1 to 4000 mm
Sensitivity: controls the recession behaviour of simulated flow.

X4 — Unit hydrograph time base [days]¶

Base time of UH1 (UH2 has base time 2·X4). Controls the timing of the runoff response.

Typical bounds: 0.5 to 20 days. Some studies use a lower bound as small as 0.04 days for very flashy catchments.

Calibration guidance¶

Initial states. GR4J is sensitive to the initial values of S and R. Use a warm-up period of at least one year (longer in arid catchments) to remove their influence.
Joint calibration. All four parameters are typically calibrated jointly because of correlations between them. X1 and X3 trade off when balancing fast and slow flow components; X2 trades off against the production store to close the water balance.
Identifiability. X1 and X4 are usually well identified. X2 and X3 can be less well identified, especially in catchments with stable baseflow.
Objective functions. NSE on flows emphasises high flows; NSE on log-flows or KGE-on-log emphasises low flows; KGE balances bias, correlation, and variability. Multi-objective calibration is common.

Strengths and limitations¶

Strengths¶

Parsimonious — only four free parameters.
Computationally cheap.
Well validated on hundreds of catchments globally.
Good general performance across a wide range of climates and catchment scales.

Limitations¶

Lumped — no spatial representation of within-catchment variability.
Daily timestep — not appropriate for sub-daily flood-event modelling without modification.
X2 absorbs water-balance closure errors; it is not a directly measurable physical property.
No snow processes; couple with a snow module (e.g. CemaNeige) upstream of inputs for snow-affected catchments.
Performance can degrade in arid catchments with long zero-flow periods, and in heavily regulated systems where the rainfall–runoff relationship is broken.

Implementation notes¶

The Kalix GR4J node passes the four parameters in the order X1, X2, X3, X4 (see the GR4J node page).
All internal calculations are in mm; conversion to ML uses the configured catchment area.
Model states (S and R) are carried between timesteps and initialised at the start of the simulation.

References¶

Perrin, C., Michel, C., and Andréassian, V. (2003). "Improvement of a parsimonious model for streamflow simulation." Journal of Hydrology 279(1–4): 275–289.

Edijatno, Nascimento, N. de O., Yang, X., Makhlouf, Z., and Michel, C. (1999). "GR3J: a daily watershed model with three free parameters." Hydrological Sciences Journal 44(2): 263–277.

Le Moine, N. (2008). Le bassin versant de surface vu par le souterrain : une voie d'amélioration des performances et du réalisme des modèles pluie–débit? PhD thesis, Université Pierre et Marie Curie (Paris 6).