The mass balance between eroded, deposited and transferred material in a cell of CIDRE. T is the rate of sediment that by-pass a cell. W is the flow width considered as equal to dx in the simulations presented here. D is the rate of sediment deposition on the cell. E is the rate of soil or river bed detachment. QsL is the sediment discharge resulting from lateral erosion (lateral to one of the black arrows corresponding to one direction of water flow). Si is the local topographic gradient or slope. qs is the incoming sediment discharge per unit width (After Carretier et al., 2016).


The landscape evolution simulations presented on this site have been carried out using the CIDRE code (e.g. Carretier et al.2016). This website has not been created as an apology for CIDRE. There are other freely usable models with remarkable performances of which a non-exhaustive list is given in the table at the end of this page. All LEMs applied over millions of years have common points (stationarity of water flows in most of the cases, general shape of erosion laws) but also different levels of simplification concerning hydrology and sediment transport which can vary. Some algorithms are frighteningly fast, while others favour the addition of certain processes (e.g. lateral erosion in CIDRE). A feature of CIDRE that exists in few models is its ability to ”spread” water runoff (Multiple flow algorithm instead of steepest descent algorithm), which can therefore diverge and more realistically account for the flow over foothills and plain areas.

In CIDRE, topography is represented by a mesh of square cells (or pixels). Each cell has an altitude and a constant size. Basically, CIDRE is injected with a rain grid or a constant value, an initial topography, a tectonic uplift rate grid or value, an erodibility grid or a value and CIDRE calculates over time the evolution of the topography, the quantity eroded or the quantity of sediments deposited on every cell of the grid.

In practice, over a time step (about 100 years), CIDRE runs the cells in the decreasing direction of the altitudes. It propagates the quantity of water fallen by the rain to calculate a water discharge flowing on each cell, and it also calculates the slopes between this cell and its lower neighbours. From the steepest slope and from the water discharge CIDRE calculates the amount that can be eroded, and the amount of moving sediment from higher up that will be deposited. In addition, the water flow can erode the lateral neighbours cells (bank erosion). Eroded and transiting material is transferred to the lower cells in proportion to the slope. Then CIDRE moves to the cell that is lower in the list and so on until all the cells have been treated. An increment of tectonic uplit is added to elevations. Then a new time step starts and the same process is repeated until the total duration of the model is reached.


Starting from an initial topography, the modification of the topography proceeds by successive time steps. During a time step, precipitation falls on the grid at a rate P [LT-1] and a multiple flow algorithm propagates the water flux Q [L3T-1], toward all downstream cells in proportion to the slope in each direction. Then the elevation z (river bed or hillslope surface) changes on each cell (size dx) according to the balance between erosion ϵ [LT-1] and deposition D [LT-1]. The erosion is different for sediment and for bedrock and ϵ is the sum of two values, one corresponding to gravitational processes without involving the runoff, usually dominating on the hillslopes, and another one associated with water discharge, typically dominating in rivers. Water flowing in one direction is also able to detach material from the cells located perpendicular to that direction to simulate river bank erosion. This erosion generates a lateral (bank) sediment discharge Qsl [L3T-1] toward the cell where the water is flowing. Finally, elevation changes also by adding an uplift U [LT-1] (subsidence if negative).

In the following, all the water and sediment fluxes are ”per unit width” [L2T-1], the width being either a calculated sub-pixel flow width, or the pixel size dx. All the simulations in this website were run using this second option.

The rate of elevation change on a cell is determined by the following mass balance equation (e.g. Davy and Lague2009Carretier et al.2016Shobe et al.2017):

∂z- =  - ϵr - ϵh + Dr + Dh - dqsl+ U             (1)
∂t                         dx

where the subscript ”r” (”river”) denotes rates associated with flowing water and ”h” (”hillslope”) denotes rates that depends only on the topographic gradient or slope S. Then we define a constitutive law for each of these components: (Carretier et al.2016)

          m  n
ϵr  =  Kq  S  for river processes               (2)
ϵh  =  κS for hillslope processes                (3)

where K [L1-2mTm-1], κ [LT-1] are erodibility parameters, m and n are lithology-dependent (different for bedrock or sediment) erosion parameters, S is the slope, q [L2T-1] is the water discharge per stream unit width, and

Dr   =  -ζq for river processes                    (4)
Dh   =  ---dx----for hillslope processes            (5)
         1- (S∕Sc)2

where qsr and qsh are the incoming river and hillslope sediment fluxes (total qs = qsr + qsh) per unit width [L2T-1], ζ is a river transport length parameter [T L-1] and Sc is a slope threshold. These fluxes are the sum of sediment fluxes leaving upstream neighbour cells while the deposition rates on a cell are a fraction of the incoming sediment.

Concerning the river processes, ϵr is known as the stream power law and derives from the assumption that ϵr is proportional to a power law of the shear stress applied by the flowing water on the river bed (e.g. Whipple et al.2000Lague2014). The deposition rate Dr is a fraction of the incoming sediment flux and this fraction (ζq) has the dimension of the reverse of a length. We call this length a transport length because it has the physical meaning of a characteristic distance over which a volume of detached material will transit downstream before being deposited. In particular, when the local q is large, few sediment eroded from upstream will deposit on the cell. The transport length depends on ζ, proportional to the reverse of a settling velocity of sediment in water (e.g. Davy and Lague2009Lajeunesse et al.2013). In instantaneous river models, ζ should be fixed by the grain size of sediment. In landscape evolution models, where the water discharge q averages the periods with and without transport, ζ is an ”apparent” parameter that can take a large range of values in real situations depending on climate variability (Guerit et al.2019).

Concerning the hillslopes processes, the philosophy is the same, except that the detachment rate ϵr and the deposition rate Dr depend only on the slope. The linear slope dependence of ϵr describes diffusion processes. Dr depends on a specified critical slope Sc: when the slope is close to Sc, the deposition rate Dr decreases rapidly, simulating in average the onset of shallow landslides. The transport length associated with gravitational processes (---dx--2
1-(S∕Sc)) is inversely proportional to the probability to deposit sediment on the cell. This erosion-deposition formulation leads to similar solutions as the critical slope-dependent hillslope model studied for example by Roering et al. (1999) (Carretier et al.2016).

Flowing water in each direction can erode lateral cells perpendicular to that direction. Little is known about the law that describes the widening rate of valley, and establishing a lateral erosion law suitable for landscape evolution models averaging processes over millennia is a challenge (Langston and Tucker2018Langston and Temme2019). Here, the lateral sediment flux per unit length qsl [L2T-1] eroded from a lateral cell is simply defined as a fraction of the river sediment flux qsr [L2T-1] in the considered direction (e.g. Murray and Paola1997Nicholas and Quine2007), assuming that lateral mobility of channels, and thus lateral erosion, increases with the flux of river sediment (Bufe et al.20162019):

qsl =  α qsr                         (6)

where α is a bank erodibility coefficient. α is specified for loose material (sediment) and is implicitly determined for bedrock layers such that the ratio of the lateral erodabilities is equal to the ratio of the fluvial ones (αloose∕αbedrock = Kloose∕Kbedrock , with K from Equation 2). If sediment covers the bedrock of a lateral cell, α is weighted by its respective thickness above the target cell.

Finally, the sediment leaving a cell is spread in the same way as water, i.e. proportionally to the downstream slopes. This procedures starts from the most elevated cell and ends with the lowest cell and is repeated in the next time steps until the end of the specified model time (Myr in our case).

Parameter values for the reference simulation

The reference simulation is a constantly uplifting domain grid of 40x40 km2 (200x200 cells of size 200 m).

Associating erodibility parameters to a particular rock is challenging. The best that we can do in the current state of our knowledge is to fix the erodibility parameters so that the relief is realistic. We impose here only one bedrock type for which the erodibility is K = 10-4 m-0.5 yr-1, and the water discharge and slope exponent are m = 0.5, n = 1. The bedrock erodibility by pure gravitational processes is κ = 10-4 m yr-1. For the sediment (previously deposited but that can be reeroded), we use K = 10-3 m-0.5 yr-1, m = 0.5, n = 1 and κ = 10-4 m yr-1.

The transport length parameter ζ is set to 1 yr m-1, and corresponds to a low value for natural systems (median at 17 yr/m Guerit et al.2019). It is difficult to link ζ with physical properties of sediment because ζ changes according to the variability of transport periods, but low values seem to correspond to temperate perennial rivers (Guerit et al.2019).

The lateral erosion parameter α to 0. Finally, the critical slope is Sc =tan(40o). The northern southern sides are open and fixed to z = 0 m. Periodic boundary conditions are imposed on the two other sides meaning that water and sediment leaving on the one side is reinjected on the other one.

U is fixed to 10-3 m yr-1 and P to 1 m yr-1.

The final time is fixed to 7 Ma for the reference simulation, a duration allowing the topography to reach a dynamic equilibrium, ie. a state where the topography does not change in average and the erosion balances uplift. For some simulations, this time is longer because the topography takes more time to reach a dynamic equilibrium.

Existing LEMs




Willgoose et al. (1991)


Beaumont et al. (1992)


Chase (1992)


Howard et al. (1994)


Tucker and Slingerland (1994)


Braun and Sambridge (1997)


Coulthard et al. (1998)


Densmore et al. (1998)


Tucker and Bras (2000)


Crave and Davy (2001)


Carretier and Lucazeau (2005)


Coulthard et al. (2013)


Goren et al. (2014)


Carretier et al. (2016)


Campforts et al. (2017)


Shobe et al. (2017)


Davy et al. (2017)


Yuan et al. (2019)

Table 1: Non-exhaustive list of published LEMs, updated from Tucker and Hancock (2010).


   Beaumont, C., P. Fullsack, and J. Hamilton (1992), Erosional control of active compressional orogens. In K.R. Mc Clay (Ed.), thrust Tectonics, 1-18 pp., Chapman and Hall, New Yotk.

   Braun, J., and M. Sambridge (1997), Modelling landscape evolution on geological time scales: a new method based on irregular spatial discretization, Basin Res., 9, 27–52.

   Bufe, A., C. Paola, and D. W. Burbank (2016), Fluvial bevelling of topography controlled by lateral channel mobility and uplift rate, Nature Geoscience, 9(9), 706.

   Bufe, A., J. M. Turowski, D. W. Burbank, C. Paola, A. D. Wickert, and S. Tofelde (2019), Controls on the lateral channel-migration rate of braided channel systems in coarse non-cohesive sediment, Earth Surf. Proc. Land., 44(14), 2823–2836, doi:10.1002/esp.4710.

   Campforts, B., W. Schwanghart, and G. Govers (2017), Accurate simulation of transient landscape evolution by eliminating numerical diffusion: the TTLEM 1.0 model, Earth Surface Dynamics, 5(1), 47–66, doi:10.5194/esurf-5-47-2017.

   Carretier, S., and F. Lucazeau (2005), How does alluvial sedimentation at range fronts modify the erosional dynamics of mountain catchments?, Basin Res., 17, 361–381, doi:10.1111/j.1365-2117.2005.00270.x.

   Carretier, S., P. Martinod, M. Reich, and Y. Goddéris (2016), Modelling sediment clasts transport during landscape evolution, Earth Surf. Dynam., 4, 237–251, doi:10.5194/esurf-4-237-2016.

   Chase, C. G. (1992), Fluvial landsculpting and the fractal dimension of topography, Geomorphology, 5, 39–57.

   Coulthard, T., M. Kirkby, and M. Macklin (1998), Non-linearity and spatial resolution in a cellular automaton model of a small upland basin, Hydrology And Earth System Sciences, 2(2-3), 257–264, doi:10.5194/hess-2-257-1998.

   Coulthard, T. J., J. C. Neal, P. D. Bates, J. Ramirez, G. A. M. de Almeida, and G. R. Hancock (2013), Integrating the LISFLOOD-FP 2D hydrodynamic model with the CAESAR model: implications for modelling landscape evolution, Earth Surf. Proc. Land., 38(15), 1897–1906, doi:10.1002/esp.3478.

   Crave, A., and P. Davy (2001), A stochastic ”precipitation” model for simulating erosion/sedimentation dynamics, Computer and GeoSciences, 27(7), 815–827.

   Davy, P., and D. Lague (2009), The erosion / transport equation of landscape evolution models revisited, J. Geophys. Res., 114, doi:10.1029/2008JF001146.

   Davy, P., T. Croissant, and D. Lague (2017), A precipiton method to calculate river hydrodynamics, with applications to flood prediction, landscape evolution models, and braiding instabilities, J. Geophys. Res. Earth Surface, 122(8), 1491–1512, doi:10.1002/2016JF004156.

   Densmore, A., M. Ellis, and R. Anderson (1998), Landsliding and the evolution of normal-fault-bounded mountains, J. Geophys. Res., 103(B7), 15,203–15,219.

   Goren, L., S. D. Willett, F. Herman, and J. Braun (2014), Coupled numerical-analytical approach to landscape evolution modeling, Earth Surf. Proc. Land., 39(4), 522–545, doi:10.1002/esp.3514.

   Guerit, L., X.-P. Yuan, S. Carretier, S. Bonnet, S. Rohais, J. Braun, and D. Rouby (2019), Fluvial landscape evolution controlled by the sediment deposition coefficient: Estimation from experimental and natural landscapes, Geology, 47(9), 853–856, doi:10.1130/G46356.1.

   Howard, A. D., W. E. Dietrich, and M. A. Seidl (1994), Modeling fluvial erosion on regional to continental scales, J. Geophys. Res., 99, 13,971–13,986.

   Lague, D. (2014), The stream power river incision model: evidence, theory and beyond, Earth Surf. Proc. Land., 39(1), 38–61, doi:10.1002/esp.3462.

   Lajeunesse, E., O. Devauchelle, M. Houssais, and G. Seizilles (2013), Tracer dispersion in bedload transport, Advances in GeoSciences, 37, doi:10.5194/adgeo-37-1-2013.

   Langston, A. L., and A. J. A. M. Temme (2019), Bedrock erosion and changes in bed sediment lithology in response to an extreme flood event: The 2013 Colorado Front Range flood, Geomorphology, 328, 1–14, doi:10.1016/j.geomorph.2018.11.015.

   Langston, A. L., and G. E. Tucker (2018), Developing and exploring a theory for the lateral erosion of bedrock channels for use in landscape evolution models, Earth Surface Dynamics, 6(1), 1–27, doi:10.5194/esurf-6-1-2018.

   Murray, A. B., and C. Paola (1997), Properties of a cellular braided-stream model, Earth Surf. Proc. Land., 22, 1001–1025.

   Nicholas, A., and T. Quine (2007), Modeling alluvial landform change in the absence of external environmental forcing, Geology, 35, 527–530, doi:10.1130/G23377A.1.

   Roering, J. J., J. W. Kirchner, and W. E. Dietrich (1999), Evidence for nonlinear, diffusive sediment transport on hillslopes and implications for landscape morphology, Wat. Resour. Res., 35, 853–870.

   Shobe, C. M., G. E. Tucker, and K. R. Barnhart (2017), The SPACE 1.0 model: a Landlab component for 2-D calculation of sediment transport, bedrock erosion, and landscape evolution, Geoscientific Model Development, 10(12), 4577–4604, doi:10.5194/gmd-10-4577-2017.

   Tucker, G., and R. Bras (2000), A stochastic approach to modeling the role of rainfall variability in drainage basin evolution, J. Geophys. Res., 36-7, 1953–1964.

   Tucker, G., and R. Slingerland (1994), Erosional dynamics, flexural isostasy, and long-lived escarpments : A numerical modeling study, J. Geophys. Res., 10, 12,229–012,243.

   Tucker, G. E., and G. R. Hancock (2010), Modelling landscape evolution, Earth Surf. Proc. Land., 35(1), 28–50, doi:10.1002/esp.1952.

   Whipple, K. X., G. S. Hancock, and R. S. Anderson (2000), River incision into bedrock: Mechanics and relative efficacy of plucking, abrasion and cavitation, Geol. Soc. Am. Bull., 112, 490–503.

   Willgoose, G., R. L. Bras, and I. Rogdriguez-Iturbe (1991), Result from a new model of river basin evolution, Earth Surf. Proc. Land., 16, 237–254.

   Yuan, X. P., J. Braun, L. Guerit, D. Rouby, and G. Cordonnier (2019), A New Efficient Method to Solve the Stream Power Law Model Taking Into Account Sediment Deposition, J. Geophys. Res. Earth Surface, 124(6), 1346–1365, doi:10.1029/2018JF004867.

Close Menu