GenExP-LandSiTes, a software for modelling
neutral agricultural landscapes
F. Le Ber (ENGEES), J.-F. Mari (U. Nancy2), K. Adamczyk (INRA), F. Angevin (INRA), C. Lavigne (INRA)
- Software name: GenExP-LandSiTes
- contact: florence.leber (a) engees.unistra.fr, jfmari (a) loria.fr
- Year first available: 2006
- Hardware required: Java platform edition 6, http://java.sun.com/
- Library required:
- package JRI, http://www.rforge.net/JRI
- software R, http://cran.r-project.org/
- spatial statistic library spastat, http://cran.r-project.org/web/packages/spatstat
- Program language: Java
- Software and supplementary material are downloadable free to use under the GPL, http://www.gnu.org/licenses/gpl.txt
Download (2011 release for windows: R and JRI required)
Download (2009 release for windows: R and JRI required)
GenExP-LandSiTes is a vector-based, spatially explicit model designed to simulate agricultural landscapes, including both their configuration (field pattern) and their composition (occurrence of the categories of land-use). GenExP-LandSiTes has been used in combination with gene flow models for evaluating GM adventitious presence in harvests at a landscape scale (Lavigne et al., 2008; Colbach et al., 2009; Le Ber et al., 2009).
Spatially explicit landscape models have been widely used for ecological or forestry studies (Turner and Gardner, 1991). They would be very useful for studying several processes occurring in an agricultural landscape, such as gene flow, erosion or spread of plant diseases. Most of the existing models are raster-based and thus are rather not adapted for agronomic purposes at a landscape scale. Indeed, agricultural models have to represent and manage geometrical patches and thus should rely on tessellation methods (Gaucherel et al., 2006). We developed GenExP-LandSiTes to fulfill this need. GenExP-LandSiTes allows to simulate various landscapes configurations and compositions, controlling the shape, area or spatial distribution of the fields.
GenExP-LandSiTes is written in Java primarily for its portability and
freely distributed through a Gnu Public Licence (Le Ber et al., 2006a).
The main features of GenExP-LandSiTes are divided in two groups.
- Stochastic field patterns generation: this includes the simulation of spatial point processes, the generation of the field patterns with tessellation methods, and the stochastic simulation of land-use;
- Post-processing: this transforms the mosaic of polygons into a more real
istic mosaic of agricultural fields and collects statistics on the simulated landscapes.
The stochastic field pattern generation
GenExP-LandSiTes can generate various landscapes simultaneously.
Each landscape is represented within a specific interface window. The first step for the user is to specify one or several areas inside the interface window in order to build various field patterns within a single landscape or to create prohibited areas (i.e. areas with non-agricultural use such as forests, buildings,...). The design of roads is also possible. In a second step, the user chooses, for each area, a specific tessellation method, a point process for the seeds generation, and a distribution of land-uses (Figures 1 and 2).
Figure 1: Area selection (red): the menu allows to first parametrise the selected area, then to calculate the chosen tessellation, and finally to visualise the different elements (fields, land cover, centroids,etc.) of the resulting landscape.
Figure 2: Voronoi tessellation to represent a landscape composed of agricultural (colors) and non agricultural areas (white). Each area can be visualised in its own way:
- Seeds are displayed in the west area (blue);
- Plot vertices are displayed in the north-east area (orange);
- Plot centroids are displayed in the
south-east area (red).
Two tessellation methods have been implemented so far.
- The Voronoi tessellation, using the "3D Hull" algorithm for Delaunay
triangulation (O'Rourke, 1998): given a set of points in the Euclidean plan, the set of seeds, a Voronoi tessellation (or Voronoi diagram) is a covering of the Euclidean plan with non-overlapping convex polygons, each of them surrounding a seed. The points in a polygon are closer to its seed than to any other seed located in any other polygon. Therefore the edges consist in the points located at an equal distance from two seeds.
- The rectangular tessellation, using the Mackisack algorithm (Mackisack and Miles, 1996). The basic principle of the algorithm is to generate the edges of the rectangles by crossing two-directional rays starting from a set of points. The area is filled in with non-overlapping rectangles sharing T-vertices.
The generation of points that serve as seeds in the tessellation process
is performed under R by the routines of the spatial statistic library spatstat (Baddeley and Turner, 2005). The JRI package2 was used, allowing communications between the Java methods and the R routines. In order to generate seeds, several options are possible.
- Generation following a Poisson process, the polygon centroids are generated assuming a Poisson process whose parameters are arbitrary set by the user;
- Generation based on a real landscape, the points are sampled in a real landscape (cf. Figure 3-left) on a one point per parcel basis (Adamczyk et al., 2007). These points serve as seeds to simulate virtual landscapes (cf. Figure 3-center). Assuming this set of points, GenExP-LandSiTes can also train stochastic point generation models by means of R routines to simulate new landscapes with similar configurations of seeds or with controlled variable configurations (cf. Figure 3-right).
Figure 3: Two types of tessellations: real field pattern (left), simulated field pattern --Voronoi tessellation-- around real centroids (center), simulated field pattern based on simulated centroids (right). The original map was provided by Institute for the Protection and Security of the Citizen (Joint Research Centre of the European Union) or AUP (Agence Unique de Paiement / French Paiment Agency CAP Support).
Once the landscape configuration is simulated, the user can allocate crops
to the fields according to various methods. As a starting point, the user chooses a distribution of land-uses. Afterwards, the land-use mosaic is built randomly according to this distribution.
GenExP-LandSiTes can also manage stochastic rules of land cover successions. These rules are carried out by a data-mining software (CarrotAge) that treats the crops successions as high-order Markov chains generated by a second-order Hidden Markov Models (HMM2) (Le Ber et al., 2006b). At time 0, GenExP-LandSiTes allocates land covers to the fields according to a distribution provided by the user (for example: 50% maize, 30% wheat, 20% others); then, at successive time slots, crops are allocated to the fields following the crop successions, e.g. (maize / wheat) or (maize / maize), according to the transition probabilities given by the HMM2.
The post-processing and the descriptors
After generating the tessellation, GenExP-LandSiTes performs the following operations:
Finally, GenExP-LandSiTes provides a library to calculate basic landscape descriptors (area, perimeter, number of vertices, centroid, shape index) in order to characterise the field patterns and to compare them with actual landscapes. The results can be represented as histograms and the user can simultaneously visualise statistics concerning several landscapes. Furthermore, the coupling with R is underway to provide high-level tools for statistical analysis of the simulated field patterns.
- Deletion of polygons exceeding landscape limits (clipping);
- Deletion of polygon sides lower than a given threshold (merging of
vertices) in order to simulate realistic agricultural landscapes;
- Selection of the polygons for a statistical study;
Adamczyk, K., Angevin, F., Colbach, N., Lavigne, C., Le Ber, F., Mari, J.-F., 2007. GenExP, un logiciel simulateur de paysages agricoles pour l'étude de la diffusion de transgènes. Revue Internationale de Géomatique 17 (3-4), 469-487.
Baddeley, A., Turner, R., 2005. Spatstat: an R package for analyzing spatial point patterns. Journal of Statistical Software 12 (6), 1-42.
Colbach, N., Monod, H., Lavigne, C., 2009. A simulation study of the medium-term effects of field patterns on cross-pollination rates in oilseed rape (Brassica napus L.). Ecological Modelling 220 (5), 662-672.
Gaucherel, C., Fleury, D., Auclair, D., Dreyfus, P., 2006. Neutral models for
patchy landscapes. Ecological Modelling 197 (1-2), 159--170.
Lavigne, C., Klein, E. K., Mari, J.-F., Le Ber, F., Adamczyk, K., Monod, H., Angevin, F., 2008. How do genetically modified (GM) crops contribute to background levels of GM pollen in an agricultural landscape? Journal of Applied Ecology 45 (4), 1104-1113.
Le Ber, F., Lavigne, C., Mari, J.-F., Adamczyk, K., Angevin, F., 2006. GenExP, un logiciel pour simuler des paysages agricoles, en vue de l'étude de la diffusion de transgènes. In: Actes du Colloque International de Géomatique et d'Analyse Spatiale (SAGEO 2006), Strasbourg.
Le Ber, F., Benoît, M., Schott, C., Mari, J.-F., Mignolet, C., 2006. Studying crop sequences with CarrotAge, a HMM-based data mining software. Ecological Modelling 191 (1), 170-185.
Le Ber, F., Lavigne, C., Adamczyk, K., Angevin, F., Colbach, N., Mari, J.-F., Monod, H., 2009. Neutral modelling of agricultural landscapes by tessellation methods -- application for gene flow simulation. Ecological Modelling 220, 3536-3545.
Mackisack, M., Miles, R., 1996. Homogeneous rectangular tessellations. Advances in Applied Probability 28, 993-1013.
O'Rourke, J., 1998. Computational Geometry in C. Cambridge University Press, 2nd edition.
Turner, M. G., Gardner, R. H. (Eds.), 1991. Quantitative Methods in Landscape Ecology. Vol. 82 of Ecological Studies. Springer.