Save people from the concrete barriers

Numerous studies found that a good view of nature has a huge influence on human mental and physical health and personal performance.

On the contrary, poor access to nature may lead to stress and diseases, and weaken one’s productivity at offices (Ulrich, 1984; Lottrup et al., 2015; Waczynska et al., 2021).

A window view is a significant medium for people’s visually access nature, especially for high-density urban cities such as Hong Kong.

Nature is in 3D, while our city in 3D, too.

For a window at 30/F, the view out is quite different from a window at 2/F in the same building.

The left figure shows how you see different views from different locations.

Alternatively, although people can physically access nature via active measures such as go hiking, jogging, and walking along the seaside, which used to be a common practice to travel beyond the concrete barriers, one has to rely more on their windows especially during the Covid time.

This study tries to answer three questions:

1) How to measure the 3D window views of nature effectively automatically at the urban scale?

2) What are the spatial patterns and reasons for the window view situations in Wan Chai?

3) Regarding both the visual and physical accessibility to nature, which buildings and town blocks need to be improved most?

Recently, a technological window of opportunity opens for this study:

• New  3D City Information Model (CIM)  opened by the Government,
• New  3D MapView APIs (Python)  opened by ESRI ArcGIS Pro, and
• New  2D/3D Computer Vision tools  opened online.

This study, therefore, proposes an automatic analysis workflow, which makes use of the advantages of the new datasets and technological breakthroughs, in terms of 3D and 2D mapping and analysis. There are 3 steps in the workflow, as shown in the flowchart:

1. Curation of data sets (1)  PlanD’s 3D CIM  on ESRI’s  3D terrain  and (2) 2D footprints, road network, and parks.
2. NVI assessment with three innovative tools: (1) ArcGIS Pro, (2)  DeepLab (ver3 ), deep learning models, and (3) our Python code.
3. Integrated analysis of (1) 3D NVI understanding and (2) NAI analytics on map.

The study area is in Wan Chai. We chose 314 buildings in 43 town blocks, as shown in the figure below. The 2D building footprints, road network, and parks were extracted and geocoded from the  Government digital map  and  ArcGIS Online .

The 3D city information model (CIM) was created on ArcGIS Pro 2.7.2 using  PlanD’s 3D photo-realistic model  on  ESRI’s 3D terrain model .

As a result, 19,837 window view photos were created automatically in the study area.

Every facade of every building, then, has a portfolio of window view photos. Each photo has 89.1 million color pixels.

4.3 Machine Learning-based NVI assessment

We applied a two-layer machine learning framework to measure the NVIs automatically.

• The first layer is a deep learning model  Deeplab v3  pre-trained on the  Cityscapes dataset .
• The second layer is a multimodal regression mapping (using the best-trained SVM, Adaboost, linear, etc.) on  Python-Orange  using a part of Wan Chai dataset.

The training-validation experiments showed that the RMSE (root-mean-squared error) of NVI was only 0.04 -- which means the training was very successful.

5.1 NVI patterns on 2D map

Statistics can show the distribution of NVIs as below.

The median NVI value (by building) is only 0.197. It shows that most of buildings have unsatisfactory nature views. Meanwhile, a few buildings' NVIs can reach as higher as 0.45. The average NVI in Wan Chai is 0.209, which is close to the median.

5.2 Possible reasons seen in 3D model

First, the buildings near the natural landscape, such as sea and mountains, tend to have higher chances to have higher NVIs.

For example, the Viewshed in ArcGIS Pro can contrast three view sites at the same altitude. The visible nature views within 600 meters are shown in green. From nearby to far away from the seaside, the sites have a decreasing WVI (and NVI). The main reason is the inter-buildings obstruction.

Secondly, the buildings with good artificial greenery management also own high-quality views.

For example, Ruttonjee Hospital has an excellent NVI (0.30~0.35), due to the two gardens around.

Last but not least, the buildings and rooms at good heights tend to have less visual concrete barriers to the nature.

For example, the Building 1 has a much higher NVI than Building 2. It is because the visual obstruction ratio (shown as the purple parts) drops down obviously over the height.

6.1 Accessibility to physical nature spaces

The occupants in Wan Chai have two ways to access nature, i.e., (1) visually and (2) physically.

The accessibility network is imported to supplement the view analysis. The buildings and town blocks with lower NVI than the average value may be prioritized in urban landscape management and optimization.

Urban residents can access nature visually and physically. The high-rise high-density 3D cities like Hong Kong challenges the traditional 2D analytics and measures. Furthermore, there exist a gap on connecting the two means of nature accessibility for urban computing.

This study utilizes the latest ArcGIS Pro and 3D city models, and presents:

1. An NVI metric from automatic 3D window view assessment, and
2. An NAI metric from an integrated analysis of both means of nature accessibility.

From the assessment results of 314 buildings in Wan Chai, three types of buildings tend to have better NVIs. Furthermore, three clusters of the lowest NAIs are pinpointed in Wan Chai. The clusters are associated with greater building ages, too. Finally, the nature accessibility issues in 5 aged zones can be recommended to Town Planning Board and Urban Renewal Authority's consideration.

The findings in this study are useful in:

1. Providing quantified shreds of evidence of the nature visibility and accessibility in a 3D city;
2. Revealing new means for urban optimization;
3. Highlighting inconvenient buildings and town blocks for planners, architects, dwellers, property developers, and other decision-makers;
4. Promoting urban greenery management and self-greenery management in Hong Kong.

Future work on Python API-enabled GIScience include:

• AI building optimization for window views, including aesthetic attributes, nature attributes, and cost attributes. 
• AI town optimization for nature accessibility, including both visual accessibility and physical accessibility.

LandsD. (2014). iB1000 Digital Topographic Map. Hong Kong: Lands Department, the Government of Hong Kong SAR.

Li, M., Xue, F., Yeh, A. G. & Lu, W. (2020). Classification of photo-realistic 3D window views in a high-density city: The case of Hong Kong. 25th International Symposium on Advancement of Construction Management and Real Estate. Singapore: Springer Nature. In press.

Lottrup, L., Stigsdotter, U. K., Meilby, H. & Claudi, A. G. (2015). The workplace window view: a determinant of office workers’ work ability and job satisfaction. Landscape Research, 40(1), 57-75. doi:10.1080/01426397.2013.829806

PlanD. (2019). 3D Photo-realistic Model. Hong Kong: Planning Department, Government of Hong Kong SAR. Retrieved from https://www.pland.gov.hk/pland_en/info_serv/3D_models/download.htm

Ulrich, R. S. (1984). View through a window may influence recovery from surgery. Science, 224(4647), 420-421. doi:10.1126/science.6143402

Waczynska, M., Sokol, N. & Martyniuk-Peczek, J. (2021). Computational and experimental evaluation of view out according to European Standard EN17037. Building and Environment, 188, 107414. doi:10.1016/j.buildenv.2020.107414