Predicting Air Temperature from Volumetric Urban Morphology with Machine Learning

TL;DR

This work introduces a fast CityGML-to-voxel workflow to generate high-resolution 3D building volumetrics and link them to 2D air-temperature rasters for per-pixel prediction. By applying Gaussian blur to encode spatial neighbors and comparing Random Forest and XGBoost, the study demonstrates that per-pixel spatial patterns can be captured even when detailed building geometry is simplified to a single height per building. Evaluation relies not only on MSE but also image-based metrics like SSIM and LPIPS, revealing that spatially-aware assessments can favor RF over XGBoost despite higher MSE. The approach provides urban planners with a scalable, raster-based framework to assess environmental impacts of urban morphology and supports integration with higher-resolution meteorological data as it becomes available.

Abstract

In this study, we firstly introduce a method that converts CityGML data into voxels which works efficiently and fast in high resolution for large scale datasets such as cities but by sacrificing some building details to overcome the limitations of previous voxelization methodologies that have been computationally intensive and inefficient at transforming large-scale urban areas into voxel representations for high resolution. Those voxelized 3D city data from multiple cities and corresponding air temperature data are used to develop a machine learning model. Before the model training, Gaussian blurring is implemented on input data to consider spatial relationships, as a result the correlation rate between air temperature and volumetric building morphology is also increased after the Gaussian blurring. After the model training, the prediction results are not just evaluated with Mean Square Error (MSE) but some image similarity metrics such as Structural Similarity Index Measure (SSIM) and Learned Perceptual Image Patch Similarity (LPIPS) that are able to detect and consider spatial relations during the evaluation process. This trained model is capable of predicting the spatial distribution of air temperature by using building volume information of corresponding pixel as input. By doing so, this research aims to assist urban planners in incorporating environmental parameters into their planning strategies, thereby facilitating more sustainable and inhabitable urban environments.

Figures (8)

• Figure 1: Workflow of the voxelisation process
• Figure 2: The left column represents the CityGML visualization of Gotha from different viewing angles, the right column represents the voxel visualisation of the same region of Gotha
• Figure 3: A brief workflow demonstration of data pre-processing steps for the machine learning training
• Figure 4: This figure illustrates the spatial distribution of building volumes (expressed in units of $\mathrm{m}^3$) and a monthly average air temperature of July at 01:00 AM (expressed in units of $\mathrm{C}^\circ$) in selected urban areas to show the effect of Gaussian blur, and the relation between urban morphology and air temperature.
• Figure 5: Comparison of the air temperature prediction results obtained from the model trained with Random Forest technique and ground truth data for the test dataset which is not included during the training process. The difference map between prediction and ground truth images given in the bottom row.