City-Scale Wall Irradiance Estimation (LOD-1, Ray-Tracing Based)

This project computes the total solar irradiance (direct + diffuse + reflected) received by each wall of every building in a user-supplied city map. Using a shapefile as input, it constructs an LOD-1 3D model of all buildings, performs GPU-accelerated ray tracing to evaluate shading, and outputs per-wall irradiance maps and heatmaps for further visualization and solar potential analysis.

The pipeline is designed for city-scale performance, leveraging a Bounding Volume Hierarchy (BVH) and a compute shader for massively parallel shadow testing.

Visualization Preview

The image above shows an example visualization of the application’s output rendered in CesiumJS.

This visualization is not part of the repository; it is simply a demonstration of how the generated GeoJSON files and irradiance heatmap textures can be used in external tools for interactive exploration.

Features

• Converts 2D building footprints + height attributes into LOD-1 extruded 3D geometry
• Uses ear-clipping triangulation for polygon meshing
• Builds a BVH acceleration structure for fast ray–triangle intersection
• Incorporates hourly EPW weather data (irradiance + sun position)
• Performs GPU-parallel ray tracing via compute shaders
• Computes direct, diffuse, and reflected irradiance per building wall
• Generates irradiance heatmaps and GeoJSON output for visualization

1. Input

1.1 Shapefile

The shapefile must contain:

• Building footprints (2D polygons)
• A building height attribute

1.2 EPW Weather File

The EnergyPlus Weather (EPW) file provides:

• Global Horizontal Irradiance (GHI)

• Diffuse Horizontal Irradiance (DHI)

• Direct Normal Irradiance (DNI)

• Solar azimuth and zenith angles

  These values are used per hour of the year.

2. Pipeline Overview

Step 1 — Load Shapefile

Extract:

• Building coordinates
• Polygon geometry
• Building height

Step 2 — Generate LOD-1 Geometry

Each polygon is extruded to its height to create walls and roof faces.

Polygons are triangulated using Ear Clipping Triangulation, which runs in:

• O(k²) time for a polygon with k vertices
• O(n) overall for city data, since k is small and fixed per building

Step 3 — Build the BVH

A Bounding Volume Hierarchy is constructed over all triangles.

Properties:

• O(n) memory usage
• O(log n) ray intersection queries
• Greatly accelerates ray tracing for shadow casting

Step 4 — Extract Hourly Irradiance & Sun Direction

From the EPW file, per hour:

• GHI, DHI, DNI
• Sun direction vector

These determine:

• Whether a wall is sun-facing
• The amount of irradiance it can receive

Step 5 — Sample Points on Each Wall

Walls are uniformly sampled.

The sampling density controls:

• The resolution of the heatmap
• The accuracy of shadow estimation

Each sampled point is sent to the compute shader.

Step 6 — GPU Compute Shader (Ray Tracing)

A compute shader launches one thread per sampled point, performing:

1. Generate a ray at the point in the sun direction
2. Traverse the BVH
3. Detect if any building obstructs the ray

Outputs:

• Whether the point receives direct sunlight

Step 7 — Compute Irradiance per Wall

For each hour, for each wall:

shadowFraction = numObstructedPoints / totalSamplePoints
cosTheta       = max(0, dot(outwardNormal, -sunDirection))

eDirect   = DNI * (1 - shadowFraction) * cosTheta
eDiffuse  = DHI * (1 + tiltFactor) / 2
eReflected = GHI * 0.2 * (1 - tiltFactor) / 2

totalIrradiance = eDirect + eDiffuse + eReflected

Diffuse and reflected irradiance use a standard isotropic sky model with a fixed ground albedo of 0.2.

Step 8 — Output Heatmaps + GeoJSON

For each building face:

• Irradiance values at sampled points
• A texture (heatmap) visualizing irradiance distribution
• A GeoJSON file linking geometry to irradiance

This enables use in GIS tools, dashboards, or BIPV potential workflows.

3. Output

• Wall-wise irradiance values (W/m² or Wh/m²)
• Heatmap textures for each wall
• GeoJSON with irradiance attributes

4. Limitations

• Uses LOD-1 geometry only (no roofs with slopes, balconies, overhangs)
• Trees and vegetation are not modeled
• Uses an isotropic diffuse model and fixed ground albedo
• Shadowing considers only building geometry

5. Technologies Used

• C# (Unity or .NET)
• Compute shaders (HLSL)
• BVH acceleration structure
• Ear clipping triangulation
• EPW weather file parser
• GIS libraries (QGIS) for shapefile processing

6. Future Work

• Add tree geometry and vegetation occlusion
• Integrate LOD-2 or LOD-3 building models
• Support anisotropic sky models (Perez, Hay-Davies)
• Incorporate multiple reflections or radiosity
• Improve sampling adaptively based on geometry