Numerical Simulation, An Art of Prediction, Volume 2. Jean-François Sigrist

scales in order to obtain an overview: from plants or planting groups, to the plot and the entire farm. This is a vast field of research in digital agriculture. Modeling by L-systems already contributes to the evaluation of certain techniques, such as agro-ecology, which are receiving increasing interest due to the growing awareness of ecological issues and the need to preserve the emerald forest?

      NOTE.– Growing plants “in-silico” with L-systems.

      The typology of a plant, the phenotype, results from the expression of its genetic heritage (its genotype), and its interactions with the characteristics of the environment in which it develops (its environment). These interactions largely determine biomass production: reconstituting the phenotype of plants is therefore a key-factor in calculating the yield of a production.

      Artificial intelligence techniques based on deep learning from imaging data can contribute to this objective by automatically and quickly performing repetitive tasks such as counting sheets. In the learning phase, it is necessary to have a database large enough to make the algorithms efficient, which is not always the case in agronomy! The databases that can be used are generally limited and campaigns to enrich them are very expensive. One solution proposed by some researchers is to generate digital plants by simulation: the variety of forms produced thus enriches existing databases at low cost (Figure 1.23).

      Figure 1.23. The virtual plants (left), obtained in silico by means of L-systems, have similar characteristics to the real plants (right), obtained in vitro

      COMMENT ON FIGURE 1.23.– The efficiency of the L-systems is such that researchers show that the simulations are able to produce a variability in the characteristics of synthetic plants close to that of real plants – otherwise the data used by the learning algorithms would not be of good quality. The researchers even demonstrate that the latter learn, with similar effectiveness, either from real data or from data produced by synthetic models [UBB 18].

      Let us conclude this chapter with the understanding that, in general, agricultural modeling addresses three scientific issues:

       – understand and predict plant growth processes;

       – assess the impact of agricultural practices in ecological and economic terms;

       – informing the policies of decision makers and farmers’ choices.

      They are thus becoming an essential tool for agricultural research, meeting the vital needs of humanity in the 21st Century: feeding populations and preserving their environment.

      1 1 Available at: https://www6.inra.fr/record.

      2 2 Data available at: http://www.fao.org/forestry.

      3 3 Available at: http://www.theplantlist.org/.