Extreme heat is the deadliest climate hazard in the United States, and its burden falls unevenly. Within a single city, air and surface temperatures can differ by more than 5 °C between a tree-lined residential street and an adjacent industrial corridor, and those differences track historical patterns of disinvestment. The available tools are mismatched to the problem: mesoscale weather models resolve kilometers, while the difference between a shaded street and an exposed one is tens of meters, and satellite thermal imagery shows where a city is hot but cannot evaluate a tree that has not yet been planted. My work builds the machinery that connects the physics to the decision.
The GDSL Digital Twin
At Purdue’s Geospatial Data Science Laboratory, I lead development of an operational, cloud-based urban microclimate digital twin. It delivers physically based simulation to any web browser, for anywhere on Earth, without the user preparing a mesh or accessing a cluster. A user selects a neighborhood on an interactive 3D globe rendered with CesiumJS; the system assembles a physical description of that place from open geospatial data and returns street-scale temperature, wind, and thermal comfort fields.
A GPU-Accelerated Physics Engine
The solver integrates the three-dimensional incompressible Navier–Stokes equations under the Boussinesq approximation, coupled to a temperature advection–diffusion equation, on GPU hardware via CuPy with an HPC backend. Urban-specific process models represent vegetation canopy drag and evapotranspirative cooling, solar shading resolved through the day, and the surface energy balance of contrasting materials. Live observations from urban sensor networks are assimilated into the simulation state through relaxation nudging, so the twin remains anchored to the real atmosphere rather than drifting into free-running simulation. Output is expressed in the quantities decisions turn on: pedestrian-level wind, thermal comfort, neighborhood heat island intensity, and the marginal cooling benefit of a proposed intervention.
Reconstructing Urban Form from Open Geospatial Data
Behind every simulation is an automated geospatial pipeline. It streams airborne LiDAR point clouds on demand through STAC-API catalogs and Entwine Point Tile indexing, crops them to the selected domain, and applies AI semantic segmentation to classify ground, building, and vegetation returns. Segment Anything Model (SAM) object detection is applied to imagery to locate individual objects in the scene, including trees, vehicles, and building elements, and their positions are carried into the simulation geometry. From these sources the pipeline reconstructs a digital elevation model, building footprints and heights, and a three-dimensional tree-canopy representation, fused with OpenStreetMap building, road, canopy, and hydrography layers. This automates the step that has historically made building-resolved urban climate modeling too labor-intensive for routine use.
Coupling to the Regional Atmosphere
A street-level simulation is only as good as its boundary conditions, which are set by synoptic weather and boundary-layer evolution that no neighborhood domain can resolve on its own. I am coupling the microscale engine to mesoscale WRF modeling so that a single framework spans regional circulation down to street-canyon radiative exchange, shading, and roughness. This is what allows one system to answer both what a block will feel like tomorrow afternoon and what it will feel like under mid-century warming.
From Demonstration to Research Program
The platform is not tied to a single test site: it runs for any location on Earth, assembling the domain from open geospatial data wherever coverage exists, and demonstrations such as Hamilton County, Indiana, establish the full chain from public LiDAR to browser-rendered microclimate fields. I am extending it along three lines: systematic validation against distributed sensor networks and vehicle-mounted mobile transects, to map not only predictive skill but where the model fails and why; coupling microclimate fields with population, mobility, and health data, to move from mapping heat to estimating who is exposed; and co-production with municipal planning departments, public-health agencies, and community organizations. Because the twin runs in a browser, it can be placed directly in residents’ hands at a public meeting rather than delivered as a consultancy report.