## A Spatial Model of Community Dynamics in Patchy Landscapes | |||||||||

## Principal InvestigatorIan R. Noble Research School of Biological Sciences |
This study investigates how the interaction between biological characteristics, habitat pattern, and disturbance regime controls the dynamics of diversity in plant communities. The main goals are to improve the theoretical understanding of the mechanisms of maintenance of species diversity in spatio-temporally variable environments, and to develop models to predict the effects of changes in disturbance regimes on landscape-scale patterns of species diversity. Our study focusses on aspects of plant biology relating to regeneration after disturbance such as seed dispersal and germination or vegetative colonization. The modelling is based on the use of functional groups for response to disturbance, characterised by common biological attributes that lead to similar responses to a given disturbance type. The dynamics of species assemblages in landscape with varying spatial patterns and disturbance frequencies is simulated in a spatially-explicit manner to take into account local interactions, dispersal, and spread of disturbance. | ||||||||

## Co-Investigators | |||||||||

## Sandra LavorelCentre d'Ecologie Fonctionnelle et Evolutive, CNRS Montpellier France | |||||||||

## Projectsp55 -VP | |||||||||

| |||||||||

## What are the results to date and the future of the work?During the past year we have been preparing the migration from entirely theroretical models (work from 1991 to 1995) to models that will include greater biological detail such as the effects of different plant morphologies on their ability to colonise space freed up by disturbances. This phase has implied substantial experimental work and data analyses so that the modelling was halted for most of the year. This activity will be resumed in 1997 with the development of our next generation of models. One application of the previous model versions to the development of reed (Phragmites australis) beds after abandonment of rice fields in southern France was however developed during this year. This application also allowed us to specifiy and code the new submodel for vegetative colonisation. Simulations showed that 1) reed colonisation is much slower than was hoped by local managers 2) the expansion of the beds is essentially dependent on rhizome growth while stolon, initially hypothesised to play a role in long distance colonisation, contribute very little due to their rarity. Our production was halted by the transition to theVPP because we had not scheduled time for algorithm adaptation. Redevelopment of algorithms needs to be carried out to improve and adapt our large number of stochastic procedures. | |||||||||

## - Appendix A | |||||||||

| |||||||||

## What computational techniques are used?The dynamics is modelled at the individual level, for a landscape consisting of a patterned lattice of habitat sites. First, a landscape pattern is generated using a hierarchical dichotomy for suitable verses unsuitable spatial units. At each time step, the algorithm sweeps across the landscape to calculate the destinations of all seeds produced by established individuals, updates the composition of the seed bank at each site, and draws the adults for the next generation. Calculations are mainly convolutions of basic operations and functions on large arrays; the algorithm also involves numerous random draws. Population maps after a fixed number of generation are stored and analyzed using algorithms of point pattern analysis. The investigation of stochastic processes requires replicated experiments. Hence, a run for a given parameter set consists of a set of 5 simulations for a given set of parameters. Inter-simulation mean and variance for a number of synthetic descriptors are calculated to analyze the dynamics of species coexistence. The program is written in FORTRAN 77. Since the algorithm relies on convolution of operations on large arrays (100 x 100), vectorizing appears to be particularly well suited to reduce calculation time. Investigating stochastic processes requires that seeds are dispersed individually and independently. Such a procedure, when in scalar form, is extremely time-expensive (several thousands calculations per time step!), and in fact takes over 2/3 of total simulation time. After reprogramming for maximum vectorization and flexibility to further sophistications, vectorization is still only partial (about 60% at present), due to unavoidable recursivity in the seed bank dynamics. We are intending to investigate in 1997 the conversion of our algorithm to FORTRAN 90. ## PublicationsLavorel, S., Gachet, S., Mesleard, F., Grillas, P. .
| |||

| |||