The proposition emerges from a space between two forms of knowing that have rarely been held together. Generations of watermen on the Chesapeake, shrimpers in coastal Louisiana, farmers along the Rhine have carried embodied, inherited knowledge of how their landscapes behave, intuitions about current, tide, sediment, season that are irreducible to data and untransmittable through instrumentation. Computational modelers, meanwhile, have built sensing and simulation capacities that extend far beyond what any human body can perceive, satellite-derived indices, real-time turbidity measurements, machine learning architectures that identify patterns across decades of data. But these two forms of knowledge have remained largely separate, the embodied knowledge cannot scale, and the computational knowledge cannot attend. Hutchins (1995) demonstrates that cognition in real-world settings is always distributed, across individuals, instruments, and environments, and that the unit of analysis for understanding knowledge production is never the individual mind but the sociotechnical system within which minds, tools, and environments interact. The landscape-as-model is the method that occupies the hybrid space between them, a framework in which distributed sensing extends embodied intuition to territorial scales, and in which embodied engagement with material systems grounds computation in the specificity of place. Neither pole is sufficient. The dissertation’s claim is that design practice, situated in that hybrid space, produces a form of knowledge that neither pole can generate alone.
Tim Ingold’s anthropology of skilled practice provides the theoretical ground for this claim. For Ingold, knowledge of the environment is not accumulated through detached observation but through what he calls “wayfaring,” the ongoing, attentive movement through a world that is itself in motion (Ingold 2011). The practitioner who knows a landscape knows it not as an object of analysis but as a field of relations discovered through engagement, the shrimper reading tidal patterns, the waterman anticipating sediment movement, or the farmer sensing soil moisture through the resistance of a plow. This is knowledge produced through “correspondence,” Ingold’s term for the responsive, ongoing attunement between practitioner and environment in which both are continually becoming (Ingold 2017). What the dissertation proposes is that computational sensing and responsive infrastructure can extend this correspondence to territorial scales without abandoning its essential character, the attentive, iterative, and constitutively entangled with the systems being known. The landscape-as-model is not a replacement for embodied knowledge but its scalar amplification and the instrumented territory as a field of correspondence in which designer, sensor, algorithm, and organism are all wayfaring together.
This chapter proposes the most radical claim in the dissertation, that the landscape itself can function as a model at 1:1 scale, a system observed, managed, and evolved through distributed sensing and responsive infrastructure, where each successive intervention produces new knowledge rather than imposing a final solution. It is a methodological proposition with specific requirements, distributed monitoring that captures what the landscape is doing at resolutions unavailable to human observation alone, responsive infrastructure capable of being adjusted as conditions change, and evolving interfaces through which the accumulating knowledge becomes available for design action. The move from laboratory surrogate to territorial prototype represents an epistemological shift, not only a technical one, from models that remain ontologically separate from the territories they represent to a practice in which the territory itself becomes the experimental apparatus.
Mediums: Water, Soil, Vegetation, Climate
To make this proposition credible, it is necessary to understand what the medium of landscape actually is. Water, soil, vegetation, and air are not abstract categories but the active materials through which future environments must be negotiated, and this medium carries its own agencies, resistances, and trajectories that complicate design intent. Water does not follow abstract rules, it erodes banks, deposits sediment in unanticipated locations, and reacts chemically with its context. Soil is not inert substrate waiting to receive seed or structure but a metabolic membrane alive with microbial life that cycle nutrients, sequester carbon, and condition what can grow. Vegetation produces microclimates, stabilizes slopes through root networks that hold substrate varying with species and season, and fundamentally reshapes hydrological regimes through transpiration and interception. To work with these materials is to enter into negotiation with agents that respond, adapt, and resist the controls imposed upon them (Bennett 2010).
Matter itself, as Karen Barad argues, is not a fixed, inert substance awaiting human manipulation but is agentive, produced through “intra-actions”, relationships in which neither humans nor materials pre-exist encounters, but co-constitute one another through practice (Barad 2007, 33). This means soil, water, and vegetation are not simply resources to be deployed for design intent but active participants whose behaviors emerge through their interactions with infrastructures, organisms, and material flows. A wetland does not function because it was designed correctly. It functions, or fails to function, because ongoing intra-actions among hydrology, sedimentation, microbial activity, plant succession, and management decisions produce conditions that either allow it to persist or push it toward collapse. Design, from this perspective, is about entering into sustained material-semiotic engagement with systems that are already underway, already evolving, already responding to forces that preceded design.
At its most totalizing, this methodology would engage trillions of agents, sensors, actuators, organisms, flows, working individually to produce collective knowledge, coordinated through computation paired with distributed sensing and recording infrastructures. But this is a maximum condition, not a threshold. The synthetic state already exists across a range of technological deployments, weather stations feeding irrigation controllers, SCADA systems managing reservoir releases, satellite-derived vegetation indices informing land management, citizen-science platforms aggregating species observations. What is missing is not the technology but the framework, a method that clearly articulates the agency of these systems as actors in the production of landscape knowledge, rather than treating them as passive instruments in service of human decision-making. The medium of landscape, engaged simultaneously at multiple scales, is already a site of continuous interaction, a relationship between articulated infrastructures that alter dynamic systems within ranges rather than through predetermined outcomes. The protocols that govern these systems must continually evolve to catalyze and resist environmental processes as conditions change (Gabrys 2016). Three components are necessary to reframe the landscape itself as the model, distributed monitoring, responsive infrastructure, and evolving interfaces.
Landscapes as Cybernetic Systems
The concept of feedback, the basis for adaptive management, originates in cybernetics, the study of “control and communication in … animal(s) and the machine(s)” (Wiener 1948). Norbert Wiener developed cybernetics during World War II to solve problems of anti-aircraft fire control, where systems must adjust their behavior based on continuous information about a target’s position and velocity. The insight that emerged is that control in complex, dynamic environments requires feedback loops that continuously compare actual performance against desired outcomes and adjust actions accordingly (Wiener 1948, 6–8).
Landscapes display cybernetic tendencies whether we design them or not. Consider how a tidal marsh maintains its elevation relative to sea level. When elevation falls, inundation increases. When inundation increases, sediment delivery increases. When sediment delivery increases, elevation rises, until equilibrium is restored, assuming sediment supply is adequate (Kirwan and Megonigal 2013). This self-regulating behavior emerges from physical and biological processes interacting through information flows, the marsh “senses” its elevation relative to water level through inundation frequency, and vegetation responds by trapping more or less sediment accordingly (Fagherazzi et al. 2012). The feedback loop exists because the physical configuration of the marsh produces it.
What computational management adds to these natural feedback loops is the capacity to process information at scales and resolutions that exceed biological agents’ capabilities, and to actuate interventions that redirect flows at territorial scales (Gabrys 2016, 14). The computational infrastructure does not replace natural feedbacks but supplements them, creating hybrid cybernetic systems where machine sensing and algorithmic control work alongside ecological self-regulation (Patten and Odum 1981).
Claude Shannon’s information theory, developed concurrently with Wiener’s cybernetics, provides a mathematical foundation for understanding communication in these systems (Shannon 1948). Shannon demonstrated that information is distinct from meaning, a sensor transmitting data about water levels is communicating information regardless of whether that information is meaningful to any observer. What matters is the reduction of uncertainty. Before measurement, many water levels were possible. After measurement, uncertainty collapses to a specific value within the accuracy of the sensor (Shannon 1948, 379–83). The value of distributed sensing networks lies in their capacity to reduce uncertainty across many variables simultaneously, providing the information flows necessary for adaptive feedback to operate across territorial scales (Hart and Martinez 2006; Rundel et al. 2009).
The Limits of Surrogate Models
Physical Models as Bounded Experiments
Physical hydraulic models, concrete basins with adjustable slopes, controllable flows, and measured sediment feeds have been central to river and coastal engineering since the mid-twentieth century. These models operated as controlled experiments where variables can be isolated, flows repeated, and outcomes measured with a precision impossible in wild rivers.
Yet physical models are constrained by fundamental scaling problems that cannot be fully resolved. Hydraulic similarity requires matching dimensionless ratios, the Froude number for gravity-driven flows, Reynolds number for viscous effects, Weber number for surface tension, but satisfying all simultaneously is typically impossible (Heller 2011). A model scaled 1:100 in length might achieve Froude similarity, meaning gravitational forces scale correctly, but fail Reynolds similarity, meaning viscous forces are disproportionately large. The consequences cascade through the system and sediment behavior deviates from field conditions, turbulence fails to reproduce, and boundary layer dynamics drift from how the river actually performs.
This problem deepens when we consider what must be abstracted or omitted. Sediment grains cannot be scaled down proportionally without changing their physical properties and fine sands become clays in the model, altering transport thresholds and deposition patterns (Le Méhauté 1976). Vegetation, biological crusts, and microbial films are critical to real-world sediment cohesion and nutrient cycling but are typically omitted, rendering the model as a hydraulic skeleton that moves water convincingly but omits what makes a river or marsh function as a living system.
Physical models also freeze time in a particular ways. They can simulate hours or days of flow, but not decades of succession, land-use change, or policy evolution. They do not capture the slow drift of a watershed’s sediment budget as upstream dams trap sand, or the gradual encroachment of invasive species that alter roughness and flow patterns over years and decades. Physical models excel at answering narrowly defined questions but they struggle with open-ended, adaptive questions about how should a system be managed over fifty years as climate, land use, and ecological communities indeterminately shift?
Infinite but Approximate
Computational models promised to overcome many of these limitations. By the 1970s, numerical solutions to the St. Venant equations enabled one-dimensional river flow simulations. By the 1990s, two-dimensional shallow water models such as HEC-RAS and LISFLOOD-FP allowed spatially explicit flood inundation mapping. By the 2000s, fully three-dimensional computational fluid dynamics and landscape evolution models began simulating sediment transport, channel migration, and delta growth over millennia (Tucker and Hancock 2010; Coulthard et al. 2002). Unlike physical models, computational models are not constrained by laboratory space or by the need to match dimensionless numbers. They can run forward in time indefinitely, testing multiple scenarios in parallel, and they can incorporate additional processes of vegetation growth, land-use change, and sea-level rise as modular components.
Yet computational models introduce their own epistemic challenges that are less visible precisely because they are less tangible. Numerical models of open natural systems can never be fully validated because they rest on simplified representations of processes that are incompletely understood, and because the initial conditions and boundary conditions required to run them are never known with sufficient precision (Oreskes, Shrader-Frechette, and Belitz 1994). Models are not mirrors of reality but highly selective tools that embody specific and pragmatic assumptions about what matters. A hydrodynamic model may represent water as a continuum governed by Navier-Stokes equations, ignoring the fact that rivers are textured by boulders, logs, root wads, and ice, which produce turbulence and drag at scales the model does not resolve. A landscape evolution model may simulate sediment flux using a simplified transport law calibrated to one river, then apply it to another with different lithologies, vegetation, and hydrology, yielding plausible outputs that are nonetheless wrong in important ways (Willems 2008; Beven 2006).
Computational models are technical instruments and infrastructural assemblages, hybrids of code, data collection networks, institutional commitments, and political negotiations. Climate models did not emerge from pure science but were co-produced with Cold War military investments, international treaty negotiations, and the gradual construction of global sensor networks. The models a “vast machines” that organize how climate is known, debated, and governed (Edwards 2010, 366–408). What gets modeled, at what resolution, using which calibration datasets, and for whose benefit are never purely technical decisions but are shaped by funding priorities, regulatory mandates, and institutional capacities. A coastal model built to justify levee construction will be structured to privilege flood stage reduction, protecting property values, while a model built to guide wetland restoration, might emphasize habitat connectivity and sediment budgets. Models do not simply represent landscapes, they actively construct the terms on which landscapes are understood, valued, and managed (Edwards 2010, 84–120).
The trade-off is unavoidable and it is often necessary to choose between generality, realism, and precision (Levins 1966). A qualitative ecosystem model can be general and realistic but imprecise. Models of site-specific processes can be realistic and precise but not general. Or like mathematical abstractions a model can be general and precise but unrealistic. This is not a failure of modeling but a structural feature of how abstraction works. The implication for environmental management is that no single model type can represent the embodied world. Models validate and produce knowledge that helps clarify assumptions, explore consequences, and generate hypotheses, but they do not replace observation, experiments, and adjustments in the field (Starfield 1997).
Every model deployed within a Promethean frame forecloses the questions it was not designed to ask. Levees built to a model’s flood stage become political commitments that resist revision when the model proves inadequate. Sensing networks calibrated to a specific set of variables become institutional fixtures whose parameters resist change, because altering them would invalidate the baseline against which years of management decisions were justified. The computational layer does not just produce outcomes. It produces outcomes that are progressively harder to undo, because each layer of analysis adds confidence to an epistemological commitment that was never examined. The landscape-as-model, the proposition this chapter develops, is not an extension of this trajectory. It is a deliberate reversal, designing the model to be surprised rather than to confirm.
From Surrogate to Prototype
Landscape as Epistemic Thing
If we agree that physical models and computational simulations cannot fully capture the behavior of complex, evolving landscapes, what is the alternative? The answer is not to abandon models wholesale but to add to the sites of experimentation from the surrogate (the model) to the prototype, the landscape itself, instrumented, monitored, and adaptively managed for knowledge production. Chapter 06 traced de Monchaux’s observation that models have historically moved inward, away from the landscapes they shape, becoming more closed to view as they become more powerful (de Monchaux 2025). The geomorphology tables reversed that movement by returning the model to material engagement. The landscape-as-model completes the reversal. The model does not merely return to the territory. It becomes the territory.
This reframing treats environmental interventions as experiments from which knowledge is generated (Holling 1978; Walters and Holling 1990). Management actions are designed to test hypotheses about system behavior. A sediment diversion is not only a flood mitigation measure but a structured experiment in land-building, monitored to reveal how much sediment is captured, where it deposits, what vegetation colonizes, and how the configuration might be adjusted as understanding improves.
The systems that constitute landscapes and the social-ecological systems in which they are embedded do not exist in stable equilibrium but cycle through phases of growth, conservation, release, and reorganization. Management interventions occur within these adaptive cycles, reinforcing existing regimes, triggering transitions to alternate states. Disturbances at one scale can cascade up or down, and interventions intended to stabilize another component can ossify the system, making it brittle and vulnerable to larger disruptions (Gunderson and Holling 2002, 33–75). This complicates the landscape-as-model approach by insisting that experimentation must attend to temporal dynamics across scales, from daily tidal cycles to decadal vegetation succession, to century scale geomorphic change and must acknowledge that interventions at one scale may inadvertently produce vulnerabilities at another.
Consider a responsive sediment diversion that effectively builds land during moderate river flows. That same diversion might reduce the frequency of small disturbances that previously maintained vegetation diversity and soil oxygenation. Over decades, this suppression of variability leads to a homogeneous, less resilient marsh that is catastrophically vulnerable when an extreme storm overtops the system. Adaptive management must not simply optimize for a single goal like maximum land-building but maintain variability, redundancy, and modularity so that the system retains the capacity to absorb the unpredictable and reorganize after disturbance (Folke et al. 2010). The landscape-as-model is not only a technical apparatus for monitoring and adjustment but also a site for cultivating adaptive capacity, with an inherent ability for both ecosystems and institutions to learn, innovate, and reorganize in response to change.
This approach does not reject abstraction. It integrates physical models, computational simulations, and field-scale prototypes into an ecology of methods. Physical models remain useful for testing specific hydraulic configurations under controlled conditions. Computational models remain essential for exploring parameter spaces, projecting long-term trajectories, and synthesizing data across scales. But the field-scale prototype provides the ground truth against which surrogates are calibrated, and it generates forms of knowledge about ecological response, social acceptability, institutional friction, and emergent behaviors that surrogates cannot anticipate (Williams 2011; Allen et al. 2011).
Hans-Jörg Rheinberger’s concept of “epistemic things” illuminates what is at stake in this reframing. Experimental systems in molecular biology function as epistemic things, arrangements of instruments, materials, and protocols that are not designed to answer pre-formed questions but to generate new questions through their operation (Rheinberger 1997). A protein synthesis experiment does not confirm a hypothesis. It produces unexpected traces, anomalies, and patterns that redirect inquiry in directions the experimenter could not have anticipated. Similarly, an instrumented wetland, a managed sediment diversion, or a monitored tidal flat functions as an epistemic thing, a dynamic system that teaches through its performance, its failures, and its surprises. The geomorphology modeling research conducted in my own practice demonstrates what this looks like in design.
The geomorphology table at REAL (2014–17) and the UVA Geomorphology Lab (2017–present), a physical-digital hybrid environment of synthetic sediment, programmable water flow, overhead depth cameras, and ultrasonic range finders on motorized rails, was deliberately not a scaled replica of any coastal landscape. Froude scaling and Reynolds number correspondence, the mathematical conditions establishing similitude between model and prototype in engineering hydraulic modeling, were abandoned. The table was its own environment, designed to generate principles through material encounter rather than to predict outcomes in specific sites. It computes simultaneously in two registers. The territory itself solves hydraulic problems, water finding paths, sediment sorting by grain size, channels forming and migrating. The sensors make that computation legible, the material computing what the digital layer makes readable.
When Leif Estrada’s thesis research at REAL tested actuated sediment insertions within the table’s flow field, the discovery was not a validated technique for replicating outcomes in the Los Angeles River. It was a principle, that redirecting sediment from a demolished concrete channel could be achieved through choreographed operations that guide outcomes within controlled ranges, with aggregated effects responsive to current hydrology while projecting future goals. The model did not confirm a hypothesis Estrada held before entering the laboratory. It generated a question the laboratory’s material dynamics made possible, how might the interaction between actuated gates and flow conditions produce channel morphologies without specifying them in advance? Embedding GIS data directly into the design model extended this further, turning the model from a static picture into a structured database that can be queried and simulated. The territory becomes an instrument of inquiry rather than a record of fact, and the landscape’s response to construction refines the hypotheses the database encodes. This is Rheinberger’s epistemic thing functioning exactly as described, producing surprises that redirect inquiry.
Andrew Boyd and Tyler Mohr’s “FIN” research at the same table extended this further, a formal language of flow enhancers, resistors, and redirectors, documented through iterative image analysis that produced topographic plans expressing landscape as a gradient between stasis and flux rather than a fixed form to be achieved. This notation system, the epistemic artifact that the table’s operation made possible, is the kind of knowledge that surrogates cannot generate, not a prediction of what the landscape will be, but a grammar for how it can behave. The prototype, in Rheinberger’s terms, worked.
This reframing has profound implications for design practice. Designers are no longer solely authors of forms but orchestrators of experimental infrastructures that generate knowledge over time. The landscape is not a passive medium awaiting inscription but an active partner in an open-ended process of inquiry (Schön 1983; Pickering 1995). Prototypes, in this sense, are not preliminary versions of final solutions but living research instruments. Computational tools should be understood less as mechanisms for optimization but as platforms for speculation and iterative refinement, enabling designers to “test multiple futures” in situ and adjust as new knowledge emerges, as Holzman (2018, 112) elaborates in relation to the methods developed through the REAL and UVA Geomorphology Lab research programs described in Chapter 3. This argument extends naturally to the landscape itself and if computational models are speculative instruments, then instrumented landscapes are speculative environments, producing data and experiences that feed back into design decisions and the adaptation of the systems themselves.
Landscapes and their human designers do not exist as separate entities that then interact but are mutually constituted in an ongoing material-discursive practice (Barad 2007, 140). An ecological restoration project is not a human intervention acting on a passive landscape but an entanglement in which plants, sediments, flows, sensors, regulations, funding mechanisms, and human labor co-produce one another. The “success” or “failure” of the restoration cannot be attributed solely to design decisions or to ecological processes but emerges from the specific configuration of relations that the project enacts. This has methodological consequences in that adaptive management is not a matter of humans learning about landscapes through observation and adjustment, but of humans and landscapes co-learning and mutually transforming within a shared practice.
Landscape as Model
To operationalize this methodology, to move from the laboratory back to the field and treat the landscape itself as the model, three technical and conceptual components are necessary, distributed monitoring, responsive infrastructures, and evolving interfaces. Each shifts the locus of knowledge production from abstract representation toward an embedded, iterative practice.
Distributed Monitoring: Sensing Across Scales
The first component is an approach to monitoring that is distributed and functions across multiple scales, from remote sensing platforms that capture territorial (watershed to planet) patterns to in-situ chemical sensors embedded in soils and water bodies that register molecular transformations. Rather than developing an “…all-seeing eye,” the network should function to express the plurality and heterogeneity of environmental conditions (Gabrys 2016). “If the satellite view has been narrated as a project of making a global observation system and seeing earth as a whole object, then more distributed monitoring performed by environmental sensors points to the ways in which the earth might be rendered not as one world, but as many” (Gabrys 2016, 14). This shift from a singular, totalizing vision to distributed sensing acknowledges that landscapes are not uniform surfaces but assemblages of processes operating at different temporal and spatial scales simultaneously.
Distributed sensing networks recognize that no single vantage point suffices to understand dynamic environmental systems. A coastal marsh might be observed through satellite-derived NDVI tracking vegetation health over kilometers, aerial LiDAR capturing millimeter-scale topographic changes across hectares, fixed tide gauges recording water level fluctuations at hourly intervals, autonomous surface vehicles collecting turbidity and salinity profiles along transects, and soil moisture sensors embedded at depths throughout the marsh platform itself, transmitting data every fifteen minutes. Each sensor type captures a different dimension of the system, spectral reflectance, elevation, water stage, suspended solids, pore water chemistry, and together they produce a synthetic, multi-scalar framing that cannot be reduced to a single dataset (Passalacqua et al. 2015; Hart and Martinez 2006; Rundel et al. 2009).
Sensor networks are not solely technical devices but are also political and infrastructural. The construction of global climate data networks required international cooperation, standardization of measurement protocols, negotiations over data access and ownership, and sustained institutional commitments to maintain stations over decades (Edwards 2010, 84–120). Similarly, distributed landscape monitoring networks are embedded in institutional ecologies, who deploys sensors, who has access to data streams, which variables are measured and at what resolution, and how data is archived and shared all reflect funding structures, regulatory mandates, and power relations. A sensor network designed by a state environmental agency to track compliance with water quality standards will be configured differently, measuring different parameters, at different locations, and reporting to different audiences, which is different than a network designed by a fishing community monitoring salinity intrusion to protect economic vitality. The landscape rendered by sensors is not reality but a landscape constituted through specific infrastructural and institutional arrangements.
Distributed sensing is more than data collection, it is the infrastructural precondition for adaptive management. Without continuous feedback adaptation is impossible. Sensors become the sensory organs of knowledge production, enabling real-time adjustments during storm events and long-term recalibrations as baseline conditions shift (Porter et al. 2005). They make the landscape legible as a dynamic field of ongoing processes that can be read, interpreted, and responded to.
The application of sensing within the environment identifies which processes count and are acted upon, defining the terms on which a landscape is known and valued (Borgman 2015). A landscape monitored primarily for flood risk will be managed to minimize peak discharges and protect property, likely at the expense of ecological variability. A landscape monitored primarily for habitat quality will prioritize species richness and ecosystem function, potentially tolerating higher flood risk. A landscape monitored for carbon sequestration will emphasize vegetation productivity and soil organic matter accumulation. The design of sensing regimes is a political and epistemological act, it establishes cultural values for knowledge production, whose questions are addressed, and the values embedded in management decisions (Tironi 2018).
The Thresholds installation (2006) at LSU demonstrates the political dimension of sensing at its most elemental. I painted a high-contrast mural on the atrium wall because the camera needed a fixed reference. By choosing its geometry, tonal range, and spatial extent, I decided what the territory would be allowed to reveal. Without the mural as calibrated reference, the blob detection algorithm would have produced noise, contrast gradients with no stable context against which to measure difference. With it, a specific kind of landscape became legible, one in which human occupation is topographic event, in which light is terrain, in which the contour line is revealed as a reading of difference between instrument and context rather than a property of the land. Changing the datum makes a different landscape emerge. A finer datum proliferates isolines into noise, a coarser datum consolidates them into generality. A different painting, a different calibration, a different threshold for what counts as contrast, would have produced a different landscape entirely. The choice of datum is always a choice about whose experience of the landscape gets registered, what phenomena are treated as information, and which landscape becomes available for design action.
Responsive Infrastructure: Small-Scale, Iterative Interventions
The second component consists of new forms of infrastructure that capitalize on responsive technologies, utilizing small-scale interventions that engage the environment through a range of physical, chemical, and biological interactions. These infrastructures provide the modifications that, when aggregated over time and space, produce territorial scale environmental changes. Rather than relying on monolithic levees designed calibrated to a single return interval or fixed channels excavated with an optimized cross-section, responsive infrastructures must be modular, adjustable, and even reversible. They are designed not to impose a fixed solution but to learn and respond through operation (Easterling 2014).
The range of such infrastructures is broad and growing. Adjustable weirs whose crest elevations can be raised or lowered seasonally in response to measured flow conditions and ecological needs. Sediment diversion structures with variable gate openings that modulate the volume and timing of sediment delivery to wetlands based on real-time turbidity data and land-building targets. Floating breakwaters that can be repositioned as shorelines migrate or as storm patterns shift. Biogeochemical treatment systems whose media and flow paths can be reconfigured in response to changing pollutant loads. These are active systems that sense conditions and adjust their behavior or are adjusted by managers informed by continuous streams of sensor data and model projections, as Holzman (2018) describes in relation to the responsive infrastructure methodology developed through the research programs documented in Chapter 3.
The Sedimachine prototype at LSU (2012) operated at the smallest useful scale of this logic. A robotic spillway with twelve servo-actuated gates, each controlled through an Arduino microcontroller, enabled programmable sequences of opening and closing that choreographed water delivery across a plexiglass surface. The gates could be scripted, repeated, and modified based on observed outcomes. A sequence that produced unsatisfactory deposition could be adjusted. The failure was recoverable. The gates shaped flow conditions, not depositional forms. The sand performed its own score according to its material intelligence, and the actual deposition patterns became the data that informed the next intervention. This programmable hydrology established, at experimental scale, the principle that responsive infrastructure demands, that operations, not outcomes, are the design lever.
Thirteen years later, the NEOM consultation (2022–25) applied the same logic at territorial scale. When GeoHECRAS hydrological modeling revealed that conventional channelization of the wadis surrounding The Line would require infrastructure widths exceeding 200 meters with hardened concrete at velocities incompatible with ecological function, the responsive alternative was not a design preference but an engineering necessity, reconceiving the traditional ephemeral wadi systems as holding areas, slowing water to sustain life and safely directing it into cisterns and caverns, while selectively ossifying some terrain to concentrate erosion where it builds productive ravines, and managing a fluctuating coastal isohaline zone through controlled water inputs from The Line’s manifold. The infrastructure was designed to be adjusted as the landscape it initiates develops, not a fixed solution but a set of initial conditions from which the landscape would construct its own subsequent configurations. In the northern landscapes, where water is routed to erode and carve ravines sheltering life from harsh climate, the topographic variation that provides shade and wind protection for biological communities is the same topographic variation that produces the visual complexity and spatial richness that make the landscape habitable for human experience. The aesthetic and the infrastructural are not separable because they emerge from the same hydrological and geological processes operating on the same terrain.
The distance between the Sedimachine’s twelve Arduino gates and NEOM’s internet of ecologies is not conceptual. It is scalar. The same epistemological commitment holds at both ends of the spectrum, that form is the outcome of designed operations rather than a predetermined end state, that failure is recoverable and generates knowledge, that the infrastructure is designed to be modified as the landscape responds. Water moves with its own intelligence, finding paths of least resistance and organizing sediment according to velocity and particle size. The divergence between engineered gate sequences and the water’s emergent patterns is where the knowledge lives.
The logic here is iterative rather than deterministic, grounded in a recognition that environmental systems are indeterminant and cannot be fully predicted in advance. Management should design for flexibility and avoid configurations that cannot accommodate the unknown (Gunderson and Holling 2002). Rather than optimal configurations produced through modeling and monolithic constructions, responsive infrastructures are deployed in phases, monitored for performance against multiple criteria, and reconfigured based on what is learned. A sediment diversion might begin as a pilot opening of modest size, operated for one or two seasons while vegetation response, sediment deposition patterns, downstream water quality impacts, and navigational effects are measured (Paola et al. 2011; Nittrouer and Viparelli 2014). If land-building rates meet targets and unintended consequences remain within acceptable bounds, the opening might be widened. If sediment deposits in unproductive locations or if salinity intrusion proves more problematic than anticipated, the structure is adjusted, the gates reconfigured, operating schedules revised, or complementary interventions added or subtracted. This approach accepts that prediction is partial and that a reliable way to learn how a complex system will respond is to intervene at a scale where failure is recoverable, to observe carefully, and adjust accordingly (Walters and Holling 1990; Westgate, Likens, and Lindenmayer 2013).
Adaptive management must cultivate an adaptive capacity, the ability of socio-ecological systems to reorganize after disturbance without losing essential functions (Folke et al. 2010). By deploying multiple small interventions rather than a single large one, by preserving modularity so that components can be reconfigured or replaced, and by avoiding irreversible commitments that foreclose future options, responsive infrastructures keep systems poised to respond to the unknown. A network of small, adjustable infrastructures distributed across a delta is more resilient than a single massive diversion because the network can accommodate local failures, can be tuned to variable ecological conditions across space, and can be incrementally expanded or contracted as knowledge increases.
Responsive infrastructures operate at scales between traditional design gestures and long-term management routines. They are not micro-manipulations of individual plants or ephemeral flow paths, nor are they territorial mega-projects whose failure is catastrophic. They occupy a productive middle ground, infrastructures substantial enough to alter system behavior in measurable ways but small and modular enough to be adjusted, multiplied, relocated, or removed without triggering cascading failures. The modularity is critical. If an intervention does not perform as expected the failure is localized and recoverable, and the knowledge gained from that failure directly informs the design of the next iteration (Allen et al. 2011; Braun 2014). Failure, in this framework, is not something to be avoided, it is a form of productive feedback that increases knowledge and articulates future interventions.
Responsive infrastructures do not simply act upon landscapes, they intra-act with them, meaning that the infrastructure and the landscape co-produce one another through their mutual engagement (Barad 2007, 33). A sediment diversion does not deliver sediment to a passive marsh. Rather, the diversion, the sediment, the flowing water, the vegetation that colonizes newly deposited soils, and the monitoring sensors that track changes together constitute an assemblage that produces marsh-building as an emergent phenomenon. The diversion’s performance is not determined solely by its hydraulic design but by the ongoing relations among all these components. This perspective suggests that responsive infrastructures should be designed with an awareness that their behavior will be shaped by, and will shape, the landscapes they engage, and that learning must attend to these relational dynamics rather than treating infrastructure as an independent variable acting on a dependent landscape.
Evolving Interfaces
The third component is a commitment to evolving interfaces, the systems for visualizing, managing, and narrating the ongoing evolution of environments alongside the lives of humans and non-humans. The interface occupies a central position in a neo-projective framework of landscape management, a framework in which landscapes are not delivered as solutions but as unfolding processes that must be continuously interpreted, communicated, and negotiated (Corner 1999; Meyer 2008). The interface is a performative, cultural project, a space of engagement for technical specialists, policymakers, and broader publics, used not only to convey information but to educate, provoke debate, and sometimes entertain. Interfaces are a design project that mediates between raw sensor data and actionable knowledge, between the specialized discourses of hydrology and ecology and the vernacular concerns of residents and stakeholders, and between present conditions and bifurcating futures (Lister 2007).
An interface might take many forms depending on audience and purpose. A real-time dashboard displaying water levels, sediment fluxes, vegetation indices, and nutrient concentrations for a managed delta or coastal wetland, updated continuously and accessible through web browsers to managers, scientists, regulatory agencies, and the public. A parametric design environment that allows engineers and landscape architects to test configurations of gates, channels, or planting strategies and render predicted hydrological, ecological, and economic outcomes across scenarios (Holzman 2018). A public-facing platform that visualizes decades of landscape transformation, annotated with narratives explaining why interventions were undertaken, what was learned from their successes and failures, and how management strategies have evolved in response. An augmented reality application that overlays historical flood extents, projected sea-level rise zones, or planned infrastructure onto present streetscapes and marsh platforms, making processes that unfold over decades or centuries momentarily visible to residents navigating their daily lives (Spirn 1984). Importantly, the interface extends beyond these examples and should be understood as a broad mandate to communicate between species and phenomena to evolve relationships.
Data infrastructures, the systems that collect, store, process, and visualize environmental information, are not neutral conduits as they actively shape scientific and political discourse. Climate models became credible not simply because they are mathematically rigorous but because they are embedded in vast data infrastructures that made their outputs visible, comparable, and actionable across institutions (Edwards 2010, 198–237). Similarly, landscape interfaces are not displays of pre-existing facts but are instruments that construct the terms on which landscapes are debated and managed. An interface that visualizes flood risk but omits habitat quality or carbon storage implicitly prioritizes certain values over others. An interface that shows only average conditions but not variability obscures the role of disturbance in maintaining ecosystem resilience. The design of interfaces is a form of argument and a means of cultural production, it proposes what matters, what should be monitored, and what futures should be pursued.
A thoughtfully designed interface can render legible the slow, complex processes that otherwise remain abstract or invisible, the sediment accreting millimeter by millimeter over generations, marsh platforms rising imperceptibly in response to sea-level rise, nutrient concentrations declining as treatment wetlands mature and expand their filtering capacity. By making these incremental changes visible and comprehensible, interfaces create publics, communities of interest, concern, and advocacy that engage with, contest, or champion landscape futures (Latour 1987). They have the capacity to transform passive inhabitants into active participants in an ongoing experiment, agents who can see the results of management decisions unfolding in real time and who can contribute observations, ask questions, and demand adjustments when outcomes deviate from expectations or values. Importantly, this concept traverses species and phenomena asking design to build pathways of understanding that may be physical or digital, a connective fabric that connects discursive means of relating.
Interfaces should visualize cross-scale dynamics and adaptive capacity, not just current states. An interface that displays only present conditions provides limited insight into how the system is emerging. A more robust interface communicates trajectories over multiple timescales (daily tidal variation, seasonal patterns, decadal trajectories), shows thresholds where state shifts might occur, and indicates the degree of redundancy and modularity in management options (Gunderson and Holling 2002). Such an interface would support not just operational decision-making but strategic reflection that questions how a management regime maintains the system’s capacity to reorganize after disturbance.
Interfaces evolve as new data streams come online and incorporate them. As new questions emerge from managers, scientists, non-humans, or publics, the communication should be reconfigured to address those questions. As management priorities shift in response to policy changes, extreme events, or accumulated knowledge, the interface should reflect those shifts, making transparent the reasoning behind decisions and the trade-offs being navigated (Tironi 2018). This ongoing evolution is a trait of initial design intent, a recognition that knowledge production is continuous, that landscapes evolve, that publics and their concerns change, and that the tools for understanding and managing landscapes must change alongside them. An interface obsessed with a single set of metrics or communications becomes an obstacle.
Together, these three components, distributed sensing, responsive infrastructure, and evolving interfaces constitute the technical and conceptual apparatus necessary for treating landscape as model. They enable experimentation at 1:1, learning through iterative practice, and adaptation across temporal scales. They shift the relationship between designer and landscape from one of authorship, in which a fixed form is imposed and expected to endure, to one of stewardship, in which an open-ended process is orchestrated, monitored, and continually adjusted in response to what the landscape itself reveals.
Landscape as the Model
Occupying the Experiment
With landscape repositioned as the model, the research agenda shifts. Rather than pursuing ever more accurate and computationally cumbersome surrogates for the landscapes we occupy the project becomes one of occupying the model itself by bringing experiments back into the field at full scale (Felson and Pickett 2005). This does not constitute a rejection of abstraction or a naive embrace of learning by doing without theoretical grounding. Physical models and computational simulations remain essential tools for exploring possibilities, testing configurations in artificial sediments or silicon before committing resources, and projecting long-term trajectories in multiple scenarios. But they are no longer the primary or sole sites of knowledge production. The prototype, instrumented with distributed sensors, equipped with responsive infrastructures, and made legible through evolving interfaces becomes the epistemic center, the place where hypotheses are grounded and where consequential knowledge is produced (Nassauer and Opdam 2008).
This repositioning carries significant methodological implications. The act of design generates hypotheses in spatial and material form, hypotheses that are tested not through discourse alone but through construction, monitoring, and evaluation (Lenzholzer, Duchhart, and Koh 2013). When the hypothesis is not merely “What if this levee were designed two meters higher?” but “How will this coastal landscape respond to a sediment management regime over a decade, and how should interventions be sequenced and adjusted to produce new knowledge?” then the primary research instrument is inhabited (Koskinen et al. 2011). The landscape becomes an experimental system to generate surprises that redirect inquiry, and to accumulate knowledge as a method to address indeterminacy (Rheinberger 1997).
The credibility of climate science rested not only on the accuracy of models but on the construction of vast infrastructures that made models testable and their outputs actionable (Edwards 2010, 366–408). Similarly, the landscape-as-model approach requires constructing infrastructures of credibility to create institutional arrangements that promote the value of knowledge production. The landscape prototype is not solely a physical site but a socio-technical assemblage. Its credibility depends on the robustness of this assemblage, the values it expresses, not solely on the ecological or hydrological “correctness” of a single intervention.
This approach also transforms how landscapes are inhabited and narrated. If the landscape functions as an ongoing experiment rather than a finished artifact, then those who live within it are not passive subjects or recipients of expert decisions but participants in an extended process of knowledge production. Interfaces that communicate change over time, that explain the rationale behind interventions, that present scenarios and trade-offs, and that solicit feedback from human and non-human species transform publics into collaborators. The landscape is no longer a black box managed by distant specialists but a shared epistemic terrain, a place where different forms of expertise meet and inform one another (Spirn 1984; Latour 1987). Inhabitants contribute what sensors miss, the timing of bird migrations, the smell of algal blooms, the locations where paths flood during particular tidal and weather combinations. These communications, when integrated into monitoring networks and fed back into management decisions, enrich the knowledge base and strengthen the socio-ecological legitimacy of adaptive interventions.
Adaptive capacity depends not only on ecological flexibility but also on social learning and institutional innovation. A landscape-as-model approach must cultivate socio-ecological resilience, an ability for both ecosystems and human communities to reorganize after disturbance, to digest feedback, and to innovate in response (Gunderson and Holling 2002, 398–412). This means designing not just for ecological robustness but for institutional adaptability, creating management agencies that can revise regulations in light of new knowledge, funding mechanisms that support long-term monitoring rather than short-term projects, and governance structures that enable diverse stakeholders to participate in decision-making. A sediment diversion might function ecologically but fail socially if the institutions managing it cannot adapt to unexpected outcomes, cannot incorporate local knowledge, or cannot negotiate conflicts.
This approach demands epistemic humility. If the landscape is the model and the model is an experiment, then surprises are inevitable and should be anticipated as normal rather than aberrant. The landscape will teach things that were not anticipated, about material behaviors, about ecological trajectories, about social responses and those teachings will require adjustments not only to physical infrastructures but to the conceptual frameworks, management protocols, and socio-cultural relationships (Pickering 1995).
This is not a failure of method but must be culturally accepted as the normal operation of knowledge production. Andrew Pickering argues that scientific work is fundamentally a back-and-forth negotiation between human intentions and material resistances, a process in which both the investigator’s plans and the phenomena under investigation are transformed through interaction (Pickering 1995, 21–67). Landscape-as-model embraces that negotiation. It accepts that landscapes possess agency, that they will push back against interventions in ways both predictable and surprising, and that the most productive response is not to force compliance but to listen carefully to what is elucidated, to adjust infrastructures and protocols accordingly, and to promote the conversation. Adaptive management, in this framework, is not a concession to uncertainty but a recognition that complex systems communicate clearly when they respond on their own terms.
The Pocomoke Sound’s coastal islands and submerging marshes are not case studies that illustrate predetermined design principles. They are testing grounds that challenge design principles, generating the friction between what the design frameworks propose and what the landscape actually does that is the source of the knowledge the studio produces. When a student prototype for stabilizing a submerging marsh edge encounters the Sound’s actual tidal dynamics, sediment budget, and biological community, what emerges is not a validation of the prototype’s design logic but a renegotiation of it. The site is the instrument of inquiry as much as the design is. This is what model-as-site means at territorial scale. The Pocomoke Sound is used by the studio in the way that the REAL table is used by the laboratory – as an environment with its own elements of novelty and surprise, where the research encounters resistance and resistance generates understanding. The landscape does not exist as a fixed object awaiting human intervention but is emergent through intra-actions with sensors, infrastructures, organisms, flows, and management practices (Barad 2007, 206). The “landscape” is not a noun but a verb, an ongoing process of material-discursive entanglement. To treat landscape as model is to acknowledge this becoming, to design infrastructures that participate in it rather than attempting to ossify it, and to cultivate the humility to learn from the knowledge that emerges.
Ecological Fitness
Toward an Evaluative Framework
If landscape is to function as model then the question of how to evaluate its performance becomes foundational. Assessment regimes in restoration ecology and environmental management rely on historical baselines focusing on a pre-disturbance condition, a pristine state to which the landscape might return. Yet the conditions that produced those historical configurations are themselves shifting and historical baselines are moving targets, and in many cases they are fictions, artifacts of an environment that no longer exists.
Novel ecosystems, systems that have crossed biotic or abiotic thresholds into configurations without historical precedent, now constitute an increasing proportion of the Earth’s surface (Hobbs, Higgs, and Harris 2009). A wetland restored to historical species composition may fail because temperature and precipitation regimes have shifted beyond the tolerance envelopes of those species. The landscape-as-model requires an evaluative framework that does not depend on returning to states that are unobtainable, but instead assesses performance in terms that acknowledge indeterminacy, change, and the impossibility of equilibrium.
The concept of ecological fitness offers such a framework. Borrowed from evolutionary biology, fitness in its broadest sense refers to those properties of organisms or systems that are explanatory of persistence and flourishing and inherently predictive of future capacity to survive under conditions that cannot be anticipated (Peacock 2011). Fitness in this sense is ecological rather than narrowly genetic and it is defined by the interactions between organisms and their environments, and the capacity of those interactions to sustain life over time.
Kent Peacock identifies three dimensions of ecological fitness that move beyond the familiar Darwinian emphasis on competition. The first is the ability to compete, to secure resources, to occupy niche space, to outperform rivals for light, water, nutrients, or territory. This competitive dimension has dominated evolutionary thinking since Darwin and remains central to how ecosystems are conventionally assessed by species richness, population densities, trophic efficiency, and biomass accumulation. But competition cannot be the only trait that can account for the persistence of complex systems.
The second dimension is an ability to cooperate through the formation of mutualistic relationships, symbiotic partnerships, and facilitative interactions where the success of one organism or assemblage depends on and at times supports the success of others. Coral reefs, mycorrhizal networks, pollination mutualisms, and nitrogen-fixing associations are demonstrations that fitness is as much about collaboration as about competition.
The third dimension, and the one that has received insufficient attention in both evolutionary theory and landscape practice, is the ability to construct through the modification, engineering, and active shaping of the environments within which fitness is expressed.
The constructive dimension resonates with landscape practice. Niche construction, as articulated by biologists studying how organisms alter their environments, refers to the ways organisms modify environmental conditions through their activities such as beaver dams that create wetlands, termite mounds that regulate microclimate, or vegetation that alters soil chemistry and hydrology (Odling-Smee, Laland, and Feldman 2003). These modifications do not benefit the individual organism alone but reshape the environment for the current community and future generations.
Algorithmic Cultivation (2019) staged this constructive dynamic at the scale of the individual organism. The data operated as “an unfiltered random catalyst for provoking the feedback loop of stimulus and response.” The robot cut, the plants responded through growth patterns, branching configurations, leaf size, color changes, that the robotic system had not specified and could not have predicted.
The plants were not passive recipients of robotic intervention. They were constructing their own growing conditions through response to an external stimulus that communicated nothing about what the optimal growing condition would be. This is constructive fitness at minimum viable scale, the organism building its own niche through response to environmental provocation, without the environmental provocation specifying what the niche should look like. The installation was designed to run for a year, technical difficulties limited operation to weeks. But even in its truncated form, it demonstrated what the constructive fitness framework predicts, that biological agency, given a sufficiently varied set of provocations, produces configurations that neither the organism nor the designer fully controls. The plant was smarter than the model.
Each experiment on the geomorphology table documented a morphological configuration that would never reoccur, even if the same experimental parameters were repeated. The sediment would organize differently, the channels would follow slightly different paths, the delta lobes would form in slightly different positions. This combination – strict experimental protocol producing unique outcomes – is the defining characteristic of complex dynamic systems. They are rule-governed but not deterministic, systematic but not repeatable, following recognizable patterns while producing singular instances. The documentation practice captures this duality. The protocol and the singular outcome, the specific morphological configuration of this experiment, at this moment, irreproducible even in principle. This duality is what landscapes exhibit at every scale, from grain-scale sorting of sediment particles to delta-scale distribution of discharge. That’s the point.
I extend Peacock’s three dimensions toward designed landscapes, proposing that constructive fitness, the capacity to build conditions favorable to one’s own persistence, describes both the marsh that accretes its own substrate and the geomorphology table that generates landforms hospitable to subsequent intervention. This extension is not a borrowed framework from evolutionary biology but an original synthesis with direct operational implications for design practice.
Extending this further, a landscape’s fitness includes its capacity to construct conditions favorable to its own persistence and to the persistence of the assemblages it supports. A constructed wetland that improves its own substrate through sediment accretion and organic matter accumulation, that gradually deepens its root networks and expands its filtering capacity, that shifts conditions in ways that recruit additional species and stabilize nutrient cycles, exhibits a constructive fitness. It is not solely surviving disturbance or competing for resources, it is actively building the conditions under which survival and flourishing are more likely over time.
This reconceptualization has practical consequences for how interventions are evaluated. Rather than asking whether a landscape has returned to a historical benchmark, we ask whether it demonstrates increasing capacity to persist, to absorb disturbance, to reorganize after disruption, and to construct conditions favorable to evolving function. These capacities are not static properties but tendencies that must be assessed over time and across multiple variables. A landscape might show declining species richness but improving nutrient retention, or it may lose certain historical species while gaining functional equivalents better suited to emerging conditions, or it might experience periodic setbacks from storms or droughts while nonetheless trending toward structural complexity. Fitness, in this sense, is multivariate and progressive, a trajectory rather than a state, an optimization towards plurality.
This framework aligns with recent developments in resilience theory and adaptive management, but adds an important dimension. Resilience is often framed in terms of resistance to change or recovery to a baseline, the capacity of a system to absorb disturbance without shifting to an alternative state. But as Nassim Taleb argues, there is a category beyond resilience, antifragility, the property of systems that improve through exposure to stressors, volatility, and uncertainty (Taleb 2012). Antifragile systems do not just survive shocks, they actively benefit from them.
The principle of hormesis is that moderate doses of stressors can stimulate adaptive responses that leave organisms or systems fitter which applies to ecosystems as well as to individual organisms (Costantini, Metcalfe, and Monaghan 2010). Fire-adapted forests require periodic burns to clear understory fuel loads and release nutrients and without fire, they become vulnerable to catastrophic conflagration. Floodplain wetlands depend on periodic inundation to deliver sediment, replenish soil moisture, and maintain the disturbance regimes that prevent competitive exclusion. A management regime that suppresses all disturbance may produce a system that appears healthy in the short term but is accumulating fragility, aggregating conditions for eventual collapse (Holling and Meffe 1996).
Ecological fitness, understood as the capacity to compete, cooperate, and construct under conditions of uncertainty, incorporates antifragility as a criterion. A fit landscape does not resist all possible disturbances but is one that has been designed to learn from disturbance, to incorporate the knowledge that shocks provide, and to reorganize to maintain or enhance functional capacity. This shifts the evaluative question from “Has the system returned to its target state?” to “Is the system developing capacities that will enable it to persist and adapt as conditions continue to change?”
A restoration project evaluated through the lens of historical fidelity might be judged a failure if it deviates from pre-disturbance species composition, even if the deviation reflects adaptation to changing climate regimes. A project evaluated by ecological fitness would be judged by whether the emergent assemblage, however novel, demonstrates the functional capacities needed for ongoing persistence such as effective nutrient cycling, soil stability, hydrological regulation, habitat provision, and/or carbon sequestration.
This also reframes the relationship between design and ecology. If fitness includes the capacity to construct, then designed landscapes are not anomalies to ecological process but extensions of it. What distinguishes designed landscapes from beaver dams or termite mounds is not that they involve construction but that they involve human intention, institutional coordination, and the capacity for reflexive adjustment based on monitoring and learning. The landscape-as-model approach leverages these distinctively human capacities of foresight, communication, technology, and adaptive management in service of ecological fitness.
The integration of computational systems, distributed sensing networks, machine learning, and autonomous management further extends this constructive capacity. These technologies do not replace ecological fitness but augment the landscape’s capacity to learn and adapt. The coupled landscape, instrumented and responsive, is not less ecological than its uninstrumented predecessor, it is more capable of the continuous adjustment that addresses indeterminacy. Evans, Bratton, and Agüera y Arcas (2026) describe intelligence itself as “high-dimensional and relational, not a single quantity,” growing “like a city, not a single meta-mind.” The instrumented landscape is this kind of distributed intelligence, a territory in which human judgment, machine learning, biological response, and material agency all participate in knowledge production that no single agent controls or fully comprehends.
This is not a claim that technology solves the problems of environmental management or that computation can substitute for ecological knowledge. Instead it suggests that the capacities for sensing, learning, and adapting that define ecological fitness can be enhanced through infrastructures that extend perception and accelerate response.
Ecological fitness thus provides an evaluative framework suited to the conditions of the Anthropocene, a framework that does not require returning to states that are unattainable, that celebrates novel ecosystems and emergent assemblages, that values the capacity to learn from disturbance rather than resist it, and that recognizes designed landscapes as participants in ongoing ecological processes of construction and adaptation. It does not abandon the goals of biodiversity conservation, ecosystem service provision, or habitat protection, but it reframes those goals in terms that acknowledge the irreversibility of environmental changes and the necessity of working with the trajectories that change has set in motion. The landscape-as-model, evaluated by ecological fitness, eschews nostalgic restoration and is a progressive form of Reflexive Stewardship, a practice oriented toward futures that are unpredictable but can be prepared for through the cultivation of capacities that are valuable under conditions we cannot anticipate. The cultivant, developed fully in Chapter 11, names the practitioner’s disposition within this practice. Not control, not optimization, but ongoing attentive negotiation with biological agency that is already working.
An Expanded Medium
Socio-Ecological and Socio-Cultural Concerns
While the focus has primarily been on biophysical processes it is essential to recognize that landscape is simultaneously a socio-cultural medium. Any intervention at territorial scale operates within networks of institutions, regulations, property regimes, cultural values, and economic pressures that shape what is possible, permissible, and desirable.
A responsive sediment diversion is not only a robotic hydraulic structure designed to build land but it is also a political artifact that redistributes resources and risks, potentially delivering land to one parish while increasing salinity intrusion or flood risk for another (Paola et al. 2011). As previously mentioned a sensor network is more than a technical system for gathering environmental data, it is also a surveillance infrastructure that raises questions about who has access to data, who controls its interpretation, and whose concerns are prioritized when thresholds are set (Gabrys 2016; Borgman 2015). An interface is also a rhetorical device that frames certain futures as inevitable or desirable while rendering others invisible or unthinkable (Corner 1999).
The landscape-as-model approach must attend to these socio-cultural dimensions as rigorously as it attends to sediment budgets, flow regimes, and vegetation succession. This means designing not only for ecological fitness but also for institutional adaptability, social equity, and cultural resonance. It means engaging publics not as obstacles to be managed through public relations campaigns but as co-producers of knowledge whose observations, values, and situated expertise enrich knowledge production and produce a plurality within the outcomes (Allen 1999; Bélanger 2016). And it means recognizing that the success of a landscape intervention is not measured solely through biophysical metrics that address hectares of wetland restored, kilograms of sediment deposited per year, or the percentage increase in species richness but must also be evaluated in terms of livability, justice, and meaning.
Does the intervention support the livelihoods of a broad range of communities and species who depend on the landscape? Does it distribute risks and benefits equitably, or does it protect favored communities while exposing vulnerable ones? Does it resonate with cultural histories and practices, or does it erase them in the name of ecological optimization?
Human institutions, cultural practices, and economic systems are integral to how landscapes respond to disturbance and reorganize (Gunderson and Holling 2002, 73–94). A project that engages communities, incorporates local knowledge, and distributes benefits equitably may succeed even if its ecological outcomes are modest or take longer to materialize, as it builds the social capital and institutional capacity necessary for generational stewardship. The landscape-as-model integrates social and ecological dimensions from inception, treating them as mutually constitutive aspects of a single ecology.
Similarly, landscape monitoring and management infrastructures are political, they encode assumptions. A sensor network developed to track ecological “health” implicitly defines what is healthy, favoring metrics aligned with conservation or restoration goals while marginalizing other values, recreational access, cultural continuity, and economic productivity that residents may prioritize. The landscape-as-model must therefore attend to political dimensions, making explicit whose knowledge is valued, whose voices shape decisions, and how conflicts are negotiated.
This expanded frame does not dilute the technical rigor of the landscape-as-model approach but enriches and grounds it. A sediment diversion that functions hydraulically but simultaneously dispossesses a fishing community that relied on particular salinity regimes and access routes is not a successful experiment, it is a failed negotiation, one that ignored critical dimensions (Spirn 1984). A sensor network that produces accurate, high-resolution data on water quality and vegetation health but excludes local ecological knowledge accumulated over generations, knowledge about seasonal patterns, species behaviors, or early warning signs that sensors do not detect is epistemologically impoverished (Gabrys 2016). The landscape-as-model is a socio-ecological-technical assemblage in which human institutions, cultural practices, and political economies are as constitutive as water, sediment, and vegetation. Its design, monitoring, and adaptive management must be correspondingly multi-dimensional, integrating social science, humanities, and community-engaged research with hydrology, ecology, and engineering (Lister 2007; Meyer 2008).
Cultivating resilience requires attention to three dimensions, resistance (the ability to absorb disturbance without changing state), recovery (the speed of return to baseline after disturbance), and transformation (the capacity to shift to fundamentally new configurations when existing ones become untenable) (Folke et al. 2010, 10–12). The landscape-as-model navigates among these three. Some interventions aim to resist change armoring a shoreline or stabilizing a channel, accepting that this may reduce flexibility. Others aim to accelerate recovery, restoring vegetation after a storm or reopening sediment supply after a drought. Still others enable transformation by reconfiguring infrastructure to accommodate rising seas or shifting land use to support nascent economies. The choice depends not only on biophysical constraints but on social values, institutional capacities, and political negotiations.
If landscapes, infrastructures, and communities are mutually constituted through ongoing intra-actions, then there is no “social” sphere separate from an “ecological” sphere, and no “technical” intervention that acts only on biophysical processes (Barad 2007, 140). Every intervention reconfigures the assemblage, not just flows and vegetation but also property relations, governance structures, cultural meanings, and corporeal experiences. The sediment diversion changes not only marsh elevation but also who has access to resources, how communities navigate their territories, what stories they tell about their relationships to land and water, and what futures they can imagine. Landscape-as-model, from this perspective, is not a project of optimizing an ecosystem according to historical criteria but a practice of co-constituting new socio-ecological configurations through iterative, material-discursive engagement. It must hold multiple forms of knowing simultaneously, the biological, the computational, the institutional, and the cultural, recognizing that design knowledge is always partial and that the full picture requires plural intelligence. When that plurality is genuinely constituted, the questions of justice, meaning, and power answer themselves.
If the landscape is a medium that is alive, responsive, and always in process then the designer’s relationship to it cannot be one of specification and control. It must be one of ongoing interaction. But what does that interaction look like? What frameworks exist for designing systems that continuously learn from the environments they are shaping? How does the feedback loop between intervention and response get formalized as method?