Dynamic vegetation models are a common tool to assess the effect of climate and land use change on vegetation. While the current development aims to include more processes, e.g. the nitrogen cycle, the models still typically assume an ample seed supply allowing all species to establish once the climate conditions are suitable. A number of species have been shown to lag behind in occupying climatological suitable areas (e.g. after a change in the climate) as they need to arrive and establish at the newly suitable areas. To reduce discretization errors, seed dispersal needs to be simulated on a sufficiently fine resolution, which in turn increases simulation time for both the local dynamics and in particular for the dispersal from each source cell to each sink cell. Previous attempts to implement migration in dynamic vegetation models have allowed simulating either only small areas or have been implemented as post process, not allowing for feedbacks within the vegetation.
Develop methods to simulate seed dispersal and local dynamics in DGVMs efficiently so that spatiotemporal vegetation dynamics can be simulated over large areas.
We combine two novel alternative methods for seed dispersal simulation with a representative cell approach for simulating local dynamics.
Seed dispersal simulation
Both methods distribute seeds between source and sink grid cells leading to individual establishment in the sink cell.
The first method (FFTM) is a convolution between the produced seeds in the simulated landscape and the seed dispersal kernel, and is based on a Fast Fourier transform.
Figure 1. Upper left panel: example of a seed source. Upper right panel: example of a seed dispersal kernel (here a non-symmetric kernel is assumed), lower left panel: transformed seed dispersal kernel, lower right panel: seed distribution after convolution.
In the second method (SMSM) we mimic the explicit movement of the seeds by iteratively shifting the seed production matrix in all directions and move seeds with a given probability. By multiplying with the seed permeability (Fig.2 left), barriers against dispersal can be simulated.
Fig. 2: Left: Seed dispersal permeability for SMSM simulation tests on an area of 100 * 100 grid cells. Each time the seed matrix is shifted, the probability of entering the new cell probability of seed fall - set to 0.00005 - is multiplied with the seed dispersal permeability of the new potentially entered cell. Right: Spread of Fagus sylvatica from top-left cell using the SMSM method with identical climate and multiplied with the spatially explicit seed dispersal permeability.
Both methods are combined with a method to decrease the computational costs required for the local dynamics in the grid cells, by simulating this local dynamics only in grid cells along migration corridors. The produced seeds are then interpolated between the corridors (Fig. 3) before seed dispersal.
Fig. 3: Sequence of simulation of migration processes in each time step
Both methods, for the seed distribution, FFTM and the SMSM, are considerably faster than the explicit dispersal from each cell to each other (Fig.4). While the FFTM is computationally faster, it does not allow for modification of the seed dispersal kernel parameters with respect to terrain features, which SMSM allows (Fig 2).
Fig. 4: Comparison of runtime in s for seed dispersal for FFTM, SMSM and explicit calculation.
The corridor method produces results similar to the full area simulation (Fig. 5) and reduces computing time proportionally to the fraction of cells with local simulations, e.g. 67% with corridors 10 km apart.
Fig. 5: Left: Seed dispersal permeability for SMSM simulation tests on an area of 100 * 100 grid cells. Each time the seed matrix is shifted, the probability of entering the new cell probability of seed fall - set to 0.00005 - is multiplied with the seed dispersal permeability of the new potentially entered cell. Right: Spread of Fagus sylvatica from top-left cell using the SMSM method with identical climate and multiplied with the spatially explicit seed dispersal permeability.
The new methods are reliable and efficient and open the potential to simulate DGVMs with migration over large areas
Lehsten, V., M. Mischurov, E. Lindström, D. Lehsten and H. Lischke (2019). "LPJ-GM 1.0: simulating migration efficiently in a dynamic vegetation model." Geoscientific Model Development 12: 893-908. doi.org/10.5194/gmd-12-893-2019
2017 - 2018