The modeling genealogy traced in Chapter 6 arrives at a question it cannot answer from within its own frame. If the MRBM’s failure was partly an epistemological failure, what kind of sensing would have to be in place for a model to produce different knowledge? The question is not technical. It is epistemological, and the epistemological stakes have political consequences. The sensing apparatus I have built across two decades of practice, from the blob detection camera reading a painted mural in a university atrium to the internet of ecologies proposed for a 170-kilometer desert territory, is not neutral infrastructure. Each instrument, each datum, each resolution threshold embeds a decision about what matters, whose landscape becomes legible, whose experience is registered, what forms of knowledge can be produced at all. This chapter examines that politics directly. The Technogeographies of Sensing, the spatial and political formations that arise when sensing technologies are embedded in territory, determine not only what can be known about a landscape but who benefits from that knowledge and who bears the cost of what the instrument cannot see. The neo-wilds that emerge in the gaps of technogeographic coverage are territories doing their own work, organizing, accreting, adapting, outside the frame of any designed attention. They are not defined by what the apparatus misses but by what they are already doing without it. Parks (2013) names these formations “signal territories”, landscapes constituted through broadcast infrastructure, where the materiality of sensing equipment produces spatial and political effects independent of the data it collects.
A Case for Highly Maintained Wildness
The coupled landscapes emerging from distributed sensing, responsive infrastructure, and adaptive management protocols represent a new category of environmental condition, one that challenges conventional distinctions between wild and managed, natural and artificial, autonomous and controlled. These are landscapes of highly maintained wildness, territories where intensive technological intervention enables, rather than suppresses, the autonomy of myriad species, processes, and material agencies. The wildness in question is not the romanticized pristine wilderness imagined as separate from human influence, but a constructed condition in which design creates the infrastructural scaffolding for diverse, unpredictable, and self-organizing ecologies to flourish (Marris 2011; Del Tredici 2010).
This concept inverts the traditional conservation logic that equates wildness with absence of human management. Instead, it recognizes that in an era of pervasive anthropogenic change, maintaining ecological dynamism often requires continuous, technologically mediated intervention (Cook 1999; Cantrell and Dawkins n.d.). The goal is to choreograph conditions under which non-human agencies can express their autonomy, not to restore a mythical pre-human baseline, adapt to novel circumstances, and generate emergent ecological configurations that were not and could not have been predetermined by designers (Morton 2009, 2012; Bennett 2010).
Wildness, in this frame, is measured not by the absence of human infrastructure but by the degree of autonomy granted to non-human agents and processes within technologically mediated environments. A salt marsh instrumented with sensors, nourished by calibrated sediment diversions, and monitored through distributed robotics has potential to be wilder, more ecologically dynamic, more resilient, and more generative of unexpected species assemblages than a “protected” marsh left to drown under rising seas because human infrastructure upstream has cut off its sediment supply. The former accepts the entanglement of human and non-human agencies and deploys technology to amplify, rather than constrain, ecological possibility (Gabrys 2016; Bryant 2014).
Computation, Control, and Designed Wildness
The relationship between computational management and ecological autonomy in neo-wilds has deep roots in landscape architecture. Ian McHarg’s Design with Nature (1969) pioneered computational approaches to environmental design through overlay analysis, using transparencies to combine multiple environmental data layers and identify optimal development locations (McHarg 1969). McHarg’s method was deterministic, inputting ecological constraints to output prescribed land uses, yet it established that computation could mediate between human intentions and environmental processes.
Margot Lystra demonstrates that McHarg’s approach was explicitly informed by cybernetics, interpreting landscapes as systems maintaining equilibrium through feedback loops and entropy reduction (Lystra 2014). McHarg sought to design with self-regulating systems rather than imposing forms requiring continuous energy expenditure to maintain. But McHarg’s cybernetic vision remained first-wave, it assumed systems tend toward equilibrium and that designers could predict and specify desired stable states (Zhang 2025). His computational overlays identified the “right” land uses as if environments were static puzzles with single correct solutions.
Lawrence Halprin, working contemporaneously, took a different approach. His RSVP Cycles, Resources, Scores, Valuaction, Performance, conceived design not as determining fixed outcomes but as establishing frameworks within which participants (human and non-human) could improvise (Lystra 2014; Halprin 2003). Halprin’s “scores” were instructions that invited interpretation and variation, more like musical notation than architectural blueprints. Halprin embraced chance and unpredictability as creative forces, aligning his practice with second-wave cybernetics’ recognition that systems produce emergent behaviors that designers cannot fully control or predict (Zhang 2025).
Neo-wilds synthesize these lineages while moving beyond them. Like McHarg, neo-wilds employ computational analysis to work with environmental processes but rather than overlaying data to find predetermined optimal solutions, algorithms continuously monitor conditions and adjust infrastructure in real-time based on feedback (Cantrell, Ellis, and Martin 2017; Gabrys 2016). Like Halprin, neo-wilds establish frameworks within which ecological processes improvise but the “performers” include not only humans but organisms, materials, and algorithms, all participating in ongoing landscape co-production (Haraway 2016; Tsing et al. 2017).
From Situated Landscapes to Datafied Territories
Benjamin Bratton’s concept of computation as “existential technology,” technology that changes not what we can do but what we can know provides a planetary frame for the technogeographic argument developed here (Bratton 2025). Drawing on Lem’s distinction between instrumental and existential technologies, Bratton argues that planetary computation, the network of sensors, satellites, oceanic monitoring systems, and supercomputing simulations is the apparatus through which humanity deduced the Anthropocene itself. Without computational sensing, there is no concept of climate change and without algorithmic processing of global data streams, there is no understanding of earth system dynamics at planetary scale. This insight deepens the politics-of-sensing argument and the technogeographic question is not only who and what gets measured, but that the very act of measurement through computational infrastructure constitutes a Copernican trauma, a forced recognition that the world does not work the way our inherited models assumed. What the sensing networks described in this chapter produce is not data about territory alone but they produce a new epistemological condition in which the observer is no longer external to the system being observed but constitutively embedded within it.
Contemporary landscapes are increasingly defined not only by their visible form but also by the invisible infrastructures of sensing, computation, and communication that render them legible to distant observers and responsive to real-time conditions. The landscape becomes a datafied territory, a hybrid entity composed of biological processes, physical materials, and streams of digital information that continuously report on the system’s state, trajectory, and thresholds (Gabrys 2016).
This transformation reframes what a landscape is. Landscape architecture has traditionally understood landscapes through visual, experiential, and spatial categories, prospect and refuge, figure and ground, sequence and threshold. Datafied territories demand additional lenses, temporal resolution (how frequently is the system sampled?), spatial coverage (which processes are monitored and which remain invisible?), and algorithmic interpretation (which patterns trigger alerts or interventions?) (Seibert 2021). The landscape becomes a computational environment, not a site where computation happens to occur, but a system whose behavior is constituted through the interplay of biophysical processes and their digital representations.
Datafied territories are also communication systems, networks of transmitters (sensors), channels (data infrastructure), and receivers (algorithms, human managers) through which information flows about environmental states (Shannon 1948). The territory is not simply observed through sensors but constituted through these communication flows. Its behavior is partially determined by the information circulating within it and the control actions that information triggers (Gabrys 2016).
Over/Under: Lausanne Jardins (2009), developed with Allen Sayegh, Edith K. Ackermann, and Marcella Del Signore, staged this constitution of landscape through data at the scale of a single plant. A specimen citrus tree occupied a plaza above a subway station in Lausanne, sustained by visible hydroponic infrastructure, an explicit homage to the 17th-century orangery tradition of keeping exotic species alive through technological prosthesis. Below, in the subway station, the specimen was digitally replicated and transformed into a shifting forest of projections across station walls, responding to the living plant’s seasonal changes. Commuters moving underground encountered a botanical environment derived from, but exceeding, the living plant above.
The installation demonstrated that the datafied territory is a different landscape altogether from the physical territory it monitors. The citrus tree in the plaza was a living specimen sustained by visible hydroponic infrastructure in a climate where it would die without it. By making that infrastructure visible, the project refused to conceal the technological conditions of the plant’s existence. Below, in the subway station, the digital forest was a transformation of the citrus tree, not a simulation, a landscape constituted through sensing and projection that could not exist without the physical plant but was not reducible to it. The sensed landscape, detached from physical presence and reconstructed elsewhere through technological mediation, carried its own aesthetic, its own spatial logic, its own relationship to time. The tree became the data source, the subway station the sensing context, and commuters the interpretive community. The territory enacts a circuit in which the sensed environment, the sensing infrastructure, and the community all participate in producing the knowledge the system makes available.
Wiener warned that cybernetic systems can exhibit pathological behaviors when feedback loops are poorly designed or when control actions operate at timescales mismatched to the processes being managed (Wiener 1948). The challenge in designing neo-wild management of Coupled Ecologies in datafied territories is ensuring that multiple feedback loops, operating at scales from milliseconds (sensor sampling) to decades (ecosystem succession) reinforce rather than interfere with one another (Cantrell, Ellis, and Martin 2017; Raxworthy 2018).
The measured landscape is an enacted landscape, brought into being through the specific practices and technologies of observation (Latour 1987, 2004). This enactment has material consequences. The unmeasured dimension remains invisible until a threshold is crossed and the system flips to an alternate state (Gunderson and Holling 2002). By that point, restoration becomes far more costly or impossible.
Branding Islands Making Nations (2016), examined this enactment at its most explicit, the production of sovereign territory through representational practice. The project’s relevance to technogeography is direct. The datafied territory does not merely observe landscape through sensing, it enacts landscape through the representational practices that sensing enables. A satellite image that registers a new landmass where ocean existed last month performs the same constitutive function as the military cartography that claims it as sovereign ground. The sensing apparatus does not describe what is there, it brings into political existence what can be governed. As Pietrusko (2020) argues, “the creation of data must be understood as an act of environmental design”, designers must engage with the upstream sites of data production, not only the downstream visualizations. The designer’s complicity in this process, the capacity to choreograph land-making operations is also the capacity to enable territorial expansion, displacement, and conflict, is not an ethical supplement to technogeographic practice but its foundational condition. Integrating GIS data imports the epistemic infrastructure of the agencies that collected it, the sensing systems, classification schemes, and resolution choices. The territory therefore appears through the lens of that infrastructure, not as a neutral reality. The gap between the sensed data layer and the underlying landscape is not a technical problem to be calibrated away. It is where the politics of sensing become visible, where the designer confronts what has been made legible and what has been left out.
Each of the six modes in Responsive Landscapes (Elucidate, Compress, Displace, Connect, Ambient, Modify) encodes a distribution of agency. Sensing configurations shape what kinds of knowing are possible, and the political consequences follow directly. Whose landscape is monitored, whose is altered, whose is controlled, these are decisions embedded in the technical apparatus before any data is collected.
Autonomy Through Infrastructure
Enabling Non-Human Agency
The concept of highly maintained wildness rests on a paradox, intensive human intervention creates the conditions for non-human autonomy. This is not the autonomy of isolation, the fantasy of nature “left alone” to function without human influence, but the autonomy of possibility. Landscapes structured to allow diverse species, ecological processes, and material agencies to express themselves in ways that are unpredictable, adaptive, and irreducible to human intention (Bennett 2010; Morton 2012).
The notion that landscapes appearing wild might be products of intensive management has precedents. Anne Whiston Spirn argues that Frederick Law Olmsted’s parks, celebrated for their naturalistic appearance, required sophisticated engineering to create and maintain, Central Park’s “natural” streams were constructed waterworks, its “wild” forests were carefully composed plantings (Spirn 1996). Olmsted understood that creating the appearance of nature in cities required technical infrastructure that remained hidden, the logic of neo-wilds, where computational management operates largely invisibly to sustain ecological dynamism.
But Olmsted’s wildness was primarily visual, it created the appearance of nature through carefully controlled compositions. Neo-wilds extend this logic beyond aesthetics to embrace functional ecological processes generating unpredictable outcomes (Cantrell, Ellis, and Martin 2017). Where Olmsted specified plant locations and maintained fixed compositions, neo-wilds establish initial conditions and then allow vegetation to self-organize through competitive sorting, with monitoring systems tracking outcomes and adjusting infrastructure if trajectories diverge from desired ranges (Felson and Pickett 2005). The wildness is not just visual but generative, a capacity to produce novel configurations that designers could not have specified in advance.
Wiener distinguished between control through rigid determination and control through feedback, arguing that truly effective systems preserve autonomy for components while maintaining coordination through information exchange (Wiener 1948). A clock controls its hands deterministically, a thermostat controls temperature through feedback that responds to actual conditions. Neo-wild infrastructure operates more like thermostats than clocks as it establishes target ranges (desired sediment deposition rates, acceptable salinity fluctuations, minimum habitat areas) but allows actual conditions to vary based on how ecological processes respond to local circumstances (Cantrell, Ellis, and Martin 2017).
This is what Wiener meant by “the human use of human beings” and creating technologies that enhance rather than suppress autonomy, that work with rather than against the self-regulating capacities of complex systems (Wiener 1988). Applied to neo-wilds, this principle extends beyond humans to encompass the autonomy of all agents including organisms selecting habitats, materials sorting themselves, algorithms detecting patterns, and communities negotiating values. The computational infrastructure does not command these agents but provides information and adjusts conditions within which they exercise their capacities (Haraway 2016; Tsing et al. 2017).
Orchestrating Multi-Species Assemblages
Highly maintained wildness requires designing not for a single target condition but for conditions that support multiple autonomous agents operating simultaneously. This means creating infrastructural scaffolds, the hydrological regimes, sediment supplies, nutrient gradients, and structural complexity that enable diverse species and processes to find niches, interact, and co-evolve (Tsing et al. 2017; Haraway 2016).
This iterative approach contrasts sharply with traditional restoration paradigms that specify a target ecosystem type, plant palette, and hydrological regime in advance. Such projects often fail because they underestimate the complexity of ecological assembly and the unpredictability of species interactions (Lister 2007). By contrast, projects designed for maintained wildness focus on creating infrastructural preconditions for self-organization (Allen 1999; Corner 1999).
Critically, this approach embraces novel ecosystems, assemblages of native and non-native species, operating under altered hydrological and climatic conditions, that have no historical precedent (Hobbs, Higgs, and Harris 2009; Del Tredici 2010). In a world where climate zones are shifting and species are migrating, the notion of restoring ecosystems to pre-disturbance baselines becomes increasingly untenable. Novel ecosystems may be the most resilient configurations available under current and future conditions (Marris 2011).
Machine Intelligence and the Appearance of Autonomy
Defining Neo-Wilds
Where Raxworthy’s viridic emphasizes landscapes that embrace processes of verdant growth, decay, and succession as primary design drivers, privileging time-based change over spatial composition (Raxworthy 2018), the cultivant, the disposition I develop in Chapter 11, names the ongoing relationship between designed intention and biological agency, where maintenance is the primary design act. The territories emerging from computational sensing and responsive infrastructure extend this logic to territorial scale. These landscapes are neo-wilds, one of the three central contributions of this dissertation, environments that appear wild to human observers, performing as ecologically dynamic and visually complex, yet maintained through continuous, algorithmically mediated interventions operating at scales and temporal resolutions that render them largely invisible to direct perception (Cantrell, Ellis, and Martin 2017). The neo-wild is cultivance made territorial, the cultivant relationship scaled beyond the individual organism or garden to the landscape itself.
Neo-wilds represent a synthesis of the cultivant sensibility with the computational management frameworks developed through adaptive management and coupled ecology (Egerstedt 2021; Gabrys 2016). They are landscapes where machine intelligence works alongside human decision-makers, ecological processes, and material agencies to choreograph conditions that maximize autonomy for non-human agents while ensuring long-term persistence of valued ecosystem functions. The result is a paradox, landscapes that are seemingly untouched, spontaneous, and wild are among the most carefully orchestrated environments humans have ever produced.
Charles Waldheim identifies “strategies of indeterminacy” as central to contemporary landscape practice, where designers establish frameworks and processes rather than fixing final forms (Waldheim 2006). Projects like Field Operations’ Fresh Kills Park explicitly reject predetermined outcomes in favor of adaptive frameworks responding to changing conditions, user appropriation, and ecological succession. Neo-wilds extend these strategies into computational territory, deploying sensor networks and algorithmic management to engage indeterminacy at multiple temporal scales simultaneously to produce hourly adjustments to water levels, seasonal responses to vegetation growth, decadal responses to sea-level rise (Cantrell, Ellis, and Martin 2017; Gabrys 2016).
Indeterminate Futures (2021), developed with Xun Liu for the Venice Architecture Biennale, is a technogeographic infrastructure built to make the unrepresentable representable. The geomorphology table produces configurations, delta lobe positions, braiding patterns, depositional front geometries, that no researcher specified and no pre-programmed rule generated. They emerge from interactions between designed conditions (flow regimes, sediment mixes, gate sequences) and the material dynamics of water, sediment, and gravity that no designer fully controls. Minting each 15 to 30 second increment of video as an NFT on the Tezos blockchain was not a market gesture. It was an archival strategy. Each token gave a singular morphological event, one that would never recur even if the same parameters were run again, a persistent identity outside any institutional repository. Each token was a record of something that had already vanished, a landscape that existed for seconds before the flow reorganized it. The archive is not a representation of indeterminacy but a practice of attending to it, building the technogeographic infrastructure through which singular, unrepeatable landscape events become available for knowledge production. What it made legible, the research had to attend to. What it could not capture, the smell of the synthetic sediment, the sound of the flow reorganizing, remained outside the record. The politics of sensing operate at the scale of the table exactly as they operate at the scale of the territory.
The Tezos blockchain was selected because it creates an archive resistant to centralized control, to institutional failure, and to the political decisions of any single institution that might suppress, revise, or fail to maintain documentation. The hic et nunc platform eventually ceased operations. The tokens remained accessible. For long-duration research whose relevance may not be apparent for decades, distributed persistence is an epistemological necessity, not a technical nicety. The Indeterminate Futures archive is an argument for distributed persistence as a component of long-duration research design.
This is the condition the neo-wilds concept names at every scale, not the absence of human influence, but the presence of human influence that has established conditions within which something unexpected can appear, what Solà-Morales (1995) recognized in the terrain vague as productive urban indeterminacy, extended here from the abandoned lot to the computationally managed territory. The designer’s role is not to specify the neo-wild but to cultivate the conditions from which it can emerge, and to build the sensing infrastructure that makes its emergence legible before it disappears.
Machine Intelligence as Mediator
In neo-wilds, machine intelligence functions as a mediator among competing processes, values, and temporal scales (Cantrell, Ellis, and Martin 2017; Olden, Lawler, and Poff 2008). Algorithms trained on years of distributed sensor data identify correlations that human observers may not see, the relationship between spring rainfall timing and summer vegetation productivity or the threshold of salinity beyond which freshwater marsh species begin to decline. These patterns inform real-time infrastructure adjustments, opening diversion gates when conditions favor land-building, closing them when ecological thresholds approach, operating continuously but unobtrusively. The patterns that matter are rarely single-variable. They operate within multivariate relationships where conditions interact to produce plural outcomes rather than singular predictions.
Crucially, machine intelligence in neo-wilds does not dictate outcomes but expands the range of possibilities of which landscapes can be imagined. An algorithm managing a sediment diversion does not specify where vegetation should grow or which species should colonize, instead it ensures that sediment is available when and where it can be captured by ecological processes, leaving the specifics to the autonomous actions of plants, animals, and materials (Cantrell, Ellis, and Martin 2017).
The Appearance of Wildness
For human visitors, neo-wilds often feel wild in ways that traditionally managed landscapes do not. They lack the visual signatures of intensive management (straight edges, uniform plantings, mowed lawns, pruned shrubs). Instead, they exhibit heterogeneity, the irregular textures, patchy distributions, and seasonal dynamism that characterize the cultural expectations of what wildness looks like (Nassauer 1995; Gobster et al. 2007).
This appearance is not accidental. While it is performance neo-wilds are designed to look wild because visual wildness is culturally valued and because it signals ecological functioning to publics who may not have technical expertise to interpret monitoring data (Nassauer 1995). But the appearance is produced through careful orchestration allowing competitive sorting rather than imposed as monocultures. The irregularity, the patchiness, the dynamism are cultivated outcomes, though specific configurations emerge from autonomous ecological processes working within infrastructurally defined ranges.
This raises questions about authenticity. Is a landscape that appears wild but is maintained through algorithmic management genuinely wild, or is it a simulation? The question assumes a binary between authentic (unmanaged) and simulated (managed) wildness that neo-wilds collapse. They are simultaneously real in that vegetation grows, animals forage, sediment deposits. They are also constructed and infrastructure enables these processes, algorithms modulate conditions, and humans design the system. The authenticity lies not in absence of intervention but in the autonomy granted to ecological processes within the system (Cantrell, Ellis, and Martin 2017; Barad 2007).
Neo-Wild Aesthetics
The Cultural Production of Neo-Wild Landscapes
Neo-wilds produce aesthetic experiences that unfold across generational timescales rather than in the immediate apprehension of form (Meyer 2008; Berleant 2012). Unlike traditional designed landscapes, which prioritize legible visual composition, neo-wilds reward sustained attention and repeated engagement across seasons, years, and generations. A marsh nourished by responsive sediment infrastructure does not reveal its full character in an afternoon but across the arc of ecological emergence with the gradual rise of marsh platforms, the colonization sequence as pioneers give way to mature communities, the return of animal species that had abandoned the site (Spirn 1984; Raxworthy 2018).
This temporal depth creates conditions for neo-wild attachment, a form of place-based belonging that emerges from witnessing and participating in landscape transformation over time (Tuan 1977). Residents who observe a marsh restoration from initial construction through decades of ecological development acquire intimate knowledge of the site’s rhythms, tendencies, and surprises. They notice when certain bird species arrive earlier in spring, when storm tides overtop newly built levees, when invasive plants are outcompeted by natives. This accumulated experience generates attachment not to a static aesthetic but to the landscape’s capacity to change, adapt, and persist (Spirn 1984; Nassauer and Opdam 2008).
Children who grow up alongside neo-wilds inherit a different environmental imagination than those raised in landscapes frozen by maintenance regimes that suppress change (Louv 2005). They internalize that landscapes are not stable backdrops but active participants in ongoing negotiations among species, materials, and infrastructures. They understand, perhaps intuitively, that the apparent wildness they experience is a product of careful management, but this knowledge does not diminish their attachment, rather, it deepens their appreciation for the complexity of maintaining ecological autonomy in an era of pervasive environmental change.
Interspecies Co-Evolution and Computational Memory
Neo-wild attachment is not exclusively human. Other species also develop relationships to these landscapes that deepen with time and co-evolution. Migratory birds that return annually to a restored marsh begin to depend on its specific configurations, channel depths providing foraging access, vegetation densities offering nesting cover, food webs supporting energetic needs during breeding or migration (Wiens 1989). As the marsh continues to evolve, the birds adapt their behaviors, selecting different foraging sites, adjusting arrival times, or modifying nesting strategies. This ongoing negotiation is a form of interspecies attachment, a reciprocal relationship in which both landscape and organisms continually adjust to one another (Tsing et al. 2017; Haraway 2016).
What distinguishes neo-wilds from “natural” ecosystems undergoing similar co-evolutionary processes is the presence of computational mediation. Machine intelligence adjusts water levels to maintain habitat conditions, modulates sediment delivery to ensure vegetation can establish, and balances nutrient inputs to prevent eutrophication while supporting productivity and at that same time builds algorithmic attachments (Cantrell, Ellis, and Martin 2017). These adjustments are informed by monitoring data tracking species responses and if bird populations decline, algorithms flag the change and prompt investigation. Over decades, this mediated co-evolution produces landscapes adapted to the particular species assemblages inhabiting them, just as species are adapted to the landscape.
The computational systems themselves also carry memory. Machine learning models trained on decades of data embody accumulated understanding of how the landscape responds to different conditions and interventions. When new managers arrive, they inherit not only physical infrastructure but also algorithmic knowledge of models that can predict how the system will likely respond to proposed changes, databases of past interventions and their outcomes, and decision-support tools encoding the reasoning of previous generations. This computational memory does not replace human judgment but provides a foundation on which new generations can build (Cantrell, Ellis, and Martin 2017; Edwards 2010).
Technogeographies
Distributed Agency and Territorial Computation
Technogeographies emerge when landscapes are understood not as passive surfaces awaiting human inscription but as multi-agent systems in which human infrastructures, sensing technologies, ecological processes, material agencies, and social institutions are all active participants (Latour 2004; Bryant 2014). Each agent operates according to its own logic, constraints, and affordances, and the landscape that emerges is a product of their interactions rather than the implementation of a singular design vision (Tsing et al. 2017; Cantrell, Ellis, and Martin 2017).
This perspective aligns with object-oriented ontology and actor-network theory, both of which insist that agency is distributed across human and non-human actors and that no single agent can unilaterally determine outcomes (Latour 1987, 2004; Morton 2013). A sediment diversion does not simply deliver sediment but it negotiates with river flows, which vary stochastically. The sediment negotiates with vegetation, which traps or releases it depending on density and species. The resulting landscape is a negotiated settlement among these agents, continually renegotiated as conditions change (Barad 2007; Pickering 1995).
Designing for multi-agent systems means that designers are participants, not primary authors. They can introduce new agents (infrastructures, species, policies, algorithms), modify boundary conditions (flow regimes, sediment supplies, legal frameworks), and observe outcomes, but they cannot dictate precisely how the system will respond (Pickering 1995; Barad 2007). This is not a failure of design but a recognition of what design can and cannot do in complex, open systems. Neo-wilds embrace this limitation, designing for conditions rather than outcomes, for possibility rather than prescription (Allen 1999; Corner 1999).
Computation Beyond Code
Technogeographies are computational not only in the sense that digital technologies process data but also in the broader sense that landscapes themselves perform computation. They process energy, matter, and information through material and biological transformations (Odum and Odum 2001). A wetland “computes” nutrient loads by uptake, transformation, and export. A sediment system “computes” deposition patterns through hydraulic sorting. A forest “computes” carbon balance through photosynthesis, respiration, and decomposition (Morton 2012; Bennett 2010).
This expanded notion of computation, encompassing biological metabolisms, chemical reaction networks, and physical self-organization suggests that landscapes have always been computational environments, long before humans instrumented them with sensors (Odum and Odum 2001). What is new in neo-wilds is the hybridization with the coupling of biological, material, and computation into integrated systems that sense, process, and respond across scales and modalities (Gabrys 2016; Cantrell, Ellis, and Martin 2017).
Acknowledging landscapes as computational environments has design implications. It suggests that interventions should be conceived not as algorithmic processes. Instructions for how the system should respond to varying inputs. A sediment diversion with adjustable gates is an algorithmic infrastructure. “If turbidity exceeds X and tide is rising, open gates to Y%, else, close gates.” These are not algorithms executed by computers alone but by the landscape itself, through its material and ecological dynamics mediated by computational management.
Wildness as Open Objectives
Rejecting Optimization
Traditional environmental management often frames projects as optimization problems to maximize habitat area, minimize flood risk, sequester the most carbon per dollar spent. These single-objective framings are attractive because they allow quantitative evaluation, but they are incompatible with neo-wilds, which value ecological dynamism, species diversity, and emergent complexity over narrow performance metrics (Cook 1999; Cantrell, Ellis, and Martin 2017).
Neo-wilds reject optimization in favor of open objectives, goals that are directional but not deterministic, that specify desirable conditions without prescribing exact outcomes (Cantrell, Ellis, and Martin 2017). An open objective might be “increase sediment accretion rates to match sea-level rise,” which can be evaluated quantitatively, but it does not specify where sediment should deposit, what vegetation should establish, or which species should colonize. These specifics are left to the autonomous actions of organisms, materials, and processes operating within infrastructural and computational constraints (Marris 2011).
Wiener emphasized that cybernetic systems must be designed for “equifinality” a capacity to reach desired states through multiple pathways rather than requiring predetermined trajectories (Wiener 1948). This aligns with Beven’s critique of hydrological models in that many parameter combinations can produce equivalent performance, and insisting on “correct” parameters is both impossible and unnecessary (Beven 2006). Neo-wilds embrace equifinality as design principle and the goal is not specifying the precise configuration the landscape should achieve but establishing conditions under which diverse configurations can all satisfy core objectives of land persistence, habitat provision, storm protection, cultural meaning.
Open objectives also acknowledge that different stakeholders value different dimensions of landscape performance, and that trade-offs among competing values are inevitable (Folke et al. 2010). A marsh restoration might increase habitat for migratory birds while reducing access for recreational fishing and it might sequester carbon while increasing mosquito populations. No single configuration satisfies all values simultaneously, and there is likely under-optimization when addressing multiple goals. Adaptive management must navigate these trade-offs through ongoing negotiation. (Gunderson and Holling 2002).
Embracing Novel Configurations
Designing for open objectives means accepting that landscapes will evolve in unexpected directions. Species that were not planted will colonize restored sites. Hydrological patterns will shift in response to extreme events exceeding design thresholds. Machine learning algorithms will identify correlations suggesting new management strategies that human experts had not considered (Hobbs, Higgs, and Harris 2009; Cutler et al. 2007).
Adaptive management under open objectives is about continuously adjusting infrastructure, monitoring regimes, computational models, and management protocols in response to what the landscape reveals (Walters and Holling 1990; Williams 2011). This iterative, responsive approach aligns with resilience thinking, which emphasizes that the goal is not to freeze landscapes in a particular state but to maintain their capacity to absorb disturbances, reorganize, and continue providing valued ecosystem services and cultural meanings (Gunderson and Holling 2002; Folke et al. 2010). Wildness, in this context, is a measure of adaptive capacity, the landscape’s ability to respond to unanticipated moments, to generate novel configurations when old ones fail, and to remain legible and inhabitable as it evolves (Morton 2012; Cantrell, Ellis, and Martin 2017).
When the Instruments Disagree
The sensing apparatus does not produce a single account of the territory. It produces several, one for each sensing modality, each resolution threshold, each institutional framework that determines what gets measured and how. Those accounts rarely agree completely. The question is what to make of the gaps between them.
The conventional answer is to treat disagreement as instrumentation error, something to be resolved through calibration, averaging, or the selection of a preferred data source. This answer misrepresents what the disagreement is. When the blob detection algorithm in the Thresholds installation reads a human body as a topographic event that no conventional map would register, it was not producing error. It was producing a different reading of the same space, one made possible by a different datum, a different threshold for what counts as contrast, a different theory of what the space is for. The gap between that reading and what a human observer would report was not noise. It was the most interesting thing the installation produced. Evidence that the space was doing multiple things simultaneously, and that which of those things became visible depended entirely on the apparatus trained to see it.
This is divergence-as-knowledge. When two sensing systems produce incompatible accounts of the same territory, the gap between their accounts is information about what the territory is doing that neither system can capture alone. The waterman on Tangier Island reads the bay’s behavior through fifty years of embodied knowledge, crab migrations, oyster bar locations, the feel of the tide against a hull. The federal monitoring network reads it through dissolved oxygen levels, nutrient loads, and water temperature, the variables that matter for regulatory compliance. These two accounts of the same bay disagree regularly, and the disagreement is not a problem to be managed by deciding which source to trust. It is data about a process, probably the interaction between subsidence rates and storm tide frequency, that neither the waterman’s embodied knowledge nor the salinity gauge is designed to capture alone. The gap between their readings is the bay telling us something neither instrument was built to hear.
Designing for divergence means building sensing infrastructures that hold multiple accounts in productive tension rather than resolving them into consensus. The NEOM internet of ecologies is constituted to produce exactly this kind of productive divergence. Biodiversity corridor sensors, species migration tracking, hydrological monitoring, and field observation generate incompatible accounts of the same 170-kilometer corridor simultaneously. When those accounts conflict, when the sensor data suggests stable migration patterns while field observation records behavioral disruption, that conflict is the research finding. The territory is communicating something that no single sensing modality anticipated. The infrastructure’s job is not to eliminate the conflict but to hold it open long enough to ask what it reveals.
Three forms of productive divergence recur across the practice. The first is calibration divergence. Different instruments measuring the same variable at different resolutions produce readings that cannot be directly compared, and the incompatibility reveals the scale at which the process being measured is actually operating. The second is modality divergence. Embodied knowledge and technical sensing produce accounts that are incommensurable because they are registering different dimensions of the same phenomenon, and the gap between them is the phenomenon’s full complexity. The third is intentional divergence. Sensing apparatus designed by different institutions with different mandates produces accounts organized around different questions, and placing those accounts in dialogue reveals the political structure of what each institution is built to see.
The datum is a design decision. So is the choice to hold incompatible readings together rather than selecting among them. That choice is the epistemological move that distinguishes an adaptive sensing practice from a surveillance apparatus.
Asymmetries of Technogeographic Power
The epistemological stakes are starkest when the same apparatus produces radically different knowable conditions. At NEOM, plural sensing produces plural knowledge. At Tangier, thin sensing produces thin knowledge, and thin knowledge produces inadequate governance. The political consequences are downstream of that epistemological asymmetry.
Tangier Island in the Chesapeake Bay is losing roughly fifteen feet of shoreline per year to wave erosion and subsidence. Its population has declined from over a thousand to fewer than four hundred. The watermen who remain carry generations of embodied knowledge about the bay’s behavior, crab migrations, oyster bar locations, tidal patterns, the slow inward creep of the marsh edge. Federal monitoring programs measure the Chesapeake’s water quality, dissolved oxygen, and nutrient loads, the variables that matter to regional regulatory compliance. They do not measure, at the resolution required, the sediment dynamics, marsh accretion rates, and wave energy patterns that determine whether Tangier Island exists in thirty years. The island’s wetware, its marsh grasses trapping sediment, its oyster reefs attenuating wave energy, its tidal flats cycling nutrients, is functioning, but it is functioning below the threshold at which the monitoring infrastructure registers it as significant. Who gets measured determines who gets protected.
The NEOM consultation (2022–25) is the chapter’s clearest demonstration of technogeography as governance. The internet of ecologies proposed for the project, an interconnected network of biodiversity hotspots, ecological corridors, and migration routes relaying real-time data through local sensors and remote monitoring, is not a monitoring system appended to a designed landscape. It is a designed sensing infrastructure whose function is inseparable from the landscape’s function. The sensing network makes the landscape legible to the management systems that will respond to it, determines which ecological conditions are visible and which are invisible to those systems, and shapes what adaptive interventions are possible. What remains outside the sensing apparatus remains outside governance entirely. The design work was inseparable from decisions about what to sense, which migration routes, which species, which hydrological thresholds. Those decisions are not purely technical. They are political. Who commissioned the design determined which ecologies would have a claim on the territory. The technogeographic question is not only who controls the datum but who decides what the sensing system is calibrated to detect, at what resolution, with what frequency, and whether the internet of ecologies serves the ecological communities it monitors or the institutional interests of the actors who control the sensing infrastructure.
The contrast is the chapter’s most important political observation. The same technogeographic dynamic, sensing apparatus determining what becomes legible and therefore what can be governed, produces radically different outcomes depending on who controls the sensing infrastructure and whose landscape is being made legible. At NEOM, sensing extends governance to territorial scale, at Tangier, the absence of sensing renders a community invisible to the systems that determine its survival. The choice of metric, sea level rise (which makes Tangier’s loss appear gradual and manageable) vs. shoreline erosion rate (which makes it appear catastrophic and urgent), determines legal standing, federal investment, and community survival. Edwards (2010) demonstrates that data always requires interpretive infrastructure, that the distance between raw measurement and actionable knowledge is never zero, and the institutions that control that interpretive distance control the politics of environmental response.
When more powerful sensing infrastructure is deployed within this same epistemological frame, the political consequences do not diminish. They intensify. The territory becomes more legible to the institutions that funded the sensors, on the terms those institutions defined, measuring the variables those institutions care about. Communities whose knowledge does not fit the sensing apparatus do not simply remain invisible. They become more invisible, because the computational confidence of the system makes it harder to argue that something important is being missed. The dashboard says the marsh is healthy. The waterman says the crabs are gone. In the Promethean frame, data outranks embodied knowledge even when that knowledge is registering a process the sensor was not calibrated to detect. More technology applied within an unexamined epistemology does not distribute legibility more equitably. It concentrates it.
Reflexive Stewardship, as this dissertation defines it, begins here, with the sensing apparatus itself. The practitioner’s sensing infrastructure reflects her vantage point. What she designs to be legible is shaped by where she stands, what questions she has been trained to ask, whose territory she has been funded to observe. Stewardship means actively cultivating awareness of what that vantage makes invisible, and what the territory has been working out on its own, beyond the instrument’s frame.
Wildness Through Maintenance
The landscapes that emerge from Technogeographies of Sensing, distributed monitoring, responsive infrastructure, and computational management are neither purely natural nor entirely artificial. They are hybrid assemblages in datafied territories, of Coupled Ecologies, that are perceived as neo-wilds in which intensive human intervention, mediated by machine intelligence, enables the autonomy of diverse species, processes, and material agencies (Cantrell, Ellis, and Martin 2017). They are wild not because humans are absent but because humans, working through computational systems, have constructed the conditions under which non-human agents can express themselves, adapt, and generate emergent configurations that exceed human intention (Marris 2011; Del Tredici 2010; Raxworthy 2018).
This is not a return to untouched wilderness, a fiction that is not sustainable in the Anthropocene, but a move toward highly maintained wildness. Landscapes that are simultaneously highly maintained and ecologically autonomous, technologically mediated and ecologically dynamic (Cantrell and Dawkins n.d.; Cantrell, Ellis, and Martin 2017). Such landscapes come from new forms of design practice that prioritize open objectives over optimization, embrace uncertainty and surprise, recognize that designers are participants in multi-agent systems rather than authors of fixed forms, and accept that the appearance of wildness may be the product of the most sophisticated management humans have ever achieved (Cook 1999; Allen 1999).
The technogeographies that support neo-wilds are infrastructures of possibility. They do not determine outcomes but expand the range of trajectories that landscapes can explore. They do not impose order but create conditions for self-organization. They do not suppress wildness but amplify it, demonstrating that in an era of pervasive human influence, the wildest landscapes are those that are most ecologically dynamic, most generative of surprise, most capable of adapting to uncertain futures and may be those we maintain most carefully, most continuously, and most intelligently (Morton 2009, 2012; Cantrell, Ellis, and Martin 2017).
The political implications have been named and the frameworks developed through practice, refracted through multiple projects, tested at multiple scales. Technogeography is not background infrastructure to the design argument. It is the design argument. If sensing constitutes territory, then designing sensing is designing territory, and landscape architecture, which has always worked with the material conditions of territorial formation, is the discipline equipped to do this work.
But what the sensing apparatus makes legible is not the territory itself. It is a medium, the landscape not as a site to be designed but as an instrument of knowledge production operating at 1:1 scale. Chapter 8 develops that claim directly. If the landscape can become a model, what does it mean to design with the model you are already inside?