The model SWIF (Soil Water In Forested ecosystems) is presented. SWIF is a model for the simulation of water flow in the unsaturated soil zone, including water extraction by roots and lateral saturated drainage. The model gives a halfimplicit finite difference solution to the flow equation. For minimizing computing time, the model chooses its integration time steps based on the rate of change of soil water conditions. The model is not a conceptually new model. However, it meets the requirements of an entirely modular programme in which new theories can easily be implemented. The hydrology of the canopy and the forest floor is described in independent modules.
Simulation results are evaluated in relation to temporal and vertical discretization, using a simulation of soil water dynamics in a forest over a period of two growing seasons. Both the site description and the simulation results are given in an accompanying paper . In another paper a sensitivity analysis of the model is given .
In order to evaluate the effects of temporal discretization of boundary condition's, the length of the interval of constant averaged boundary conditions was varied from one hour to one month. The simulated annual transpiration is found to be dependent on this interval for a wet year, but not for a dry year. When the boundary conditions are taken constant for intervals shorter than one day, simulated soil water contents show a highly dynamic behaviour during and directly after heavy rainfall. Then small integration time steps are required and computing time increases strongly.
For the evaluation of the effects of vertical discretization, the number of soil layers was varied from 5 to 40. The yearly transpiration appears to be almost independent of the layer thickness as long as the horizon boundaries are not altered, but the temporal distribution of transpiration and water contents are affected. The required computing time for the simulation of the water balance of a full growing season increases from 2.5 minutes for a simulation of five layers to 18 minutes for a simulation of 40 layers.
When model results are evaluated on a yearly basis, they appear to be not very sensitive to the temporal and vertical discretization. Only if research aims require more detailed model results on a shorter timescale, higher computing times are to be accepted.