Pruned Precedents & Growing Potentials

Studying a niche interest within a niche discipline can be a challenge. Especially when a much larger subject (architecture) develops an overbearing fascination with both that interest (computational design) and that discipline (landscape architecture). This disciplinary mixer has ramifications for my research: if I’m aiming to show that we can use computational design in a way that is distinctly landscape architectural then that begs the question of what differences would create such a distinction. And so, I find myself drawn into the arguments for disciplinary autonomy and the ever-present question of ‘what exactly is landscape architecture?’

Many recent answers to this question rely on critique: that this or that is that wrong way of thinking about or working within landscapes. Here we often see a reaction to the obscuring influence that architecture exerts on the discourse surrounding the design of landscapes and cities. Where these critiques get especially interesting is that many of them take issue with the methods that I seek to appropriate in my own work. Typically, I agree with the gist of these commentaries: that there are pragmatic and conceptual failings to how these methods are used. However, the question often remains whether these flaws are irredeemable, and if not, what kind of improvements can be made.

Hence this post. It’s the first in a series that aims to identify these critiques, where I think they strike true, and how this influences the direction of my own work. The starting point for today’s post is Julian Raxworthy’s recently completed doctoral thesis: Novelty in the Entropic Landscape.

This new disciplinary proximity is most evident in a fascination with change and time, expressed in terms such as “dynamism”, “mobility”, “process” and “flexibility” that have featured prominently in publishing in both areas since the mid-1990s. This body of thinking and practice I identify as the “Process Discourse”…

…The rise of the process discourse has seen the adoption of ecological models of process, and their generalisation into algorithms, which are now incorporated into architectural-design generation process in order to give designs some of the qualities of dynamism that natural systems possess. Architectural genres such as parametricism and datascape work to model dynamic landscape forces in the design-generation process, but do not change when made as actual structures in reality. I argue in this dissertation that by simulating change, projects underpinned/informed by the process discourse do not exhibit the key philosophical property of change, which is the spontaneous emergence of novelties. This property is recognised in the language used by the process discourse by terms such as “emergence”, but is effectively ignored in the simulatory models of architectural form.¹

Here, straight from the first page of the first chapter, is why his Julian’s work is of interest. His simultaneous investigation of the process-centric conceptions of design in both digital architecture and landscape urbanism is one that has been largely unexplored, despite the close ties between these two schools of thought.

When I first looked at this intersection, I thought that process-driven simulations can be useful to landscape architects because they can help us understand and generate ‘emergent’ outcomes within complex systems.² So where do Julian and I diverge?

Our apparent differences appear in the second chapter of the thesis, where Julian covers and criticises the contemporary ‘process discourse.’

He begins with an excellent discussion of the use of layering as a generative design strategy, as developed by Daniel Libeskind, Ian McHarg, Bernard Tschumi, OMA, and James Corner. Thereafter, he introduces the present state of the process-discourse as espoused by landscape urbanists, and how this has influenced landscape architecture and urban design. Here he highlights³ the different strains of thought that exist on either side of the Atlantic, where Penn, Harvard, and The Landscape Urbanism Reader square off against the Architectural Association and A Manual for the Machinic Landscape.

I’m going to skip past the discussion of landscape urbanism’s conceptual flaws, as I generally agree with Julian and particularly Peter Connolly’s^{4 5} assessment that it imports overly-architectural⁶ concepts which fog its claims to be a ‘new lens by which to view the contemporary city’. Instead, I will focus on this chapter’s discussion of the particular techniques promoted by this discourse.

I have sympathy for the charge that landscape urbanism’s methods suffer from a ‘scientism’ whereby site analysis sets up a ‘Cartesian sieve’ that rejects all unquantifiable phenomena present. While the qualitative and affecting elements of landscapes are important and often marginalised, it isn’t a gap that I’m addressing in my own work. Instead, I’m staying put on the systems-theory and ecology side of the fence.

Which is not to say that the problem of instrumentality is not a problem. But given that the technical/objective approach will always have some place within the design process, it’s far easier to limit the aims of my own work to ‘do these tools provide useful insights into the technical workings of landscape systems’ rather than to also make the case that they can simulate subjective aspects of a design, the design process, or the design’s users.

This brings us to the third model of landscape, which is landscape as virtual simulation of processes. Driven by its desire to reveal hidden systems, this superficial view of site ignores other systems and concrete site features or specificities, such as property ownership or geographical features, limiting information to that which offers abstract potential with computer modelling.⁷

That said, while digital models do typically exclude many of the cultural and political factors present in a site, this is rarely due to any inherent limitations of CAD software. If I put on my Agency of Mapping hat on, it seems that modelling software can establish a ‘field’ of cartographic conventions that encompasses many of the legal and geographic features mentioned in the above quote. I’m sceptical that digital models could simulate the subjectivities that drive these systems; but I think that their present states could be represented as well as with traditional methods.

There is often an all-to-brief leap between mapped information and designed form. A leap that designers take when they become so enamoured with their site research that it becomes an aesthetic driver for the actual design. This jump takes two forms. The first is through a direct translation, whereby information informs form, creating a 3-D ‘datascape’ that can then come to pass as a landscape, building, or city. As Julian writes:

In Reading MVRDV, Lootsma here is treating the representation as neutral and confusing visualisation, for existence. If the datascape is the visualisation of the effects of forces, then the world as it exists is a visualisation of the forces already. If the datascape makes these forces visible in a different way to the real, then it does so in a less specific way than the real does.⁸

But the second type of leap is much more interesting and promising. In this instance, the information gained during the mapping process is analytic: it has a logic that attempts to understand and simulate processes within a site rather than merely depict them as they currently appear. Here we often see algorithms employed as tools to help understand a site, such as using a recursive hill-descent operation to map the flows of water within a catchment. Unlike the datascape, these techniques could claim to create a useful understanding of how landscape forces operate rather than just visualise their effects.

But when it comes to actually designing something, these analytic processes can be inappropriately reapplied to the task of form generation. As Julian notes, landscape urbanists and computational architects often share a peculiar form of biomimicry, where they will abstract the model of a natural process away from its context — a re-purposing that aims to produce change and movement as an aesthetic effect or “operational metaphor”⁹ rather than as a functional state.¹⁰ Which can be fine. Circle pack, voronise, or delaunay your plans if you think it looks good. But recognise that your use of these algorithms is rhyme without reason. If you fetishise and reify the process of structuration in this way, you can’t then claim that the structure of your design emerges from anything except your aesthetic interests.

The opposite strategy is to only apply computational processes where they drive a functional aim. Where simulation and generation direct and evaluate how a design works, rather than how it appears. This performance-driven approach is much harder. It requires care and competence, especially given the greater scale, complexity, and dynamism of landscape systems compared to those of buildings. But it seems that more and more landscape urbanists are taking this approach. Most of Julian’s examples draw from the early work (circa 2003) of the AA’s landscape urbanism programme, while their more recent work — and that of the GSD and Penn — do make much better use of sophisticated and well-applied environmental models to inform their designs.

However, even when using simulations appropriately, the problem of how to produce form still rears its head. Once one has the means to evaluate how form affects function, what exactly do you do with that knowledge? All too often there is another gap in how well-applied simulations are then re-purposed to generate flawed forms. Looking again to the work of landscape urbanism programme, I wrote that:

The work of the AA similarly employs diagrammatic strategies to generate formal interventions. That said, their diagrams build upon a more rigorous analysis and understanding of the existing landscape that leads to an adaptive complexity whereby their designs tend to bend and grow according to a contextual application of an overarching logic. However, the architectural nature of these forms leaves these design solutions largely static and inflexible; an imposition that imposes a new equilibrium upon the landscape. This acts to neuter both the ability of the existing landscape systems to change, and the ability of the design itself to change in response to future problems. The design enacts a short-term reconfiguration, rather than a long-term remediation.¹¹

Similarly, Julian writes:

Landscape form that results from these mappings tends to then treat the abstract flows of mapped data as sculptural forms. When the geometries become concrete, the lack of understanding of landscape in landscape-architectural terms becomes apparent. This is demonstrated by difficulties with program or use, since propositions have no understanding of landscape types. Street, park, or garden elements are allocated to forms on the basis of their resemblance to the features of the geometry, such as patios allocated to areas, or walls to vertical graphs.¹²

This highlights that even though computational analysis can be a tool to enhance the design process, it is no replacement for a understanding of landscape phenomena drawn from disciplinary knowledge. Simulations can help us understand the problem domains in which we work, but the solutions to those problems are often conceptually flawed even when ‘validated’ through simulations. For example, one of Julian’s key criticisms is that the work of the AA ignores the slow and gradual trajectories of landscape systems.¹³ But this seems like a conceptual roadblock that can be divorced from their use of digital tools. Moreover, improving those tools to better take account for temporal processes would probably discourage a tabula rasa approach by helping designers to work with — rather than supersede — the systems present on-site.

Thus far I’ve been reflexively defending the present or future capabilities of landscape modeling. Changing tack, I’d like to look at the intent behind Julian’s criticisms. Namely, that each critique highlights a failing to embrace landscape’s potential for ‘real change.’ He examines how to actualise this potential through three case studies.

The first case study — the Bordeaux Botanical Garden by Mosbach Paysagistes¹⁴ — is a park that produces “novel material outcomes” by harnessing landscape systems, as exemplified in the soil mounds of the ‘Environment Gallery:’

The Environment Gallery both represents a landscape shaped by erosion as well as literally being shaped by erosion as part of the proposition. To erode is to destroy imperceptibly, little by little and, in geology, to gradually wear away. The immediacy of this expression and its geomorphological usage is equivalent because erosion works at scale. This assists the interpretive rhetoric of the project because the same processes that affect the mound do, in reality, affect the geology, the main difference being time frame. In considering the project’s manipulation of erosion, it’s important to remember that “the science of hydrology (water) is inseparably interwoven with geomorphology”, so any discussion about soil is a discussion about the action of water. Erosion results from the exposure of soil to water and is affected by a range of factors, including the inherent stability of the soil and its stabilisation by vegetation. Erosion can be caused by raindrops that dislodge particles, and also by overland flow that results from water moving across a soil, which is the main agent at these botanic gardens. The erodability of a soil will depend on its angle of repose or the angle at which its inherent structure can support itself. This structure results from a combination of sand, silt and clay, the friction between the relative particle sizes, and also the coherence that results from the cation-exchange capacity, which determines how the particles cohere in the presence of water, with clay particles cohering most. Overland flow occurs when water is directed across a surface at a velocity to break these bonds, and where the water causes erosion that moves a soil past its angle of repose. Overland flow is guided by topography and creates two types of landform: erosion landforms, where the erosion process shapes an existing mass; and deposition landforms, where a new mass is created from the eroded material deposited elsewhere¹⁵

I’ve introduced this particular case study because, as Julian notes, it is a landscape that exhibits designed change without the practices of post-construction maintenance and modification that characterise his other examples. Instead, the designer of the Botanic Gardens has successfully anticipated the possibilities for change during the design process. How did this happen?

As I have shown, the Bordeaux Botanic Garden uses the relationship of inorganic materials and the meteorological environment to create change and stimulate growth…

The fact that this is delivered through the construction documents prior to implementation makes the Botanical Gardens unique among the case studies, which are otherwise all ongoing and nonrepresentational. Since a key argument of the dissertation is that the emerging novelty from processes is best engaged with directly through gardening-like processes, the botanic garden is an important exception that demonstrates that change can be facilitated though conventional project-delivery mechanisms. The gardening-based projects can deal with emerging issues over time whereas the conventional project must predict them and use contingency to factor them in. Correspondingly, while the gardening project can be loose, the conventional project must be precise. This precision can restrict possibilities and extrude the future in a uniform way, while gardening does not just mitigate differences but works with and optimises them into exciting novelty. Nonetheless, the way that paths, edges and configuration have been detailed at the Botanic Garden still allows change while covering contingencies.¹⁶

Here the landscape is designed to evolve in a relatively open manner through a highly controlled focus on details and tectonics. This level of engagement is rare, and a credit to Mosbach’s imagination and execution. Yet, I wonder whether she would have benefited from using computational tools to better imagine and evaluate how to write the score for this “deliberately orchestrated entropy.”¹⁷ Developing a simulation for the erosion process may be a complex task and may only produce somewhat-accurate results. Even so, such a tool should allow the designer to better understand how some choices in materials and detailing could facilitate and direct the future outcomes of erosive processes. Perhaps Mobach’s knowledge of these factors was so intuitive that she had no need for such tests — but it would certainly help me if I needed to perform the same task. At the least, it might lessen the quantity of research and experimentation needed to understand how these processes operate.

Similarly, while introducing his second case study, Julian discusses how the modernist landscape architect James Rose expertly understood how different plants morphologies can be combined to create particular levels of transparency and enclosure.¹⁸ This level of engagement with the formal properties of plants contrasts with how planting plans are often designed in the present:

Planting plans effectively predict growth by the spacing they specify between plants, which includes the physical dimensions of growth in deciding how much space to allow. The planting plan uses predictions of mature size to determine appropriate spacings for plants when they are planted. The predictive model of growth used in the planting plan, which aims for a fixed future mature condition, denies the very thing that makes such an idea unique: growth. A landscape designer tends to see an outcome of growth that varies greatly from the original intention as a catastrophe¹⁹

While devising a planting plan will always be an exercise in best-guessing, it seems like another instance where simulations would be useful because the accuracy of our predictions can be improved by better taking natural variation into account. This is not to dismiss Julian’s suggestion that the careful hand of an expert gardener is needed to guide growth once the design is implemented; but rather that we should also enhance the designer’s ability to understand and evaluate the range of possibilities they set in play. Including an element of constrained randomness into growth predictions would create a multitude of probabilistic futures instead of relying on a single fixed average. Including temporal elements into these predictions would help designers imagine how the non-mature state of the landscape would look. Using more detailed methods of visualisation could help them to explicitly understand and test — as James Rose did — how morphology and growth affect spatial qualities. Taken as a whole, these improvements could enable a design process would actually begin to become properly morphogenetic in both the biological and computational sense. Which is why I’ve (crudely) worked on this problem in the past and am currently revisiting and improving my attempts.

This is all in service of saying that computation is one of the answers to the ultimate question of Julian’s thesis: “how can landscape architecture be practised to allow it to best manipulate its materials’ inherent capacity for change?”²⁰

His own answer emphasises two aspects — tendency and feedback — that are key to an improved practice. Tendency being “a way of thinking about the design process that recognises that design is a form-making process inherently tied to a prediction of an end, or later state, but where novelty is encouraged to develop over time”²¹ and feedback a “continuing, real-time involvement in a process … when the output of the process is fed back into another iteration of the process as an input.”²²

The kind of feedback Julian advocates is analogous to that of the gardener; a craft where there is “specificity, in material terms, at particular moments in time.”²³ Although simulations often recursively parse their results, the fact that they are virtual guesses rather than real results means there is an obvious gap because of their “inability to represent and simulate a process before the process has operated.”²⁴ Unless environmental sensors become widespread and embedded, there isn’t much opportunity for computational tools to directly engage with real-time feedback.

In contrast, the use of computation to examine ‘tendencies’ in the design process seems promising. Yes, simulations “can never predict the novelty that occurs in real time in the world”²⁵ and the “quantification of all variables in open systems are impossible.”²⁶ Despite this, they can be useful aids in designing complex landscape systems that change over time. While both computation and intuition are flawed instruments of prediction, the latter is an order of magnitude faster at generating representations and results that help us to evaluate the former. This speed is the very thing that makes the computational design process qualitatively different to traditional methods because it can dramatically reduce the effort needed to search, test, and retrace steps within the ‘design space’ that encompasses all possible routes to a final design.²⁷

Ironically, the ‘design space’ of architecture is flatter than that of landscape architecture because it demands a greater consideration of how the temporal dimension is affected by each design decision. It is in these interlocking complexities of form and time where we find a significant challenge that computation can alleviate. Whether we use the embodied practices that Julian advocates, or the analytic simulations that I’m investigating, attention must be paid to how we can cultivate our design process so that it enables us to cultivate changing landscapes. Which is to say that landscape architects should become better gardeners of both our virtual²⁸ and our real design environments.

Update: I’ve had some great responses from Julian, writing on his blog, and Rob Holmes, writing on Mammoth.