Enable JavaScript to view this interactive presentation.
Download the offline presentation
Open the HTML in a current desktop browser. Select Full screen. Right Arrow, Page Down, Space, or Enter advance; Left Arrow and Page Up go back. Home and End jump to the opening and closing. F toggles fullscreen, N opens speaker notes, R replays the current scene, P pauses or resumes motion, including when the computer initially requests reduced motion; B blacks out the screen. Every scene advances with Next; no on-slide choices are required. There is no talk timer. Fonts, map data, and models are embedded, so the downloaded HTML works without internet.
The talk targets 4 minutes 40 seconds, leaving twenty seconds within a five-minute slot. The 27 scenes are short visual builds in one story, not 28 separate topics. The script moves continuously through the visual examples, with no audience participation pauses. Notes contain the spoken script and section timings. Let the day cycle and the three individual ABM examples complete. The heat walker starts after a 0.6-second introduction. The first ABM scene introduces three labeled decisions, then expands to 180 people. The PDF and PowerPoint in the submission package are static fallbacks; use the HTML for animation.
One familiar walk introduces a broader research approach. A shared time budget produces unequal reach. Personal preferences change the route. A three-dimensional environment changes exposure. Individual decisions become an agent-based population, and a proposed crossing and shade become a planning experiment. The cost slide asks whether the outcome justifies the investment. The closing presents broader areas of work: the built environment, environmental exposure, and simulation and planning tools. It returns to the idea that an ordinary walk can become a research question.
The street and building geometry comes from OpenStreetMap; route corridors and the selected junction come from Dakota’s supplied sketches. Satellite vegetation informs a relative heat proxy. Walking speeds, waits, preferences, emissions, weather, budget, and demand are illustrative assumptions. These visuals demonstrate a research workflow. They are not a calibrated navigation service, local air-quality forecast, temperature measurement, or completed investment appraisal.
The three isochrones preserve the same network, origin, and scale as V6–V8. Assumed paces are 1.35 m/s, 0.80 m/s, and 0.95 m/s. The selected junction adds an assumed 60-second delay; steps have a 1.5 time factor. The fifteen-minute boundary uses interpolated Dijkstra travel times with an 18 m cartographic buffer. Its reported area is the area of that buffered reachable network (0.845, 0.234, and 0.369 km²), not a claim that all enclosed parcels are accessible.
General, Access, Interest, and Beauty originate in Dakota’s path_examples.geojson. The supplied lines identify corridors, not a surveyed navigation network. OpenStreetMap has missing connections; local alignment within 12 m and explicitly assumed connections preserve the intended corridors. Geometry, entrances, opening hours, gates, gradients, steps, and passage availability are not fully verified. “Regular” is the reference route in the user’s sketch, not an observed population average or a claim that every computer would select it.
path_examples.geojson
Access uses the familiar informal crossing at the point where the sketch intersects Songdogwahak-ro 84beon-gil (local coordinates approximately 148.1, −172.9 m). This is distinct from the frustrating regular junction at 126.665656° E, 37.378600° N. Comfort first follows the regular route through that junction and then minimizes a shelter-weighted travel cost on the allowed network. Covered passages and simplified 15:00 building shadows supply the shelter preference. Interest also approaches through the regular junction before following the shopping detours; its interior connection is schematic. Beauty follows the supplied park-side sketch.
All route times use 1.2 m/s and the selected junction delay, rounded to whole minutes on screen. Shopping and café stops are not included. The exact embedded times preserve decimal minutes. The same complete route geometry drives the distance, time, colored line, and walker endpoint. White dashes on the Comfort route indicate covered segments; no route is truncated at fifteen minutes.
The NDVI texture is removed. Green space now uses mapped park and water polygons with mapped walking paths. No raster greenness is painted across the city. Missing park detail remains a limitation of the source map.
Solar azimuth and elevation use NOAA’s fractional-year equations for 21 June 2026 at 37.379° N, 126.663° E, UTC+9. The displayed sky arc is diagrammatic; shadow directions and lengths use calculated coordinates. Footprints come from OSM. Heights use mapped height, building levels × 3.2 m, or type-based defaults. The shadow model simplifies buildings into extruded footprints and does not resolve individual tree canopy.
The morning-to-afternoon route switch is an explicit decision scenario: a regular morning journey is replaced by a route computed with a greater preference for shelter. It is not an empirically estimated behavioral response, nor proof that solar position alone causes the choice. The heat scene now uses the cached Sentinel-2 L2A observation S2C_52SBG_20250710_0_L2A (10 July 2025). Cloud, shadow, and snow masking precede 10 m NDVI calculation; a 15 m smoothing kernel reduces sensor-scale speckle. Vegetation inferred from NDVI, mapped paved roads and water, and computed 15:00 building shade supply a relative outdoor heat-exposure index. The morning-to-evening overlay recomputes shade at half-hour intervals and interpolates between 27 fields from 06:00 to 19:00. A hypothetical warming cycle peaks at 14:00. The proxy is 0.23 + 0.51 × day − 0.32 × vegetation × (0.35 + 0.65 × day) − 0.28 × shade × sin(solar elevation) + 0.10 × road proximity × day − 0.18 × water, clipped to 0–1. Day = max(0, sin(π × (hour − 6)/16)). These numerical weights are assumptions. A single color scale is used throughout. The 5 m display grid does not add satellite detail beyond the original 10 m observation. NDVI is an input only: parks retain their clean map styling. Colors have no temperature units and are not validated PET, UTCI, or health-risk estimates. Unshaded walking excludes stationary waiting.
The embedded wind field is precomputed with a D2Q9 BGK lattice-Boltzmann CFD solver on a 640 × 620 grid of 5 m cells. Building footprints above 2 m act as solid, no-slip obstacles with halfway bounce-back at their boundaries. The scenario assumes a 2 m/s wind toward 16° north of east. The domain boundary prescribes a uniform far-field wind and density, with a 120 m relaxation layer to reduce reflected numerical pressure waves. This outer layer lies beyond the central walking area. A constant effective viscosity represents unresolved mixing. The last 2,000 of 16,000 flow steps supply a mean velocity field.
This is a simplified 2D horizontal CFD slice displayed within the 3D city. It can illustrate deflection, sheltered wakes and street-channel flow. It is not validated 3D RANS/LES or a local forecast, and does not resolve airflow over roofs, vertical recirculation, atmospheric stability or thermal buoyancy. The grid is fine enough to make building interactions visible, but no grid-convergence or meteorological validation study has been performed.
One passive pollutant is emitted along mapped motor roads, with relative source strengths based on road class and mapped or assumed lane counts. A modest queue multiplier applies only near mapped OSM traffic signals. User-marked presentation junctions no longer manufacture hotspots. Traffic volumes, fleet composition, emission factors and concentrations are unknown; all strengths are scenario assumptions.
The first pollution scene is a still-air diffusion-and-loss baseline with the same sources and impermeable building boundaries. Wind is then added. A conservative MUSCL/minmod finite-volume scheme with SSP-RK2 time integration transports pollution around impermeable buildings, with a 3 m²/s diffusion coefficient. Continuous road emissions remain active in every snapshot. An assumed exchange timescale varies from 4 to 30 minutes according to opposing-building height relative to street width, so more enclosed cells clear slowly. This is an uncalibrated closure for unresolved ventilation, not measured canyon residence time. Boundary outflow, source input and removal are included in its mass budget. Nine snapshots show a 30-minute transition on one common color scale. The animation compresses time; the flow tracers are visual markers. It uses no separate industrial or heat particle types. Purple particles make ongoing sources visible; they are visual markers and do not encode concentration. The transport scene holds the still-air baseline briefly, then fades in the same airflow shown on the preceding slide while concentration shading updates through the transport snapshots. It uses no additional pollutant trails; changes in the shading represent modeled redistribution. Concentration colors use the same contrast mapping in the baseline, transport, and route scenes. The field values and route calculations have not changed. The transport field includes source input, outflow, diffusion, and removal; the per-step numerical update residual was below 4 × 10⁻⁸ relative to stored mass. This checks update roundoff, not real-world accuracy.
The horizontal map retains the same solved wind and pollutant transport as the previous presentation. Wind vectors are now stored with 0.000244 m/s precision; the earlier display rounded to 0.03125 m/s and could hide very slow motion. The displayed streamlines integrate the solved field and remain outside solid building footprints. Speed colors use one 0–5.5 m/s scale. The main map’s concentration estimates and ten-route exposure comparison are unchanged.
A pool of 68 candidates is generated from travel-time, exposure-weighted, and intermediate-destination searches. Ten loop-free routes, capped at 38 minutes, are selected to span distinct network corridors. The global exposure-optimum on the permitted graph is included. The list tests these ten routes automatically, then highlights the lowest-exposure result. Its values use the regular route’s total modeled exposure as 100. This is not a claim to have evaluated every possible behavioral alternative.
The selected walking scene minimizes the integral of the final modeled relative concentration over travel and waiting time using Dijkstra’s algorithm on the permitted baseline walking graph. It excludes the informal crossing. A small common background term prevents zero-cost edges. The walker retains 1.2 m/s. Where a mapped passage enters a building footprint, the nearest outdoor concentration is assigned: no unsupported indoor filtration benefit is invented. This is the least-exposure path among this graph’s alternatives under one wind and source scenario, not a recommendation based on measured air quality.
The heat chart accumulates modeled minutes in direct sun along the actual route. The air chart accumulates relative exposure, with the regular route’s complete dose defined as 100 units. Colored curves advance with the same route clock as the walker; dashed curves show the regular reference. These are model outputs, not sensor readings.
There are 180 synthetic people, divided equally among time-, shade-, and amenity-oriented preferences. They have different origins, destinations, and walking speeds. A fixed seed makes the experiment repeatable. One third of trips start at campus; others connect neighborhood anchors. Each person chooses a path minimizing their own weighted travel cost. The city is shared, but individual priorities differ. There are 172 distinct original routes and 171 routes after the plan.
The introduction first follows three original journeys, selected from the same people highlighted after the intervention. Their priority labels describe the model’s assumed preferences, not real survey quotations. The remaining agents then appear around these three examples. Route replays compress travel time for visibility and should not be used to compare walking speeds. The baseline scene shows the original journeys. After the proposed crossing and localized shade are introduced, the same people, destinations, and preferences are rerun. Thirty-six people select a different route. This is individual route-choice ABM; people do not interact with each other, crowd, or delay vehicles in this demonstration.
The three highlighted cases are selected examples from the simulation, not estimates for demographic groups: person 9 saves 3.86 minutes; person 97 chooses an 8.43-minute longer walk while reducing time in direct sun by 7.27 minutes; person 5 has no change. The highlighted route replays separately from the background population so its decision is visible. The first two examples round to four minutes saved and seven fewer minutes in sun with an eight-minute longer walk. Some unchanged routes still benefit from added shade. A route change is not itself a measure of welfare.
Across all 180 people, mean walking time falls from 17.69 to 17.31 minutes (22.4 seconds saved) and mean direct sun falls from 11.50 to 11.04 minutes. Twenty-three people have at least a five percent reduction in their own weighted route cost; this diagnostic is retained in the data but the main slide uses the more understandable route-change count and three examples.
Localized additional shade is assigned within 24 m of proposed tree points, reducing the remaining exposed fraction by 0.65. The intervention graph explicitly joins duplicate node IDs at the same coordinate in the supplied shortcut sketch. This changes only the proposed connection, not the baseline isochrones.
The slide compares annualized cost directly with the monetary value assigned to time savings. Under these illustrative assumptions, time savings alone exceed the annualized cost. The cost slide assumes ₩150 million upfront, a 15-year life, and ₩10 million annual upkeep: ₩20 million per year using straight-line annualization. It assumes 1,000 daily trips with the same origin, destination, and preference mix as the synthetic sample, 300 days per year, and ₩12,000 per hour for time. These are invented discussion inputs, not local construction quotes, official values of time, or measured demand. Applying the paired population’s mean time saving gives about ₩22.4 million annual time benefit; approximately 894 daily walks would cover annualized cost through time savings alone. No discounting, uncertainty analysis, or valuation of shade, health, safety and vehicle impacts is included. This is a transparent workflow example, not an investment appraisal.
The intervention formalizes the Access shortcut as a proposed controlled crossing/full-junction connection. It is not a design drawing, engineering feasibility finding, or recommendation to cross there today. The direct route receives an assumed 30-second wait, compared with 60 seconds at the regular junction. Its shorter distance is the principal reason for the estimated time saving. Traffic operations, safety, vehicle conflicts, and accessibility would need evaluation.
The next scene adds proposed shade near the drawn tree points. Both the individual crossing route and population experiment use the same localized shade assumption: an additional 0.65 of the remaining exposed fraction within 24 m of those points. Shade does not increase walking speed. This is a simplified canopy scenario, not an engineered planting design.