3D modeling for real estate valuation in China


Rapid urban development in China has transformed cityscapes, with a dramatic increase in tall buildings. However, current 2D real estate valuation methods neglect the vertical dimension, even though this aspect contains important metrics that impact the value of the property – such as types of views, range of view, and light of the property. day. 3D modeling of geographic data can help change this in high-rise and high-density areas, going beyond 2D to allow the identification of complex property features in the vertical dimension as well. This article presents research to explore a 3D modeling approach using remote sensing data for a more precise evaluation of properties.

Accelerating urbanization has put great pressure on land in China, posing the challenge of precisely defining property characteristics and values. However, the integration of GIS and real estate valuation are mainly based on two dimensions. When it comes to dealing with the complexity of vertical developments in urban areas, current methods are unable to describe in detail the 3D features of a property and provide precise information on spatial variations between properties. Given the considerable rise in real estate prices and technical advances in GIS, 3D modeling is becoming more and more relevant for real estate valuation in urban areas in order to better describe the complex characteristics of properties, especially in the vertical dimension. , as recent research in China shows. .

Study zone

Xi’an, the capital of Shaanxi province, is located in west-central China. Xi’an is the educational, political and economic center of Shaanxi Province and northwestern China. It also attracts tourists from all over the world. Xi’an was chosen for the research project because of its data availability and its real estate market, which has been particularly flourishing in recent years. While it is both expensive and time-consuming to study the effects of 3D indicators, the anticipated neighborhood-wide results can be useful, giving potential buyers a quick estimate of the price of real estate in a specific neighborhood. and also making the study feasible in terms of calculation and data collection.

Figure 1, The 2D indicators included in the 2D model (Blue: significant at the significance level of 0.05; Gray: not significant).

Superior level of detail

3D modeling and its interactive functions for property valuation have so far not been very popular with the municipal government and real estate developers in Xi’an; it has only been explored by real estate developers in a few large cities (eg Shenzhen). Xi’an municipal government uses some 3D modeling, but mainly in the field of city planning and the use cases are limited due to low levels of detail (LOD) influenced by huge cost, low technology and public security policy. However, 3D modeling can be developed to a higher level of detail to serve more purposes in the future (eg emergency management). Real estate developers use 3D modeling for architecture and landscaping, but rarely use it for business purposes as they believe / perceive traditional 2D sales tools to be efficient and intuitive. Hukou’s policy reform has contributed to substantial population growth in Xi’an and thus creates higher demand for housing. Now that the Xi’an real estate market is a seller’s market, real estate developers have no impetus to develop new technologies. Therefore, an appropriate 3D modeling approach must be determined taking into account local circumstances.

Figure 2, The 3D modeling workflow.

Explore the 2D-based method

In this research project, a regular hedonic price model with 2D indicators was first applied. Generally speaking, most research has concluded that changes in house prices can be generalized by different statistical models. However, the R2 of the 2D methods applied in this research, ordinary least squares (OLS) and geographically weighted regression (GWR) were too weak to generalize a model (0.111 in OLS and 0.217 in GWR), which meant the current 2D indicator models (Figure 1) could not explain the variations in house prices. The reason may be the use of first-hand property price data that was generated under the fixed price and purchase restriction policy established by the Xi’an municipal government, including the he main objective was to stabilize real estate prices. Therefore, prices may not show a significant impact of geographic locations and surroundings.

The 3D modeling process and the results

The preparation of a 3D model was divided into two main stages: the classification of the land cover based on the remote sensing data and the construction of the structure of the 3D data (Figure 2). The multispectral images used for the classification came from the Chinese satellite Gaofen-2, with a resolution of 4m, and Sentinel-2, with a resolution of 10m. Classification was performed by a support vector machine and validated by visual interpretation (Figure 3). The construction of the 3D data structure was done in ArcGIS to create height attributes for the 2D floor plans, which could be directly linked in CityEngine. In the 3D modeling process, two high-rise neighborhoods were explored. The 3D metrics taken into account were view type, sky view factor (SVF), sunshine, and property orientation. These were determined on the basis of local knowledge acquired through a questionnaire sent to the residents of Xi’an. The first three indicators were simulated by functions integrated in CityEngine: field of view and sunlight analysis. The orientation of the property has been defined as a dummy variable. The regression results revealed that the R2 of the 3D method (0.451) was significantly higher than those of the 2D methods. SVF, sunlight, and property orientation were found to have a significant influence on property prices at the significance level of 0.05. The results revealed that the 3D method could better explain changes in house prices than the 2D methods. The 3D visualization in CityEngine also showed excellent performance regarding the quantification and analysis of 3D indicators (Figure 4).

Figure 3, The classified image of the study area with an overall accuracy of 87.5%.

3D model validation

The regression results of the 3D model were validated by a Leave-one-out cross-validation and the results were found to be quite good, with a 9.76% error predicting the house price. However, no research has ever established a threshold or developed an approach for this type of model validation. The analysis of the 3D indicators was entirely simulated in software and it would be tedious to collect them in the field. Therefore, validation of the 3D model will remain an important issue to be addressed in the future.

A way forward

This research has proven that the 3D modeling approach, with its vivid visualization and powerful calculations, is effective for high quality real estate appraisal. It shows a way forward in which property assessment can be conducted more comprehensively, accurately and cost effectively with high resolution remote sensing data and 3D data structure support. Moreover, it can be widely implemented and adapted to other fields, for example to set the price standards for future property tax in China. The researchers recommend more exploration in two aspects in particular. First, the visualization and quantification of 3D indicators can be further developed. Accuracy can be improved with higher LOD. In addition to this, different calculation techniques can influence the research results, so there is a need to develop an efficient method to quantify 3D indicators. Second, this research contained a tremendous amount of manual labor. Therefore, further exploration of automated techniques is recommended.


This work was supported by ITC and we thank Dr Jong Wang and Dr Xia Li for data support.

Further reading

3D scenarios of the studied districts: http://bit.ly/2IEA3Wy / http://bit.ly/2Iyl4NJ

Figure 4, Real estate price variations of a sample neighborhood represented in 3D at the neighborhood scale.