A scattering of seeds grows into an algorithmic forest.
ANS Studio developed a “Seed Scattering System” to create a ‘natural’ distribution of plants for the garden surrounding an office building. The ‘natural’ stipulation here is not understood in terms of an untended garden or of a naturalistic formal pattern, but as a distribution that locates plants and chooses species in a manner that replicates the end-result of the localised growth and succession processes that occur in a forest. This was not the sole focus though — the model made heavy use of parameterised goals and constraints to allow for other design criteria to affect the distribution.
What distinguishes this project from many other approaches is the sophistication of the modelling process and the use of the model as the key design driver spanning from site analysis to design documentation. Rather than using landscape form as the key site of design investigation (and have analysis performed in response to changes) the model itself embodied the process of form development, with the designer instead choosing amongst possible solutions and adjusting input weights.
The model itself performed a number of steps when creating a possible design. Broadly speaking the first phase was in identifying how environmental conditions, such as soil composition, building shading, and wind sheltering, affected different portions of the site. Follow from this the design logic was developed, whereby the designer could adjust parameter’s values and possible layout patterns for how the plant placement would respond to the environmental conditions. Finally the system would take all of these into account to create the planting plan, with the algorithm’s primary outputs being the ‘seed’ points that represented a plant with a particular spacing and species optimised to the given local conditions.1 The location of the pathway system occurs after this distribution and is optimised to work around root systems.
By developing the bulk of the design within a parameter system, the way in which planting plans were developed changed. The designers were not looking to “manipulate geometries or compositions of tree groupings but to design the fundamental rules that underlie them.”2 Doing so also forced an explicit trade-off between performance-driven and aesthetic-driven design criteria through the weighted parameters of the model. Modifications to the design criteria later on in the process can be easily accommodated by regenerating the design solution with the new parameters, such as when the entrance spaces needed to become more prominent.3
As the design criteria could be encompassed in a relatively complete manner by choices in the vegetation — and their direct relationship to pathing — the design logic could be encapsulated neatly in a weighted model, particularly when both the initial and final site conditions were flat in formal terms. However, the use of a singular model becomes much more difficult as the design criteria become more multivalent, and come to encompass wider ranges of criteria. In these cases, and in cases where the design has more formal effects, the use of formal representations to link between different performances criteria becomes key. That is to say, rather than have a single model encompass and generate a design from the first to last stages of the design process, a formal representation (e.g. a surface in a CAD program) are used to store the results of these models or to negotiate between multiple models investigating different criteria.
Tsukasa Takenaka and Aya Okabe, “Development of the Seed Scattering System for Computational Landscape Design,” International Journal of Architectural Computing 9, no. 4 (2012): 431.↩
Takenaka and Okabe, “Development of the Seed Scattering System for Computational Landscape Design,” 434–435.↩
Takenaka and Okabe, “Development of the Seed Scattering System for Computational Landscape Design,” 432.↩