Computational co-assembly calculating with nature to grow Islands

Abstract

Current approaches to coastal resilience often fail because they rely on prediction and control in an era of deep climatic uncertainty. This research proposes an environmental architecture that serves as an adaptive mediator, facilitating co-assembly between human intervention and natural systems. This design methodology shifts the paradigm from computed solutions to computing systems that continuously adapt via environmental feedback. Using a high-throughput computational pipeline, we explore a parameter space of geometric configurations, designing submerged structures that modulate hydrodynamic energy to direct sediment transport and grow a nascent island. These structures were deployed as a Real-World Lab in the Maldives, serving as a full-scale test of this Co-assembly approach to resilience. Over a 12-month period, the structures facilitated the accretion of nascent landforms, demonstrating an adaptive alternative to dredging. The results illustrate the potential for computational design to instrumentalize environmental science, creating architectural systems that do not merely inhabit a site but actively compute with its physical forces to build resilience over time.

Keywords

computational design self-assembly NBS CFD coastal resilience adaptive capacity CHANS

Introduction

Architectures of resilience in deep uncertainty

Architecture regulates the relationship between human systems and their environments¹ as a primary function. Our buildings and infrastructure, including coastal protection systems, shelter us from environmental threats while facilitating access to ecosystem services. The resilience of these systems can be measured by the persistence of these mediating functions in the face of external change.² Driven by climate change,³ environmental fluctuations increasingly subject the built environment to intense and unpredictable perturbations,⁴ and as Earth systems become less stationary,^5,6 historical baselines no longer provide a basis to plan for future change. In the face of deep uncertainty, resilience in the built environment is becoming more difficult to achieve.

As part of the global adaptation to deep environmental uncertainty, this research tests the computational capacity of interactions between human and environmental systems for atoll island coastal management. These islands are increasingly using armoring and mechanical dredging to manage emergent land within their atolls. Studies assessing environmental exposure and coastal resilience in the Maldives have found these interventions to be maladaptive over moderate and longer timescales.^7,8 Dredging, in particular, smothers living corals, killing the organisms that produce a majority of the sediment the islands are composed of, and undermining the islands’ intrinsic ability to adapt and grow with rising seas. A 2019 study found that the engineered response to climate change, intended to stabilize these shorelines, has degraded the adaptive capacity of 80% of the Maldives’ inhabited islands.⁹

Design aims & research questions

This project proposes and tests a computational design methodology that instrumentalizes the interface between infrastructure and environment, imbuing that boundary with the capacity for co-assembly. We ask the following questions: Can directed, ocean-driven sediment transport replace mechanical dredging while preserving the complex sediment pathways crucial to island self-organization and long-term resilience? Can we engineer submerged geometries to consistently direct sediment deposition and island growth? How can computational design leverage Real-World Labs to optimize for contexts whose complexity far exceeds our capacity to fully resolve numerically?

Nature-based

Efforts to achieve resilience through stability and robustness are failing, and coastal resiliency will increasingly rely on adaptive capacity.¹⁰ Natural and Nature-based Features (NNBF) have emerged as an adaptive alternative, building coastal resilience by guiding environmental systems to provide protective ecosystem services.¹¹ The shift towards NNBF is limited in contexts under immediate environmental threat because it can be difficult to model, predict, and monitor.^12–14 Additionally, NNBF approaches have been difficult to implement in archipelagic contexts such as the Maldives because ocean and sediment dynamics vary from island to island, preventing the dissemination of “one size fits all” approaches.^15–19

The site

The Maldives runs down the middle of the Indian Ocean from North to South, resting upon the Chagos–Laccadive Ridge, a vast submarine mountain range supporting coral reefs. The coral, growing just below the surface, is the primary source of sediment, providing the primary substrate of which the islands are composed. The Maldives has two distinct seasons. From May to November, winds and waves prevail from the southwest; from January to March, winds prevail from the northeast. December and April are considered transitional. Currents from this cycle are a primary driver of sediment transport at the target site in the South Male atoll, creating a bi-directional flow regime, with the prevailing direction shifting by approximately 180° every 6 months.²⁰

Architecture as a real world lab

This project’s computational design workflow focused on calibrating the interactions between the engineered structure and the environmental dynamics at its site. This methodology shifts the design product from computed solutions to computing instruments that extend the computational process out into the site, leveraging environmental feedback. The desired emergent landform is “computed” by the site itself, allowing it to adapt to changing conditions over time. This process follows the emergence of naturally occurring atoll islands, leveraging their intrinsic form-finding capacity to generate a persistent landform that performs like a natural island, with complex, balanced seasonal oscillation, and minimal net sediment inflow/outflow over longer timescales. Combined with ongoing monitoring and remote sensing, this collaboration with natural processes provides a Real-World Lab²¹ investigating the form-finding capacity of co-assembly at the active interface between human and non-human systems. This approach reorients the temporal scale of coastal infrastructure to its environment. Unlike the rapid, often transient land creation via mechanical dredging, this co-assembly process operates on timescales aligned with the atoll’s cyclic dynamics.

Contribution

This research proposes and tests a “co-assembly” approach to adaptive infrastructure for coastal resilience. We describe the computational pipeline used to rapidly explore a complex parameter space by connecting parametric modeling output directly to a high-throughput computational fluid modeling process. This digital process is used to calibrate the interaction between underwater geometries and the flow, moving formfinding to the “Real-World Lab” and leveraging the computational capacity of in situ physics to overcome computational limits of numerical simulation in this complex context.

Literature review

Resilience

Holling established the concept of ecological resilience in 1973 as a system’s capacity to absorb disruptions and reorganize without losing core functions.²² This differs from engineered resilience, defined as the rate of return to equilibrium. The definitions of ecological resilience and adaptive capacity in the coastal engineering literature have continued to evolve over the past two decades. The field now includes a variety of frameworks for understanding resilience at the intersection of human and environmental systems, such as the socio-ecological systems (SES) framework.²³ These frameworks largely recommend NNBF, which increasingly has demonstrated its context-dependent effectiveness.¹⁵ The Dutch Sand Motor (Zandmotor) mechanically deposited a large volume of sediment at a single location to be distributed by ocean currents over time.²⁴ Similarly, SCAPE’s Living Breakwaters integrate engineered wave attenuation with oyster reef restoration to foster physical, ecological, and social resilience.²⁵ These hybrid strategies are supported by findings from the North Frisian barrier islands,²¹ which emphasize the need to align coastal management intervention with long-term morphodynamic processes.

Maladaptation along small island shorelines

The Maldives is expanding its land area by dredging sediment, but such reclamation can harm marine ecosystems and, in the long term, make the country more vulnerable to rising sea levels.²⁶ Between 2004 and 2014, 18% of islands moved to a more disturbed type, with inhabited islands showing even more dramatic changes, where 47% moved to a more disturbed type in that decade. During this period, dredging in the Maldives has increased dramatically. The number of islands affected by sediment dredging from the island reef flat increased by 49% during the 2004-2014 period. Dredging activities have severely compromised coral reef ecosystems, undermining their two primary ecosystem services: (i) wave energy attenuation, and (ii) carbonate sediment supply to the island.²⁷

This process risks locking the Maldives into a “hard path” of coastal protection, driven by reinforcing feedback in the interactions between human and natural systems. Human pressures (such as dredging) on the shoreline and reef flat undermine the natural services of the reef-island system, reducing the island’s capacity to adapt naturally and, in turn, increasing the rate of human intervention and the associated pressures. Many Maldivian islands “now exhibit an altered-to-annihilated capacity to respond to ocean-climate pressures” over the medium to long-term.²⁸ 44% of 608 sampled islands have already at least partly lost the capacity to naturally adjust and adapt to climate-related ocean change.²⁷

Coupled human/natural systems at engineered coasts

Understanding the feedback between human activity and natural processes within hybrid systems is increasingly important for resilience in architecture and infrastructure, and a variety of frameworks have emerged to understand its complex interplay. The Coupled Human and Natural Systems (CHANS) framework²⁹ conceptualized the human-nature relationship through tight local linkages, with bidirectional interactions and feedback driving non-linear dynamics. Socio-Ecological-Technological Systems (SETS) frameworks³⁰ are frequently used to analyze the resilience, vulnerability, and adaptive capacity of coastal environments by examining larger-scale feedback mechanisms between socio-economic and ecological systems. CHANS and SETS have historically been deployed as analytical tools to describe existing conditions rather than as a generative methodology for new interventions.

Resilience-based frameworks like Panarchy³¹ emphasize the ability of a system to withstand disturbances while maintaining its core functions, often focusing on concepts like cross-scale resilience and the role of biodiversity. The progression of these frameworks over time shows an increasing sophistication in how they represent the coupling of human and natural systems, moving from cause-and-effect models to more complex representations that acknowledge non-linearities, feedback loops, and the blurring of boundaries between human and natural components.³² For example, Gunderson and Holling’s model³³ describes how fast, small-scale variables can transform slow, large-scale systems—a dynamic they term “revolt.” This research specifically focuses on these multi-scale dynamics, leveraging their computational formfinding capacity. The engineered interventions are transducers that organize multiscale fluctuations in fluid kinetic energy into continuous incremental sediment accretion.

Computational frameworks in landscape design

Similarly, computational design in landscape architecture is undergoing a shift from geometric representation towards modeling dynamic, indeterminate systems. Bradley Cantrell introduced the concepts of “responsive landscapes” and “cyborg ecologies”,³⁴ arguing that sensing technologies and feedback loops enable infrastructure to curate ecological flows rather than resisting them. The designer has shifted from modeling objects to choreographing metabolic processes. Similarly, Karen M’Closkey and Keith VanDerSys highlight how digital simulation tools enable designers to visualize and manipulate the “patterns of flux” inherent in environmental systems.³⁵ By engaging simulation with environmental dynamics as a generative engine, these frameworks lay the groundwork for architectural intervention as an instrument that modulates the interface between fluid dynamics and engineered structure.

Underwater geometry and sediment transport

To leverage the computational capacity of environmental dynamics, one must define the physical rules governing its material inputs. The transport of sediment by a flowing fluid is governed by flow dynamics and the shape and mass of sediment particles.³⁶ This relationship was captured by Shields in 1936 as the dimensionless Shields parameter, which defines the shear stress needed to suspend sediment and initiate transport. This sediment entrainment threshold is effectively the “activation energy” for the system’s form-finding process and can be plotted as a function of the sediment’s Shields parameter and the flow’s Reynolds number³⁷; the ratio of inertial to viscous forces. Subsequent research^38,39 has added nuance to the Shields parameter, capturing the contributions of sediment shape, bed slope, and other sediment properties.

In shallow atolls, flows are nearly always a combination of current and waves, and the interaction between these two forcing mechanisms is not fully captured by the Reynolds Number of the flow. In nearshore observations, wave-current flows resulted in at least 20% more bedload transport than wave-dominant flows and at least 80% more bedload transport than current-dominant flows.⁴⁰ In each of these conditions, kinetic energy was found to be proportionate to the bedload flux, suggesting it may be the most useful proxy for bedload sediment transport. Sediment transport is typically categorized into bed-load transport and suspended-load transport, based on the amount of contact the particle has with the bed during transport. In current dominant flows with low-mass sediments, suspended load may exceed bedload,⁴¹ but approximately 75% bed-load transport, 25% suspended-load transport^42,43 is typical for nearshore and reef flat conditions.

A geometry’s capacity to direct transport and trigger sediment deposition relies on the following observations:

In combined wave-current flows, transport is proportionate to kinetic energy. “Regardless of flow type, bedform migration rates were directly proportional to the total kinetic energy contained in the flow field.”⁴⁰

Sediment accumulation occurs where the sediment transport rate is lower than the surrounding area. “divergence(convergence) of the transport vectors [is] a proxy for erosion(deposition)”⁴⁴

So, geometry in a flow creates areas of lower kinetic energy, which in turn direct converging sediment transport vectors and deposition when sediment supply is present.

Summary: Systems thinking to sediment transport

This review traces a path from the conceptual models of resilience and coupled systems to the empirical science of sediment transport, areas of inquiry bound together by the multiscale dynamics of the atoll island context. This blend of abstract systems thinking and tangible material mechanics provides a jumping-off point in the literature for this multidisciplinary investigation into the use of feedback mechanisms in coupled atoll systems to build, rather than degrade, long-term resiliency.

Methods

Numerical simulation approach

The numerical simulations were intended to calibrate system interactions rather than optimize a specific state. The architectural intervention, as an instrument, was tuned to modulate the environmental dynamics to drive morphological change. Numerical simulation provided the first step in a multi-stage pipeline, moving from high-throughput sensitivity testing to physical flume validation and full-scale deployment (Figures 1-30).

Figure 1.

Simulation funnel, numeric simulation, flume testing, leading to a full-scale installation in the “Real World Lab” site.

Figure 2.

A. Simulation workflow B. Visualizing progress C. Batch of experiments.

Figure 3.

A. pipeline acyclic graph B. single run image C. Parallel Coordinate Plot.

Figure 4.

Rapid simulations, visualizing variance in short-term sediment accretion due to ring parametric variation.

Figure 5.

A. transport sensitivity to sediment size B. attenuation effect, illustrating shift in sediment transport regime (erosion to deposition) with the Hjulström Diagram.

Figure 6.

Sediment mix from site, setup in FLOW-3D® HYDRO.

Figure 7.

Accretion for onsite sediment mix, compared with single size sediments.

Figure 8.

2D Mesh inlet length’s effect on mean velocity of the fully developed flow within the ring structure.

Figure 9.

Normalized Sediment accretion sensitivity to ring structure alignment with flow direction.

Figure 10.

Sensitivity of mass averaged kinetic energy within the ring to wave height. Kinetic energy varies inversely with ramp height, even with large waves.

Figure 11.

A. average sediment accretion per m² is highest with 4-5m gaps B. as gaps increase, ring area becomes larger and more circular.

Figure 12.

Spacing contributions to accretion levels out at 30m.

Figure 13.

Two day simulation with current velocities measured onsite, showing bedford evolution over a series of current oscillations.

Figure 14.

Vertically recirculating, bidirectional flume design.

Figure 15.

Vertically recirculating, bidirectional flume in the lab.

Figure 16.

Symmetric ramp configuration in bidirectional tank, with oscillating current.

Figure 17.

Ramp rings configuration from numerical simulation, validated in a bidirectional flume.

Figure 18.

Installation of first full scale unidirectional ramp in Real-World Lab.

Figure 19.

The ring structure experiment was sited within a future island reclamation project, so all captured sediment offsets that planned and approved dredging.

Figure 20.

Installation of the ramp ring.

Figure 21.

Ramp ring over 5 months, visually documenting bedform evolution.

Figure 22.

Underwater documentation of sediment accretion within the ring, at three and 6 months.

Figure 23.

Net bathymetry change from 2023 baseline survey after 12 months.

Figure 24.

Net bathymetry change from 2023 baseline survey after 18 months.

Figure 25.

Numerical simulation of flow velocity attenuation in an engineered array.

Figure 26.

Self supporting, self-shoring synthetic array prototyping.

Figure 27.

Conceptual diagram of resilience island lifecycle.

Figure 28.

Prototyping consolidating approaches (vegetation, natural fiber structures) in Real-World Lab.

Figure 29.

Modular, incremental system for low-cost, low-risk, adaptive shoreline management.

Figure 30.

Calculating in Real-World Labs, an approach to overcome the limits of numeric computation and model adaptive responses to complex environmental challenges.

Large batches of simulations were used to systematically isolate and test sensitivity to geometric variables such as gap size, spacing, and orientation to decode the rules linking site-specific sediment entrainment and deposition to the structure’s physical parameters.

Sediment transport modeling with FLOW-3D®

The geometry calibration process used FLOW-3D® HYDRO (Version 2025R1)⁴⁵ to model the interactions among geometry, waves, current, and sediment. FLOW-3D’s Volume of Fluid (VOF) method tracks the free surface of a fluid and simulates fluid-solid interactions. All simulations used a Reynolds-Averaged Navier–Stokes (RANS)-based finite-volume method to model turbulent flows by decomposing the flow variables into mean and fluctuating components. To effectively model multi-scale turbulent energy dissipation (ε), the re-normalization group (RNG) approach was used to model ε in all simulations.⁴⁶ In Flow3D Hydro simulations that included sediment transport, the initiation of sediment motion was governed by dynamically calculating the critical shear stress criterion or Shields number.

Flow3dX pipeline

The simulation pipeline used FLOW-3D(x) to link parametric models from Grasshopper3D directly to FLOW-3D Hydro fluid simulations. This automated the systematic exploration of a complex parameter space with batch workflows and continuous results analysis.

Simulation pipeline 1 - Exploring ring configuration parameter space

Objective

Simulation Pipeline 1 broadly and rapidly explored the parameter space for ring-like configurations of discrete segments to understand the contributions of different arrangement features to sediment accretion within the ring.

Setup

Flow Velocity 0.5 m/s | Cells: 75,000 | Duration: 60 s | Runs: 30.

FLOW-3D(x)’s “Design of Experiments” DOE feature was used to generate a set of 30 simulations, testing variation in:

• Number of sides (4-100)

• Gap size ratio (0.1-0.9)

• Ring aspect ratio (0.5-2)

• Gap orientation ratio shifting gap proportion from flow aligned to flow perpendicular (0.1-0.9)

A sampling volume within the ring was used to capture the mass-averaged kinetic energy, and net sediment accretion was visualized for each run. Mass-averaged kinetic energy was used as an optimization target for these short-duration simulations, while the sediment visualization provided a sense of how the flow is interacting with the structure. This pipeline suggested:

• Wide parallel gaps correlated positively with reduced kinetic energy

• Perpendicular gaps correlated negatively with reduced kinetic energy

• More segments (smoother oval) correlated positively with reduced kinetic energy

• Width (in the dimension perpendicular to flow) correlated positively with reduced kinetic energy

The results suggested that flow alignment matters, a critical insight for designing in a bilateral flow regime.

Simulation pipeline 2 - Sediment size sensitivity analysis

Objective

Atoll reef flats include a wide range of sediment particle sizes, which is an important contributor to the shear stress required at the seabed to lift and entrain sediment. The set of simulations tested sensitivity to sediment size and compared this sensitivity against theoretical expectations.

Setup

Flow Velocity 0.5 m/s | Cells: 75,000 | Duration: 3600 s | Runs: 3

We tested three sediment sizes: 0.05 mm (very fine), 0.5 mm (average), and 5 mm (gravel). The model showed high sensitivity.

• very fine sediment eroded away

• average sediment was transported and accreted within the ring

• gravel was not transported (no accretion or erosion)

These results align with expectations from the canonical Hjulström diagram (Hjulström, 1955).

Given this high sensitivity to sediment size, samples were collected on-site and used to create a representative sediment mix for subsequent sediment transport simulations.

The simulated accretion results from the onsite sediment mix were very close to those from an average sediment size of 0.5 mm. This result validated the usefulness of prior single-sediment-size numerical simulations and single-sediment-size flume experimentation.

Simulation pipeline 3 - Meshing sensitivity

Objective

To reduce computation time for large site models, we embedded a high-resolution 3D simulation mesh within a coarser 2D “shallow water” mesh; in particular, we used the shallow-water simulation at the flow inlet to establish a vertical flow-velocity profile. To understand how this simulation approach affected flow attenuation, we varied the length of the shallow-water mesh at the inlet.

Setup

Flow Velocity 0.5 m/s | Cells: 75,000 | Duration: 360 s | Runs: 3

“The shallow water mesh” inlet was tested at 3 different lengths:

• 12 m

• 16 m

• 20 m

The mean flow velocity within the ring for each inlet length was:

• 0.35 m/s

• 0.33 m/s

Flow velocity varied 5% in this “shallow water mesh” inlet length range. Given this minimal variation, the most computationally efficient 12-m inlet was used for subsequent simulations.

Simulation pipeline 4 - Flow angle sensitivity

Objective

The bidirectional structure is expected to be sensitive to the angle of the incoming flow direction. In the flume and in initial numeric simulations, flows were perfectly bidirectional, but flows on the site will vary. Two-day flow velocity measurements at the site recorded 10-degree shifts in the average hourly flow direction. To measure the structure’s flow alignment sensitivity, it was rotated by 15-degree increments, measuring sediment accretion for each flow angle.

Setup

Flow Velocity 0.5 m/s | Cells: 85,000 | Duration: 3600 s | Runs: 3 | Sediment: site mix.

Performance dropped as the flow angle approach 15°, suggesting the intervention will only be effective where prevailing currents oscillate within a 30-degree range.

Simulation pipeline 5 - Wave interaction sensitivity

Objective

Our simulation approach assumes the current to be the dominant forcing mechanism for sediment transport at the site, and waves are omitted in numerical simulations to reduce simulation duration and computational complexity. To validate this simplifying assumption, a 2D wave simulation was used to test the contribution of waves to energy attenuation within the ring structure.

Setup

Flow: 0.5 m/s | Cells: 480,000 | Duration: 25 s | Runs: 16 | waves: stokes/cnoidal.

The test set included 16 simulations, varying barrier size and wave size, and measuring the resulting mass-averaged kinetic energy attenuation as the difference in kinetic energy outside and inside the structure.

This simulation set confirmed that the energy attenuation effects of the structure increase with wave height and with structure height. This supports the assumption that omitting waves in other simulations is conservative, because the presence of waves increases the attenuation of kinetic energy within the structure relative to that outside it. This increases the energy gradient and results in converging sediment transport vectors.

Simulation pipeline 6 - Ring gap optimization

Objective

The initial P1 batch of simulations, exploring the ring parameter space, suggested sensitivity to gaps between segments. Due to installation considerations, the site installation included 3 segments. P6 was used to understand and optimize the small gaps between the ramp segments at either end to maximize sediment accretion within the ring.

Setup

Flow Velocity: 0.5 m/s | Cells: 2,000,000 | Duration: 30 s | Runs: 7

A physically accurate parametric version of the ramp structure, with a parametric sampling area was used to test out the impacts of varying gap size.

Unlike prior simulations, conducted before a final geotextile module was determined, each ramp segment was a fixed size and shape, matching the planned site installation. Gaps were in the set of simulations, and two measurements were recorded: the normalized total sediment accretion and the area-averaged sediment accretion.

As the gaps between fixed modules widened, the total area within the ring increased. As a result, although the total accretion increased continuously with gap size, the area-averaged sediment accretion peaked at a gap of 4-5m. This suggests an optimal gap size for focusing sediment accretion, and that more spread out structures with larger gaps may continue to capture more sediment, diffused over a larger area.

Simulation pipeline 7 - Ring spacing optimization

Objective

The initial batch of simulations explored the ring parameter spacing and suggested sensitivity to large gaps parallel to the flow direction. Site conditions constrained this spacing to 30 m, so simulation pipeline 7 further investigated the initial findings using fixed ramp modules matching the planned site installation and spacing distances between 4 m and the 30-m site constraint.

Setup

Flow Velocity: 0.5 m/s | Cells: 2,000,000 | Duration: 30 s | Runs: 13 | Sediment Mix: from site.

We modeled arrays of open-ring structures based on engineered geotextile units and used FLOW-3D’s full sediment transport model. The objective was to maximize the “sediment accumulation fraction” within a parametric sampling volume for a 60-s simulation with a constant 0.5 m/s current. This optimization study revealed a strong positive correlation between sediment accretion and the spacing between the two halves of the structure. In a symmetric bidirectional flow, sediment accretion increased with spacing, leveling off as spacing approached 30m.

Simulation 8 - Long-term bedform evolution simulation (2 days)

Objective

Insights from all prior sensitivity testing informed a 2-days simulation of the proposed structure using site-specific bathymetry, flow conditions as measured over 2 days with tilt sensors, and the site’s sediment profile. The simulation modeled one characteristic tidal day of flow in each of the two dominant seasonal directions to capture the initial bed-form evolution within the ring when exposed to an oscillating bidirectional current. This full-scale site simulation was limited to 2 days by computational capacity, requiring 20 days to compute.

Setup

Flow Velocity: from site | Cells: 140,000 | Duration: 2 days | Runs: 1 | Sediment: onsite.

To reduce computation, a small 3d mesh was embedded within a 2d “shallow water” mesh, with inlet and outlet spacing at 12 m, based on prior sensitivity testing.

The progression over 48 h captured the nonlinear evolution of the bedform. The initial sediment deposition creates nascent bars, which modulate the subsequent flow-velocity gradient and the resulting accretion. At 26 h, newly formed bars from the first flow direction are influencing the deposition patterns of the reversed flow. This result from this site-specific simulation, while limited in duration, suggests that the results from prior testing translate to the site and that the oscillating monsoon current can be expected to contribute net sediment each season.

Flume experimentation

Setup

A vertically recirculating flume was designed to model current dominant sediment transport within the constraints of a 3-m tank, enabling longer duration, physical experiments to validate the numerical results. A raised platform contains the sediment, flanked by flow straighteners at either end.

Wall-mounted pumps were set up to run recirculating flow in either direction. The pumps were connected to a programmable controller, allowing each experiment to include a varying sequence of flow conditions.

All experiments were run with 0.3 mm quartz sand. The sediment bed was leveled between experiments and 3d scanned before and after. Three yellow spheres were mounted to the tank side walls to support an automated 3d-scan alignment process.

Bidirectional flow

The programmable pump configuration supports adjusting or fully reversing the flow direction. An oscillating current was used to simulate the seasonally oscillating currents on the site.

The observed patterns qualitatively support the hypothesis that symmetric ramp structures in a bidirectional flow trigger net accretion through subsequent flow cycles. This was followed by tests of ring configurations, based on results from the numerical simulation pipeline. Observations of ramp geometries in the flume captured the effects of gaps and spacing on transport pathways into and out of the ring and qualitatively validated the transport patterns observed in numerical simulations.

Site installations (Real World Lab)

Unidirectional flow

A single geotextile ramp was installed in October 2019 (Song et al., 2021) as the first full scale experiment in the Real World Lab site. The triangular prismatic bladders were fabricated by Tencate and installed in the Maldives on the shallow reef flat at low-tide depths ranging from 1 to 2m. A sand pump on the end of an excavator was used to fill the geotextile bladders. The ramp geometry was oriented such that the lower end faced east and the higher end faced west to attenuate east-to-west currents during the monsoon season.

Bidirectional flow

The unidirectional ramp triggered significant sediment accretion during the East-to-West monsoon seasons, demonstrating onsite sediment supply and supporting the deployment of a larger, bidirectional ring structure nearby. When a developer received permits to reclaim a new island, the bi-directional structure was sited within the reclamation area, such that all passively captured sediment would replace planned future dredging.

The bi-directional ramp ring was composed of geotextile ramp modules, similar to the initial unidirectional ramp, and was fabricated and installed using the previously tested methods.

Field monitoring and data uncertainty

Hydrographic and topographic data acquisition

Surveys were conducted in April 2025 and October 2025 at the Growing Island Project Site (Emboodhoo Finolhu, South Male Atoll) to characterize the seabed morphology and quantify sediment accumulation relative to the 2023 baseline. A hybrid acquisition approach was employed to resolve the land-sea interface, specifically around the submerged sand-filled bladder structures.

Deep-water bathymetry was acquired using a Sonarmite single-beam echo sounder (SBES) integrated with a CHCNAV Global Navigation Satellite System (GNSS) receiver. The transducers were pole-mounted to the starboard side of a shallow-draft vessel (4.5–6.0 m length; 0.3–0.6 m draft). The survey followed a systematic grid pattern with a line spacing of approximately 20 m. Data acquisition was timed to coincide with high tide to maximize coverage in the shallow lagoon environment and ensure navigable depth for the vessel.

To address the “white ribbon” zone (the area too shallow for the vessel but too deep for standard optical remote sensing), a topographic “walkaround” survey was utilized. This involved a surveyor traversing the perimeter of the sand bladders and shallow banks, carrying the CHCNAV GNSS rover on a pole. This method utilized Real-Time Kinematic (RTK) positioning to capture precise edge measurements and crest elevations of the artificial structures, ensuring ground-truth continuity between the submerged and emergent terrain.

Geodetic control and data processing

Horizontal positioning was referenced to the World Geodetic System 1984 (WGS84), projected to Universal Transverse Mercator (UTM) Zone 43N. Vertical control was referenced to the Maldives Land Survey Authority (MLSA) Mean Sea Level (MSL) datum (Male’ PD).

Corrections were applied in real time using the Water Solution Continuous Operating Reference Station (WS-CORS) network situated in Male City, which provided differential corrections to the rover units. Raw CSV point data (comprising Northing, Easting, and Elevation/Depth) were processed in Autodesk Civil 3D to generate a seamless point cloud and Digital Terrain Model (DTM).

Bathymetric measurement uncertainty

Quantitative error bounds are calculated based on the equipment specifications and environmental constraints (described above in the field report).

• Vertical Uncertainty (σ_z):

The total vertical uncertainty (TVU) is a summation of GNSS vertical error (typically ±0.02 m for RTK), SBES depth accuracy (1% of depth), and dynamic vessel motion (heave/roll/pitch).

○ Topographic (Walkaround): Estimated σ_z = ± 0.03 m.

○ Bathymetric (Boat-based): Estimated σ_z = ± 0.04 m.

• Horizontal Uncertainty (σ_xy):

Using the WS-CORS network, horizontal positioning accuracy is estimated at ±0.02 m + 0.5 ppm. However, pole tilt on the moving vessel introduces a lever-arm error.

○ Estimated σ_xy: ± 0.10 m.

• Interpolation and Volumetric Error:

The 20 m survey line spacing introduces interpolation uncertainty when generating the DTM, particularly in areas of high rugosity near the sand bladders. Volumetric uncertainty (V_err) for sediment accretion analysis is calculated as:

V_{e r r} = A \times \sqrt{σ_{z 1}^{2} + σ_{z 2}^{2}}

Where A is the surface area of the change, and σ_z1 and σ_z2 represent the vertical uncertainties of the 2023 and 2025 surfaces, respectively. Consequently, reported volume changes below a vertical threshold of ±0.15 m should be interpreted as within the margin of error of the single-beam methodology used.

Hydrodynamic instrumentation

Current velocity and direction were quantified using a shallow-water Lowell Instruments TCM-4 Tilt Current Meter. The instrument operates on the drag-tilt principle, in which flow velocity is derived from the inclination of a buoyant, tethered logger relative to the vertical axis. To resolve the mean current velocity in the presence of wave orbital motions, the instrument was configured for high-frequency burst sampling. Data was recorded over a continuous 48-h period to capture multiple tidal cycles.

The sampling protocol used a 2-min burst interval, with each burst comprising a 45-s acquisition window sampled at 16 Hz. This high sampling rate (720 samples per burst) enables statistical removal of high-frequency wave noise, isolating the steady-state current vector.

Hydrodynamic measurement uncertainty

The measurement uncertainty for drag-tilt meters is non-linear, as sensitivity decreases at the extremes of the tilt range. Based on the manufacturer’s calibration for the TCM-4, the velocity error bound is a function of the measured speed magnitude:

U_{v e l} = \pm (2.0 c m \cdot s^{- 1} + 0.03 \cdot V_{m a g})

Directional uncertainty is estimated at ± 5.0° for tilts exceeding 10°. It should be noted that during slack tides where velocity drops below the instrument’s detection threshold (typically < 2 cm/s), directional data is considered indeterminate due to the dominance of buoyancy over drag forces.

Results

Bilateral flow attenuation and directed sediment transport

In the first season of the east-to-west dominant current, the ring structure performed as expected based on findings from the unidirectional structure, capturing sediment from the east to west transport, as evidenced by rapidly forming bars.

Within the ring, sediment accretion transformed the bathymetry within 3 months, and by 12 months it was inches below low water at its highest points.

Unlike prior unidirectional structures, the ring structure continued to drive sediment accretion through the subsequent season, in the prevailing west-to-east current. The following bathymetric surveys captured the seasonal shifts and the continuous sediment accretion within the target area.

Discussion and next steps

Sediment structure interaction

Unidirectional

The initial unidirectional structure continues to perform effectively when aligned with prevailing currents. In the lab, we see applications for optimized unidirectional systems in fluvial flows and other conditions characterized by a single prevailing flow vector. This suggests the potential to compute using other unidirectional flows, such as lava flows, mudslides, and avalanches.

Bidirectional

For the Maldives and other sites dominated by oscillating seasonal or tidal currents, a next step is to generalize the ring structure approach for a variety of sites. The existing configuration performs most effectively when oriented within 15 degrees of the prevailing current direction and is therefore suited to locations where a prevailing bidirectional current varies in direction by 30° or less. The CFD pipeline can be adapted to explore a broader solution space of possible ramp configurations and to identify those suited to other bidirectional flow conditions, including variations in flow direction and asymmetries in time-averaged flow velocity. Deployment across a variety of sites will ultimately require an adaptable parameterization that responds to these varying conditions.

Isotropic

Moving beyond bidirectional systems, we have drawn upon research studying vegetated flows^47,48 to develop isotropic structures that can modulate sediment transport in flows operating in any direction.

This investigation aims to develop coastal construction systems that improve access and increase shoreline stability without disrupting critical sediment transport pathways. This is an ongoing effort to design architectures that mediate, rather than obstruct, contributing to the shoreline’s long-term adaptive capacity.

Growing islands

The project stakeholders in the Maldives are encouraged by the first-year results from the Real World Lab. The Bidirectional structure captures net sediment each season in patterns that align with numerical simulations. We are working with the Maldivian government to identify future test sites for this approach.

We view the installation of these structures as triggers, intended to cultivate the emergence of stable, persistent islands through continuous environmental feedback. In the lab and on-site, we are also testing systems to support subsequent steps following an island’s emergence. This includes floating systems that position native vegetation to consolidate sediment as it accumulates and reaches the water’s surface.

Additionally, we are developing intervention systems to maintain fully consolidated and inhabited islands. These systems allow residents to safely and incrementally manage sediment transport along their shorelines, extending architecture’s mediating function to the management of the coastal edge.

Ecological scope and monitoring

While the primary focus of this study is the morphological evolution of the sediment structure, we acknowledge that the intervention operates within a sensitive coral reef ecosystem. The results presented here are limited to physical sediment transport and hydrodynamic attenuation. Ecological monitoring is currently underway to track coral health in the immediate vicinity of the accretion zones and to measure biodiversity recruitment on the engineered substrate, to validate the project’s “nature-based” designation, and ensure that the localized sediment accretion does not negatively impact adjacent reef health.

Conclusion

Architecture as instrument of inquiry

This research tests the “Real-World Lab” approach, extending the computational design process into the world, where local hydrodynamic and biological processes contribute their computational capacity to formfinding through morphodynamic evolution.

This enables the field installation to directly test dynamics too complex and non-linear to simulate numerically. The installation thereby provides a lab that can continuously improve our understanding of the dynamics at the interface between systems. This approach engages architecture in the collection of scientific evidence and expands authorship over the computational process, in alignment with calls by Pablo Lorenzo-Eiroa⁴⁹ and others.

This computational extension provides the potential to overcome the limits of digital computation in deep uncertainty. Enlisting the collaborative information-processing capacity of both local environmental systems and human stakeholders creates the conditions for co-assembly, building resilience through the active mediation of multi-system interaction.

Footnotes

Acknowledgments

This project was part of a long-term research collaboration between Invena PVT LTD , MIT’s Self-Assembly Lab , Sanken Overseas and Momentum Resorts PVT LTD . All bathymetric surveys were conducted by and the associated data provided by Charrette Studio , Male, Maldives. The research was additionally supported by MIT Architecture, MAD, MIT SA+P.

ORCID iD

Walter Zesk

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by USAID, grant number CAP-FAA-010-Invena.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Ulber

Mahall

. Adaptive architecture as mediator between humans and Earth. AGATHÓN Int J Archit Art Des 2019; 6: 94–103.

Walker

Holling

Carpenter

, et al. Resilience, adaptability and transformability in social-ecological systems. Ecol Soc 2004; 9(2): art5. https://doi.org/10.5751/es-00650-090205

Deser

. Certain uncertainty: the role of internal climate variability in projections of regional climate change and risk management. Earths Future 2020; 8(12): e2020EF001854. https://doi.org/10.1029/2020ef001854

The University of British Columbia School of Architecture and Landscape Architecture, McCarty J and Rysanek A . The university of British Columbia school of architecture and landscape architecture. The New Normals: Architecture Under Climate Change Uncertainty. 2020 AIAACSA Intersect Res Conf CARBON, 2020, pp. 213–222.

Hendry

Pretis

. All change! the implications of non-stationarity for empirical modelling, forecasting and policy. Econom Model Forecast EJournal 2016, Available from: https://www.semanticscholar.org/paper/All-Change!-The-Implications-of-Non-Stationarity-Hendry-Pretis/4be10260078699aa22a3997983daa42b9e225f7a?utm_source=consensus

Milly

PCD

Betancourt

Falkenmark

, et al.

Stationarity is dead: whither water management?

Science 2008; 319(5863): 573–574. https://doi.org/10.1126/science.1151915

Piggott-McKellar

Nunn

McNamara

, et al. Dam(n) seawalls: a case of climate change maladaptation in Fiji. In: Leal Filho

, editor. Managing climate change adaptation in the Pacific region. Cham (Switzerland): Springer; 2020. p. 69-84. Available from: https://doi.org/10.1007/978-3-030-40552-6_4

Nunn

Klöck

Duvat

. Seawalls as maladaptations along island coasts. Ocean Coast Manag 2021; 205: 105554. https://doi.org/10.1016/j.ocecoaman.2021.105554

Duvat

VKE

Magnan

. Rapid human-driven undermining of atoll island capacity to adjust to ocean climate-related pressures. Sci Rep 2019; 9(1): 15129. https://doi.org/10.1038/s41598-019-51468-3

10.

Chester

Allenby

. Toward adaptive infrastructure: flexibility and agility in a non-stationarity age. Sustain Resil Infrastruct 2019; 4(1): 1–19, Available from. https://www.tandfonline.com/doi/full/10-1080/23789689.2017.1416846

11.

Directorate-General for Research and Innovation(European Commission) . Evaluating the impact of nature-based solutions: a handbook for practitionersi. Publications Office of the European Union, 2021. Available from. https://data.europa.eu/doi/10-2777/244577

12.

Chausson

Turner

Seddon

, et al. Mapping the effectiveness of nature-based solutions for climate change adaptation. Glob Change Biol 2020; 26(11): 6134–6155. https://doi.org/10.1111/gcb.15310

13.

Geukes

Van Bodegom

Van Oudenhoven

APE

. Setting the stage for decision-making on nature-based solutions for coastal climate adaptation. Ocean Coast Manag 2024; 247: 106916. https://doi.org/10.1016/j.ocecoaman.2023.106916

14.

Ruckelshaus

Reguero

Arkema

, et al. Harnessing new data technologies for nature-based solutions in assessing and managing risk in coastal zones. Int J Disaster Risk Reduct 2020; 51: 101795. https://doi.org/10.1016/j.ijdrr.2020.101795

15.

Saleh

Weinstein

. The role of nature-based infrastructure (NBI) in coastal resiliency planning: a literature review. J Environ Manag 2016; 183(Pt 3): 1088–1098. https://doi.org/10.1016/j.jenvman.2016.09.077

16.

Greenlees

Cornelius

. The promise of panarchy in managed retreat: converging psychological perspectives and complex adaptive systems theory. J Environ Stud Sci. 2021 Sept 2021; 11(3): 503–510. https://doi.org/10.1007/s13412-021-00686-1

17.

Yang

Ligmann-Zielinska

Dou

, et al. Complex effects of telecouplings on forest dynamics: an agent-based modeling approach. Earth Interact 2022; 26(1): 15–27. https://doi.org/10.1175/ei-d-20-0029.1

18.

Tompkins

Adger

. Does adaptive management of natural resources enhance resilience to climate change? Ecol Soc 2004; 9(2): 90210. https://doi.org/10.5751/es-00667-090210. Available from: https://www.ecologyandsociety.org/vol9/iss2/art10/

19.

Singhvi

Luijendijk

van Oudenhoven

APE

. The grey – green spectrum: a review of coastal protection interventions. J Environ Manag 2022; 311: 114824. https://doi.org/10.1016/j.jenvman.2022.114824

20.

Kench

Brander

. Response of reef island shorelines to seasonal climate oscillations: south maalhosmadulu atoll, Maldives. J Geophys Res Earth Surf 2006; 111(F1): 2005JF000323, Available from. https://doi.org/10.1029/2005JF000323

21.

Kempa

Karrasch

Schlurmann

, et al. Design and insights gained in a real-world laboratory for the implementation of new coastal protection strategies. Sustainability 2023; 15(5): 4623. https://doi.org/10.3390/su15054623

22.

Holling

. Resilience and stability of ecological systems. International Institute for Applied Systems Analysis, 1973, p. 23. (Research report (International Institute for Applied Systems Analysis)).

23.

Refulio-Coronado

Lacasse

Dalton

, et al. Coastal and marine socio-ecological systems: a systematic review of the literature. Front Mar Sci 2021; 8: 648006. https://doi.org/10.3389/fmars.2021.648006

24.

. The sand motor: a nature-based response to climate change. Findings and reflections of the interdisciplinary research program Naturecoast 2019.

25.

Palinkas

Orton

Hummel

, et al. Innovations in coastline management with natural and nature-based features (NNBF): lessons learned from three case studies. Front Built Environ 2022; 8: 814180. https://doi.org/10.3389/fbuil.2022.814180

26.

Chase-Lubitz

The Maldives is racing to create new land. Why are so many people concerned?

Nature. 2024; 628(8009): 704–8.

27.

Magnan

Duvat

VKE

. Towards adaptation pathways for atoll islands. Insights from the Maldives. Reg Environ Change 2020; 20(4): 119. https://doi.org/10.1007/s10113-020-01691-w

28.

Arnall

. Where land meets sea: islands, erosion and the thing-power of hard coastal protection structures. Environ Plan E Nat Space 2023; 6(1): 69–86. https://doi.org/10.1177/25148486221101461

29.

Malvarez

Ferreira

Navas

, et al. Storm impacts on a coupled human-natural coastal system: resilience of developed coasts. Sci Total Environ 2021; 768: 144987. https://doi.org/10.1016/j.scitotenv.2021.144987

30.

Markolf

Chester

Eisenberg

, et al. Interdependent infrastructure as linked social, ecological, and technological systems (SETSs) to address Lock-in and enhance resilience. Earths Future 2018; 6(12): 1638–1659. https://doi.org/10.1029/2018ef000926

31.

Allen

Angeler

Garmestani

, et al. Panarchy: theory and application. Ecosystems 2014; 17(4): 578–589. https://doi.org/10.1007/s10021-013-9744-2

32.

Masselink

Lazarus

, et al.

Defining coastal resilience. Water

2019; 11(12): 2587. Available from. https://www.mdpi.com/2073-4441/11/12/2587

33.

Gunderson

Holling

(eds). Panarchy: understanding transformations in human and natural systems: Island Press, 2002, p. 507.

34.

Cantrell

Holzman

. Responsive landscapes: strategies for responsive technologies in landscape architecture. London: Routledge; 2015. Available from: https://www.routledge.com/Responsive-Landscapes-Strategies-for-Responsive-Technologies-in-Landscape-Architecture/Cantrell-Holzman/p/book/9781138796652

35.

M’closkey

Vandersys

. Topography as the groundwork for landscape design—interview with karen M’Closkey and Keith VanDerSys. Landsc Archit Front 2023; 11(4): 46–59. https://doi.org/10.15302/j-laf-1-030040

36.

Pähtz

Clark

Valyrakis

, et al. The physics of sediment transport initiation, cessation, and entrainment across aeolian and fluvial environments. Rev Geophys 2020; 58(1): e2019RG000679. https://doi.org/10.1029/2019rg000679

37.

Shields

. Application of similarity principles and turbulence research to bed-load movement: California Institute of Technology, 1936. (Translate from German). - References - Scientific Research Publishing. Available from: https://www.scirp.org/reference/referencespapers?referenceid=1363329

38.

Deal

Venditti

Benavides

, et al. Grain shape effects in bed load sediment transport. Nature 2023; 613(7943): 298–302. https://doi.org/10.1038/s41586-022-05564-6

39.

Pender

Shvidchenko

Chegini

. Supplementary data confirming the relationship between critical shields stress, grain size and bed slope. Earth Surf Process Landf 2007; 32(11): 1605–1610. https://doi.org/10.1002/esp.1588

40.

Wengrove

Foster

Lippmann

, et al. Observations of bedform migration and bedload sediment transport in combined wave-current flows. J Geophys Res Oceans 2019; 124(7): 4572–4590. https://doi.org/10.1029/2018jc014555

41.

Babiński

. The relationship between suspended and bed load transport in river channels. IAHS-AISH Publ, 2005. Available from: https://www.semanticscholar.org/paper/The-relationship-between-suspended-and-bed-load-in-Babi%C5%84ski/d524d9d2b3c1223063b8251d1fc3e7cef14c603a

42.

Paul

. Relative quantities of suspension versus bed-load transport on beaches. SEPM J Sediment Res 1978; 48, Available from: https://pubs.geoscienceworld.org/jsedres/article/48/3/921-932/97200

43.

Zampol

Inman

. In: Seymour

(ed). Discrete Measurements of Suspended Sediment: Springer US, 1989, pp. 259–272. Available from: https://doi.org/10.1007/978-1-4899-2531-2_24

44.

Hopkins

Elgar

Raubenheimer

. Flow separation effects on shoreline sediment transport. Coast Eng 2017; 125: 23–27. https://doi.org/10.1016/j.coastaleng.2017.04.007

45.

Flow Science, Inc . FLOW-3D HYDRO, version 2023R1. https://www.flow3d.com

46.

(PDF) . Development of turbulence models for shear flows by a double expansion technique. ResearchGate, Available from: https://www.researchgate.net/publication/23602939_Development_of_Turbulence_Models_for_Shear_Flows_by_a_Double_Expansion_technique

47.

Nepf

. Drag, turbulence, and diffusion in flow through emergent vegetation. Water Resour Res 1999; 35(2): 479–489. https://doi.org/10.1029/1998wr900069

48.

Nepf

. Flow and transport in regions with aquatic vegetation. Annu Rev Fluid Mech 2012; 44(1): 123–142. https://doi.org/10.1146/annurev-fluid-120710-101048

49.

Lorenzo-Eiroa

Sprecher

. Architecture in formation: on the nature of information in digital architecture. Routledge, 2013. Available from: https://doi.org/10.4324/9781315890128