Computational modeling for climate change: Simulating and visualizing a resilient landscape architecture design approach

Current landscape architecture digital tools

As climate change becomes an increasingly common aspect of planning and design of urban spaces, landscape architects face challenges of great ecological complexity. Landscape architecture software workflows, already lagging due to slow adoption of Building Information Modeling (BIM), are now further pressured by the demand to use sophisticated software workflows in addressing complex climate change scenarios. ¹ These pressures require adaptation of traditional design processes in order to create aspirational designs which can continually cool ever-hotter urban areas as well as tolerate extreme weather events such as coastal storms. ² This raises the question of what digital techniques landscape architects must assemble in order to accomplish these tasks. Without a single piece of software that can address the full complexity of designing for the landscape, we propose the assembly of several existing digital tools to import scientific climate and environmental data and create simulated three-dimensional (3D) environments for testing design solutions.

Current software workflows for visualizing impacts of climate change do exist, and the urgency of climate change requires that we attempt to model these unknowns as best we can. Landscape architects still use computer-aided design (CAD) software on a daily basis, usually AutoCAD ³ for design development and documentation. They also use geographic information systems (GIS) software to document existing physical and demographic conditions. The common use of GIS software in landscape architecture has broadened the CAD workflow to include geospatial aspects, arguably the strongest “intelligent” tools used by landscape architects, and their interoperability with CAD systems is strong and established. ⁴ Furthermore, landscape architects also make use of parametric design tools for creating responsive algorithmic design, for instance, the Rhino ⁵ tool Grasshopper, which is effective at inputting dynamic data and rapidly outputting different spatial forms. For instance, Grasshopper can be used to perform daylighting analysis, commonly used for interior architectural temperature analysis, but also used for exterior conditions. ⁶ Grasshopper can also be used to generate morphology influenced by repetitive patterning, such as the creation of crystalline and glass-like shapes which respond to numerous parametric inputs, a tactic which we will describe later in this article as a way of creating cooling channels of urban air flow. ⁷

Environmental solar energy simulations are usually created using standardized weather files which are based on current data and do not take future climate predictions into account. However, climate-change-adapted weather data simulation is possible and has been attempted for several years, with many established methods. One approach is to use available end-user software tools, such as CCWorldWeatherGen ⁸ and WeatherShift™, ⁹ to generate future projection weather data that can be used for executing building and environmental performance simulation. However, these tools do not operate similarly and therefore may produce widely varying results. In addition, in the case of WeatherShift™, the software only modifies the most significant meteorological parameters, an aspect that is of major importance in situations when less significant meteorological variables need to be included.^10,11 Another approach is to “morph” existing EnergyPlus™ ¹² data with UK Met Office Hadley Center general circulation model (GCM) predictions for a “medium–high” emissions scenario (A2). This approach makes use of the existing CCWorldWeatherGen software, placing certain limits on it to eliminate unrealistic results. Although there are some small issues with this workflow, most notably the tendency to underestimate the impacts of climate change, it is considered to be a practical approach to deriving weather files suitable for climate-change impact assessment in the built environment.^10,11 A third approach is to synthesize weather data sets out of several climate scenarios, based on 1-year weather data which can be generalized for 30-year periods. This approach produces scientifically valid energy performance assessments while keeping the number of simulations much lower than other methods. ¹³

Some computational design researchers have begun to address the area of information-driven landscape modeling, developing algorithms that emulate and visualize the behavior of natural systems. The Nature of Code ¹⁴ provides a framework for software simulation of natural forces in spatial environments, including gravity, friction, and velocity. Testing the Waters, ¹⁵ a project by Peg Office of Landscape Architecture, explores wetland suitability strategies that are directly guided by energy and elevation parameters, modeled using GIS, computational flow dynamics, and parametric software. Work in digitally simulating the flow of rivers has produced valuable ideas regarding the use of robotics to manage sedimentation. ¹⁶ Hydrology researchers have long been using computational models to visualize potential loss of coastal cliffs using GIS software, using a virtual reality workflow to communicate these potentials to lay audiences. One example is the Soft Cliff and Platform Erosion (SCAPE) model, ¹⁷ which uses GIS and a computational algorithm to visualize future cliff erosion scenarios based on erosion, wind, and sea level rise. The development of Ladybug Tools ¹⁸ for Grasshopper has allowed designers to optimize their designs using true environmental data which can be seamlessly input into the 3D model to generate more realistic feedback. ¹⁹ The development of these open-source tools has advanced to the point where integration of computational fluid dynamics (CFD) now permits the modeling of air flows within the Grasshopper environment. This is a particularly important development for landscape architects, who rely on accurate modeling of exterior environmental conditions. ²⁰

Cast study 1: simulating erosion, sea level rise, and storm surge

The main goal of this workflow was to computationally simulate and visualize future conditions of a site undergoing erosion caused by sea level rise, create an animated model of a major coastal storm within these future conditions, and develop a responsive landscape architecture design proposal. In order to meet this goal, we needed to create a digital model that more accurately reflected the interrelated impacts which drive erosion: waves, tides, and changes in sea level rise. ¹⁷ One impediment was that the existing tools used to model and predict sea level rise are not dynamic ecological models, and as such are not informed by the adaptive responsiveness of coastal ecologies, ecologies which can often reform and adapt to encroaching water. ²⁷ We aimed to overcome this obstacle using a range of geospatial, image editing, 3D modeling, and animation tools, which enabled us to model the impacts of sea level rise, storm inundation, and erosion as a dynamic, evolving set of relationships.

Out of necessity, we made intentional assumptions. Although it is known that erosion rates change frequently, we were unable to predict how these rates would change in the future. Because of these unknowns, we opted to used fixed values, setting a low erosion rate of 0.2 m/year and a high erosion rate of 0.5 m/year. One important note is that these rates are greater than the average recorded rate in that area since 1939, ²⁸ but are in sync with the rates of sea level rise used in similar digital simulations of coastal erosion in other coastal erosion studies. ²⁹ A second assumption we made was that the case study site was composed of a single site material, eliminating distinctions between hardscape and permeable surfaces. This was necessary to establish areas of control where we could be certain no material differences were contributing to irregularity in the simulations. The importance of modeling differentiated in situ site materials, such as variations in rock strength, permeability, and drag, will be an important component to include in future development of this workflow.

We began the erosion process with a digital elevation model (DEM) from the United States Geological Survey’s (USGS) National Map Viewer. ³⁰ We imported the DEM into in ArcMap GIS software. Once the DEM had been acquired, we needed to computationally erode the coastline at these set rates. We discovered that although the DEM could easily be exported to a 3D modeling software program, the 3D modeling tools did not allow us to methodically reshape the coastline with any geospatial accuracy. Resorting to the accuracy and reliability of two-dimensional (2D) image editing, we opted to work directly with the DEM image file and erode the coastline in the raster image that was used to generate the 3D model. To accomplish this 2D erosion, we used the Geographic Imager plugin, which allowed the spatial properties of the image to be automatically updated and maintained during raster image editing. This allowed us to clone stamp the coastline pixel data further inland at geospatially accurate distances between the sample and paint locations (Figure 3).

OPEN IN VIEWER

Figure 3.

Digital elevation model showing current and eroded conditions.

After creating several modified DEMs representing various erosion scenarios, we exported the DEMs from Geographic Imager and back into ArcMap, where they appeared as a series of layers on top of the original DEM. The modified DEMs were exported as American Standard Code for Information Interchange (ASCII) point data and imported into Rhino as a terrain mesh. We created a flat plane representing a simplified measurement of sea level height above current mean tide, intersecting the planes with the eroded terrain meshes. Sea level heights are projected to increase between 0.8 m and 1.8 m by 2100. We have used these approximations to create three scenarios, with 2 ft of sea level rise being the most conservative risk scenario, 4 ft of sea level rise being an intermediate risk scenario, and 6 ft of sea level rise being the most extreme risk scenario. ³¹ These visualizations show that ocean inundation combined with shoreline erosion could lead to the breaching of coastal barrier islands and resulting saltwater inundation into freshwater ponds (Figure 4).

OPEN IN VIEWER

Figure 4.

Visualizations showing sea level rise rates coupled with increasing erosion.

Static imagery has some limitations when used to represent dynamic landscape elements. This is especially true when simulating dramatic natural events such as heavy wind and water. To create a more realistic model, our next task was to create an animated simulation of coastal waves. We first identified the potential maximum height of a storm surge in this specific area of the Rhode Island coastline, referencing the Sea, Lake, and Overland Surges from Hurricanes (SLOSH) ³² model from NOAA’s Hurricane Modeling System to provide storm surge ranges for the site’s general vicinity. SLOSH is a computerized model which estimates and visually displays storm surge heights for historical, hypothetical, or predicted hurricanes. These estimates make use of data on atmospheric pressure, size, forward speed, and track data, which are used to create a model of the wind field which generates the storm surge. The SLOSH model is especially useful because it responds to the physical features of a delineated area such as shoreline and inland hydrology, depth, and infrastructure. We used a SLOSH modeling approach called the probabilistic approach, which incorporates statistics of past forecast performances to generate an ensemble of SLOSH runs. ³³

Once we had identified the maximum possible storm surge height, we turned our attention to modeling the storm surge water body in motion. Several software tools exist which can animate fluid particles in 3D space. In evaluating the available software, two options emerged, which were useful because they could model fluid particles in 3D space with a high degree of technical accuracy and visual legibility. The first option, RealFlow, ³⁴ offers rigid and soft body dynamics solvers, making it ideal for large-scale simulations such as floods or oceans with breaking waves, as well as small-to-medium ocean surfaces. ³⁵ The second option, Bifrost, a tool within Maya, is a procedural framework that uses a fluid implicit particle solver to create simulated liquid effects. Bifrost also includes an Ocean Simulation System, which can create realistic ocean surfaces with waves, ripples, and wakes. Ultimately, we selected Bifrost because of its seamless integration within the Maya 3D modeling environment, allowing the smoothest exchange of 3D models with other CAD software. Bifrost’s credibility as a hydrological simulation tool comes from its usage of the Navier–Stokes solver to generate fluid particle motion, written here in pressure–velocity variables

∇ … u = 0 ( ∂ ∂ t + u . ∇ − ν ∇ 2 ) u = g − 1 ρ ∇ p + f

(1)

where ∇ = ( ∂ / ∂ x , ∂ / ∂ y , ∂ / ∂ z ) is the gradient operator in 3D, ( ∂ / ∂ t + u . ∇ − v ∇ 2 ) is the frequently recurring convection–diffusion operator, and f is an external force per unit mass.

The Navier–Stokes equations are commonly used to describe the motion of viscous fluid substances, and their usage for animation and rendering of complex water surfaces allows preservation of much of the realistic behavior of water while also allowing a sufficient degree of control necessary for animation and rendering environments. ³⁶

The first step in modeling the storm surge water body was to import the existing current-day terrain model from Rhino. Because of the computational power required to produce the animations, we divided the site into smaller sections and focused on a single topographic slice. Using a Bifrost simulation at the average height of 11 ft above sea level—that of a category 1 storm surge—we intersected this moving liquid plane with the eroded terrain mesh. To generate further turbulence, we added a cross wave and programmed in periodic fluctuations in the height of the waves. This further amplified the wave action and the distance that the storm surge could travel ashore.

After running initial animations showing a coastal storm surge, our next goal was to apply the wave behavior to an eroded coastline, simulating a future scenario that could occur due to sea level rise. To accomplish this, we imported the edited terrain models and ran animations replacing the existing terrain model with ones showing erosion, factoring in higher wave elevations due to sea level rise. We observed that the tiny remnant of the coastal barrier island is easily breached by storm surge, removing any remaining protection to the heavily populated inland area (Figure 5). Moreover, this complete loss of the barrier island, and the influx of coastal waters, dramatically changes the composition of the adjacent ecosystem.

OPEN IN VIEWER

Figure 5.

Animation stills showing a category 1 hurricane breaching the eroded barrier island.

Using geospatial data showing ecological zones along the study area, we mapped specific ecological areas onto our simulations. These indicate that the advanced level of shoreline retreat could significantly reduce or eliminate areas of inland swamp, as well as brackish and freshwater marshes. With these losses in mind, we set out to develop a design strategy which would preserve these reduced ecologies through gradual inland relocation. This is an uncommon tactic, as most sea level rise design interventions aim to mitigate damage in situ using barriers to keep water out or inlets to allow water to circulate around built form. In contrast, the design strategy emerging from our research prioritizes the full relocation of major ecological areas, while aiming to have a minimal impact on existing built form. Within the study area, our research indicates that the largest inland water body facing total loss is Winnepaug Pond, a swamp and salt marsh fed by Weekapaug Beachway. Our models show that sea level rise and erosion will eventually convert this water body to coastal beach, a change made even more dramatic due to loss of barrier islands (Figure 6). With this in mind, we ask the radical question: could a landscape design possibly recreate this lost ecology in a nearby inland area?

OPEN IN VIEWER

Figure 6.

Current and future conditions showing biodiversity loss.

We identified a suitable area to relocate this water body. Our site selection was based on finding an area that was relatively undisturbed by development, large enough in area to accommodate the water volume of Winnepaug Pond, and lower in elevation than the surrounding areas. These criteria led us to select an area of inland mixed forest and swamp which connects to the Newton Swamp Management Area. Although significant further analysis would be necessary to determine real-life site suitability, as an initial proof of concept, this level of analysis is sufficient to model a spatial design strategy.

We located a series of depressed land areas which could gradually be inundated, with connecting channels between each basin. The first identified basin was Doctor Lewis Pond, which is only a few hundred feet from the closest area of predicted inundation from sea level rise, and is separated from the inundation area by a single road. We re-graded the edge of the pond to connect to the edges of Winnepaug Pond, proposing a bridge over the connective inlet (Figure 7). Winnepaug Pond is predicted to become the eventual shoreline edge, and as such this new grading effectively connects the inland basin of Doctor Lewis Pond to the Atlantic Coast. This re-grading allows the flow of coastal water into a larger inland basin, passing underneath a bridge and impacting two structures and a small portion of a road.

OPEN IN VIEWER

Figure 7.

Existing conditions showing inland basin (left) and proposed re-grading to allow ocean water to flow in (right).

We ran this model through our previous Bifrost particle simulator to visualize the possible force of a coastal storm on this reshaped landscape as it flowed through the newly shaped channel into the inland basin. The resulting animation indicates that although the narrow inlet causes extreme turbulence of the moving water, the water does flow primarily into the basin and does not redirect upward to residential areas. One benefit of this level of turbulence at the inlet is that the slowing of this water will reduce inland erosion. Yet, a critical drawback is that it could likely erode the inlet shoreline—already quite steep due to re-grading—causing harmful levels of erosion. This is a drawback that will need to be explored further, perhaps testing methods of slowing down the water through smaller estuaries or widening the inlet (Figure 8).

OPEN IN VIEWER

Figure 8.

Modeling of a category 1 hurricane flowing into the re-graded inlet.

Case study 2: simulating urban heat island effects

Urban heat islands are metropolitan areas that are significantly warmer than surrounding rural areas due to human activity. The main contributing factors to urban heat islands are the presence of hardscape and concrete facades and ground materials, often comprised of darker colors which absorb and hold sunlight and increase the immediate surface temperature. This absorbed heat is then released in the cooler hours of the evening atmosphere, increasing the overall ambient temperature. The widespread use of air conditioners in densely urbanized areas are also important contributing factors, releasing heat into the atmosphere and contributing to the formation of ozone and smog. Urban temperatures can be up to 5°F–17°F warmer than surrounding suburban and rural areas, amplifying energy demand for building air conditioning, and in turn leading to greater CO₂ emissions into Earth’s atmosphere. ³⁷

With this urgency in mind, we set out to create tools and typologies to assist landscape architects, urban designers, and urban planners in mitigating urban heat island effects and reducing human reliance on artificial cooling systems during the hottest months of the year. In order to understand current conditions of urban heat island effects at a large scale, we first performed a surface temperature analysis of the entire City of Boston, MA. We used Dragonfly (a component of Ladybug Tools) to download recent data and infrared images from the USGS website and import this information to our Rhino model. To visualize, identify, and colorize the infrared image with surface temperature data, we used Ladybug to link the aerial image with aerial Landsat ³⁸ data, where the basic infrared details were interpreted by Ladybug and Dragonfly and represented as a gradient colored by temperature range. The resulting visualization allowed us to evaluate the impact of urban heat island effects throughout the City of Boston and to identify a smaller case study area. A comparison of three neighborhoods shows varying surface temperature impacts informed by building shadow and surface material. We ultimately decided to study the Roxbury neighborhood, which included high surface temperatures as well as a higher building density than some other neighborhoods with similarly high surface temperatures (Figure 9).

OPEN IN VIEWER

Figure 9.

Surface temperature analysis comparison showing three neighborhoods in Boston.

Once we visualized the existing surface temperature, we proceeded with a sun and shade study. This study did not visualize the specific path of shadows cast by buildings into open space, but instead averaged the mean surface temperature which resulted from sun and shade in open areas. Each analysis was done over the course of a single month, geolocating the Rhino model to the correct geographic coordinates and orientation. Using Ladybug to identify the sun path over these set specific periods of time based on EPW data, we were able to access and apply weather data from several monitoring stations. This workflow enabled us to identify the areas of shade and sun exposure that existed across the study area within Roxbury.

These analyses showed us the major areas where surface temperature would be the hottest over the course of summer months. We identified these areas as locations where design intervention could occur to cool the surface temperature. To test design interventions which could most effectively improve these conditions, we chose two common strategies for cooling outdoor air temperature: small-scale improvements to hardscape materials and more widespread improvements to outdoor air flow. We then proceeded to use two digital workflows to test these strategies.

The first workflow of visualizing temperature differences in small-scale hardscape materials relied on computationally analyzing the impact of albedo and surface temperature. Albedo can be defined as the ability of a material to reflect solar energy. To visualize these surface temperatures based on the characteristics of their corresponding surface materials, we used the Grasshopper plugin Honeybee, a component of the Ladybug toolkit. The reliability of these tools are based on their underlying calculation of albedo, which is written here as

T = K s x 1 − ( a l b e d o ) 4 σ 4 + GREENHOUSE EFFECT

(2)

where K_s is solar insolation (“solar constant”) = 1361 Watts per square meter, σ is Stefan–Boltzmann constant = 5.670373 × 10^–8 Watts/m² K⁴, (m = meter, K = Kelvin), and GREENHOUSE EFFECT is 33°C.

This formula enables Honeybee to include a small number of material presets, which are embedded with data for albedo, thermal capacity, conductivity, and heat absorption. The albedo values for some of these materials are as follows:

The next step was to associate ground surface textures with specific material properties, selecting and applying Honeybee presets to Rhino materials. In addition, natural elements such as trees, grass, and soil were geometrically identified to allow the Honeybee algorithm to differentiate hardscape and softscape materials. The resulting visualizations demonstrate the differences in surface temperature of the ground based on the material that was applied, indicating that soil, dry, and semi-dry sand have the highest albedo and therefore the lowest surface temperature in hot conditions (Figure 10).

OPEN IN VIEWER

Figure 10.

Simulation showing various materials applied to the ground plane and their resulting surface temperature.

Computationally modeling urban air flow relied on differentiating between the density, shape, and formation of buildings and the energy and seasonal direction of air flow. To understand the seasonal air flow direction, we used Ladybug to input EnergyPlus™ data to create a wind rose analysis. It is important to note that the simulations were based on current EPW data and therefore do not account for the effects of climate change on the EPW values we used. We chose this approach to avoid introducing speculative data into a new workflow which was still in a development and evaluation stage. The inclusion of well-documented methods of generating predictive EPW-based climate data will be an essential piece of our future workflows.

Once we understood the general wind direction over the course of 12 months, we selected the average hottest month of August to use as a baseline wind flow pattern for testing design strategies. Butterfly, another component of the Ladybug toolkit, relies on OpenFOAM, an open-source CFD software C++ toolbox for the development of customized numerical solvers. The CFD framework is based on the same Navier–Stokes solver as was used to create the storm surges in Bifrost and was a critical part of successfully modeling urban air flow. ²⁰ To create this simulation, we used the Butterfly Grasshopper definition to situate the building geometries within a wind tunnel and then used the parameters within the definition to direct a series of vectors which passed across the model by way of a mesh grid. The visualization indicates the direction and speed of the vectors; as vectors were forced to change direction, slow down, or even stall, the visualization would colorize them as yellow or red. As the vectors were able to continue their path with minimal speed decrease or change in direction, they were colorized as blue. After the first simulation was complete, we adjusted the values within the Grasshopper definition, which include proposed region, wind speeds, and test period windows of time, and re-ran the simulations. The definition also provided an option to visualize air pressure, which is significant because higher air pressures lead to decreases in outdoor human comfort. The visualization shows significant portions of the study area with high surface temperature and air pressure. In the study area’s existing condition, we observed that air flow was most negatively impacted where smaller buildings were densely arranged with contrasting orientations, and we observed that air pressure was most negatively impacted where “pinch points” occurred—small areas of open space between multiple large building masses (Figure 11).

OPEN IN VIEWER

Figure 11.

Detail of air flow simulation, showing “pinch points” between large building masses.

Once the base conditions of urban heat island effects were visualized, these same processes were used to redesign the study neighborhood as an urban design at the master plan scale. Recent research has indicated that the arrangement of a city’s streets and buildings plays a central role in how hot or cool it gets, with findings that show that a traditional urban grid, called a “crystalline” urban form, experiences more extreme heat buildup than an irregular “glass-like” formation. ³⁹ Therefore, we decided to follow the “glass-like” formation in an attempt and redesign the neighborhood with a focus on building and street layout, coupled with simple material adjustments that would raise the overall albedo. The underlying principles of this redesign involved arranging the buildings to be as parallel to the summer wind direction as possible, with paths and roads crossing through the urban grid to channel this summer wind through open spaces.

A major advantage of using Grasshopper has been the integration of parametric tools into the visual simulation engine provided by Ladybug Tools, which played a critical role in our design decision-making. This allowed us to create variations of one environmental factor such as wind speed and visualize its impact on other environmental factors such as wind direction and air pressure (Figure 12).

OPEN IN VIEWER

Figure 12.

Parametric simulation of variable wind speed at 3, 6, and 9 m/s, combined with design variations and visualizing resulting effect on wind direction and air pressure.

Inserting different building and landscape formations and parametrically changing the environmental values allowed us to take advantage of the parametric capabilities of Grasshopper to confidently arrive at the most impactful design solution. Comparing the original visual simulations with simulations using the new “glass-like” urban design shows substantial improvement in the areas of surface temperature, air pressure, and air flow (Figure 13).

OPEN IN VIEWER

Figure 13.

Simulations of existing conditions (left) and proposed building arrangement (right) with new hardscape materials.