The Sky View Factor (SVF) is a key parameter in urban climate modelling, which quantifies the fraction of the visible sky from a given point and allows for the estimation of incident solar radiation, thermal comfort, and urban heat distribution. High resolution in SVF computation is essential for microclimatic studies in the canopy layer, where detailed representations of urban environments are crucial for understanding variations in heat exposure. Traditional SVF calculation methods often rely on sequential processing of shadow projections, however, these methods are computationally intensive and time-consuming, particularly for urban climate analyses at high resolution.

These computational challenges are further magnified when incorporating complex components such as vegetation, where tree crowns exhibit intricate geometries, partial transparency, and permit sky visibility from beneath the canopy. These properties require detailed modeling to account for light penetration and obstruction. This significantly increases the computational cost of Sky View Factor calculations and extends runtime to hours or even days, depending on scale and resolution.

This study introduces a novel GPU-accelerated ray tracing approach for SVF calculation, designed to address the computational limitations of traditional methods for large-scale analyses. By utilizing NVIDIA GPUs and the CUDA programming framework, the method applies parallel computing to perform ray tracing across the full range of azimuth and altitude angles. It estimates SVFs by systematically weighting blocked rays based on their spatial contributions to the hemisphere.

The accuracy of the developed method is validated through two complementary approaches. First, modelled SVF values are compared against theoretical expectations derived from idealized geometric environments. Additionally, a test case on a neighbourhood in Rotterdam is conducted to compare the results of the developed method against those obtained using an established SVF estimation technique using a serial approach. In addition to accuracy, computational efficiency is evaluated by comparing processing times across different study area extents with those of a CPU-based implementation. The proposed GPU workflow achieves a 99% reduction in processing time compared to traditional shadow casting methods performed on a CPU, while maintaining similarly high resolution and accuracy.

Urban Multi-scale Environmental Predictor (UMEP) is a climate service tool focused on urban climate simulations, using meteorological, surface, and land cover data to model a variety of climate indicators. Among these, Mean Radiant Temperature (MRT) can be computed for outdoor thermal comfort analyses, by using the Solar and LongWave Environmental Irradiance Geometry (SOLWEIG) model. A critical input for SOLWEIG is a set of Sky View Factor (SVF) maps, which quantify the fraction of visible sky at each point of the urban environment under study. However, generating these SVF maps from high-resolution digital surface models is computationally expensive and represents a significant limitation in the use of UMEP for large-scale urban studies.

Thus, this study addresses the above limitation by introducing a GPU-accelerated workflow for SVF calculation, leveraging the powerful nature of NVIDIA GPUs and PyCUDA to enable parallelized ray tracing. The proposed method incorporates anisotropic SVF calculations and accounts for vegetation canopies, which are a critical factor in accurate calculations of urban climate parameters. By replacing the CPU-based SVF calculations currently integrated within UMEP, the proposed GPU-based workflow achieves a 99% reduction in processing time while maintaining accuracy and compatibility with SOLWEIG requirements.

The proposed method was applied in the Rotterdam case study, demonstrating its usability within the UMEP climate service tool. The reduction in computational time significantly accelerates pre-processing for MRT calculations, enabling the modelling of city-large areas at a 1-meter resolution. This advancement represents a step forward in optimizing urban climate modelling workflows, enhancing their scalability and usability for researchers and practitioners.