Balancing Construction Costs and Environmental and Social Performances in High-Rise Urban Development: A Generative Urban Design Approach

Abstract

The urban design process is complex and interdisciplinary, especially in the context of high-density cities with high-rise buildings. The design of high-rise buildings requires input from a variety of stakeholders in the city, who often represent conflicting requirements and interests. However, conventional design approaches struggle to address this complexity. This study introduced a generative urban design approach and applied it to a case study for high-rise urban development in Guangzhou, China. Specifically, 7500 urban forms were generated with variables such as street networks, block offset, building typology, and green space, and then were evaluated and optimized via multi-objective evolutionary algorithms. A total of 30,000 performance values were generated through various simulations. This study also presented a two-round, quantitative evaluation process against eight performance objectives from environmental, social, and economic aspects, including urban density, green space area, Daylight Potential, construction cost, Heat Stress, Green Space Accessibility, View Interest, and Shadow Impact.

1. Introduction

Cities are facing significant challenges due to rapid urbanization, climate change, and resource depletion. Therefore, designing future cities is becoming extremely complex and requires interdisciplinary collaborations. One solution that cities have been investigating is increasing density. Developing high-rise buildings is not the only way to increase urban density, but has become a viable solution, especially in fast-growing cities in Asia. The concentration of people in high-rise developments has made significant impacts on the environment, in both positive and negative ways, e.g., greater overall operational energy efficiency and lower carbon emissions vs. greater air pollution and decreased daylight access [1,2]. In addition, tall buildings present significant social challenges, such as reduced community interaction and access to green spaces. Many studies have examined both the environmental and social performances of high-rise developments qualitatively and quantitively, but very few were able to bring economic indicators, such as construction cost, into the equation when assessing the performance of high-rise developments. Particularly, designers often do not prioritize construction costs enough in early design stages, which might lead to potential cost overruns. Estimating the construction costs for tall buildings is even more complicated due to the complexity of the projects and limited data availability. Urban design is a very complex process, especially in the context of high-density cities with high-rise buildings, so it requires participation from a variety of stakeholders who might represent different requirements and even conflicting interests [3,4]. However, conventional design approaches struggle to address this complexity. Specifically, only a very limited number of design proposals are often developed and refined through manual iteration and experience without rigorously testing the full range of possible design schemes by quantifying different performance objectives, such as land use, zoning, density, transportation, building morphology, energy consumption, and carbon emissions [5,6]. Therefore, new design approaches and tools are very much needed to better equip urban designers to address various challenges from environmental, social, and economic aspects.

In recent years, the generative design approach, which is a framework that uses computational design methods integrating simulation tools and evolutionary algorithms, was developed to generate, evaluate, and evolve a very large number of design options at various scales [7,8]. The study introduced in this paper presented a generative urban design approach to generate urban forms and evaluate and balance their environmental, social, and economic performance. Specifically, a case study for high-rise urban development in Guangzhou, the capital and largest city of Guangdong Province in southern China, was conducted using this approach.

2. Literature Review

2.1. Construction Cost of High-Rise Buildings

It is generally accepted that the construction costs increase as the building height rises on a floor area basis. However, very few empirical studies on the relationship between construction cost and the height of tall buildings have been conducted. Recent studies have found that the relationship between the construction cost and building height was represented by a U-shaped curve, instead of a simplistic linear regression model [9,10,11]. The U-shaped curve suggested that the construction cost per square meter decreased initially as the number of stories increased, but it started to rise after passing the bottom point.

The construction cost of high-rise buildings is greater than low-rise buildings mainly because of the greater costs for vertical transportation, mechanical systems, tower cranes, on-site labor, etc. [12,13]. The number of studies on high-rise construction costs, especially for buildings over 100 m tall, is extremely limited, largely due to data availability. Two recent studies on the relationship between high-rise construction costs and building height confirmed a U-shaped curve [11,14]. D. H. Picken and Ilozor examined 24 residential buildings in Hong Kong ranging from 3 to 39 stories (height ranging from 9 m to 112 m) completed in the early 1990s, and found that the construction cost declined until the height reached 100 m, and then raised again [11]. Blackman and Picken analyzed 36 residential buildings in Shanghai ranging from 2 to 37 stories (height ranging from 6 m to 115 m), completed between 2000 and 2007, and found that the construction cost declined until the height reached 24 m, and then raised again [14]. The raw data available in the literature were used to predict the construction cost in this study.

2.2. Multi-Objective Evolutionary Algorithms

Evolutionary algorithms are population-based metaheuristic optimization models that generate design solutions paired with competing performance metrics. There are two main types of design optimization in evolutionary algorithms in the generative design process: single-objective optimization and multi-objective optimization [7]. However, multi-objective evolutionary algorithms have been proven to be more effective in more complex urban contexts. A successful optimization run using objective evolutionary algorithms relies on the relationships between three key metrics: the parameters informing the morphology (genes), the morphological characteristics being generated (the phenotype), and the performance objectives (fitness functions) [15].

An increasing amount of generative urban design research has applied multi-objective evolutionary algorithms to generate and optimize urban forms paired with a set of quantifiable performance objectives, and to conduct and visualize comparative analysis on different design solutions [16,17,18]. However, most of the studies conducted using multi-objective evolutionary algorithms ended up with a one-round selection process from the algorithm’s output. However, this one-round selection analysis limits the understanding of the design proposals within pre-determined performance objectives, which can only be a few given the capacity of the tools that are currently available. Some very recent research included two-round selection stages, but the performance metrics in the second stage were evaluated subjectively. Therefore, this study examined more performance objectives via multi-objective evolutionary algorithms, by integrating a two-round selection process with two different sets of quantifiable objectives to find optimal solutions. Also, most of the studies focused on social and environmental performances, such as energy, carbon, daylight, mobility, and comfort, but very few were able to integrate economic indicators into the overall design performance. Therefore, the research methods introduced in the Research Methodology section below will fill these two significant research gaps by integrating construction costs and a wide range of environmental and social performance indicators into a generative urban design workflow.

3. Research Methodology

3.1. Urban Form and Performance Objectives

This study began with the creation of a parametric urban model using Urbano, an urban mobility modeling plugin for Rhino/Grasshopper, with data loaded from GIS and OpenStreetMap (OSM) [19,20]. Then, an algorithm was created to generate the urban form, with variables such as street networks, block offsets, buildings, and green spaces, and meanwhile examining the performance of these urban forms through various Grasshopper plugins. This study conducted a two-round, quantitative evaluation process. Specifically, the study integrated 4 performance objectives, including urban density, green space area, Daylight Potential, and construction costs, into the evolutionary algorithms in the first-round selection process, followed by another round of evaluation by using a different set of 4 performance objectives, including Heat Stress, Green Space Accessibility, View Interest, and Shadow Impact (

Table 1. The performance objectives used in the two-round, quantitative evaluation process.

No.	First-Round Evaluation		Second-Round Evaluation
	Performance Objective	Measurement	Performance Objective	Measurement
1	Urban Density	Floor Area Ratio (FAR)	Heat Stress	Universal Thermal Climate Index (UTCI) on the street level
2	Green Space Area	Green Area Ratio (GAR)	Green Space Accessibility (GSA)	The walking distances connecting the green spaces within the site from the surrounding neighborhoods
3	Daylight Potential	The ratio of the building facades with direct sun exposure	View Interest	ratio of the building facades with direct view access to the river
4	Construction Cost	The total cost of constructing all the buildings	Shadow Impact	Direct sun exposure to the buildings outside the site boundary

).

3.2. Construction Cost Estimates

The research used data from the study on Hong Kong high-rise construction costs [11] introduced in the Literature Review section due to the geographical proximity to the study area in Guangzhou, and then conducted a quadratic regression analysis, which is a statistical method to model a relationship between two sets of variables. Specifically, a quadratic regression model (y = 0.0809x² − 16.13x + 2062.6, R² = 0.32) was created to predict the estimated construction costs of the buildings, with a height up to 250 m. The development of buildings taller than 250 m has been strictly restricted since 2022 by the Chinese government. Therefore, the research used 250 m as the height limit of buildings (Figure 1). As the Hong Kong study was conducted in 2001, the construction cost growth over time was factored in based on the Building Works Tender Price Index (BWTPI) [21], and then the prices were converted to United States Dollars (USD) from Hong Kong Dollars (HKD). The Hong Kong study was conducted in April 2001, so this study used 2001 Quarter 1 and the latest available 2023 Quarter 3 data from the Building Works Tender Price Index (BWTPI) to project the estimated construction costs for today (Hong Kong Architectural Services Department, n.d.). The study used 0.13 as the conversion rate from Hong Kong Dollars (HKD) to United States Dollars (USD).

3.3. Multi-Objective Optimization

A multi-objective evolutionary algorithm was developed using Wallacei, an evolutionary and analytic plugin for Rhino/Grasshopper. Through this algorithm, the performance of each design objective of every single iteration was evaluated and optimized. Then, the pareto front solutions were used, which are a set of solutions that are non-dominated, representing the fact that the optimal trade-offs between different objectives when improving one objective cannot be obtained without worsening another. Selecting pareto front solutions helped the selection process focus on the groups of efficient options and make tradeoffs within this set, rather than considering the full-size population (the total number of iterations). Further, a K-means clustering algorithm was applied to the pool of the pareto front solutions, so these solutions were organized into clusters based on their similarity. In this process, the solution closest to the center of each cluster was selected. Lastly, a few additional solutions were selected, including the most optimal solution for each performance objective, the “Average Solution” (with the closet equal weighting between all the performance objectives) and the “Ideal Solution” (with the closest optimal result for all the performance objectives). The “Average Solution” is considered a non-existent optimal solution in the objective space, with the closest equal weighting between all the design objectives. Similarly, the “Ideal Solution” is also considered a non-existent optimal solution in the objective space, and thus the selected solution is the one nearest to this point based on a Euclidean measure between the two [15].

4. Case Study: High-Rise Urban Development in Guangzhou, China

4.1. Site Analysis

The site is in the International Business District, a fast-growing, high-density area in Guangzhou, China. Situated alongside the Pearl River in the south, the site is also adjacent to a canal in the East (Figure 2). The area of the site is approx. 73,000 m² and is expected to be developed into a high-rise urban complex, with a building height limit of 250 m. Therefore, the key focus of the design was to increase the urban density but also balance environmental and social implications. One of the major challenges for a high-rise, mixed-use development is the shadow caused by the buildings, so increasing daylight access was one of the key design strategies. Both the river and canal are important ecological and environmental assets to the site, so the design strategies also included an increase in the visual connections with the water as well as the accessibility to open spaces. Due to accelerated climate change, heat has significantly impacted our society, including a rise in heat-related deaths, for many cities nationally and internationally [22], and this is particularly true for cities with hot and humid climatic conditions, such as Guangzhou. Thus, this study also implemented strategies to mitigate the heat impact. There are very limited green spaces in the surrounding urban context, so the design also focused on creating more green spaces, not only on the ground level but also through the rooftops.

4.2. Form Generation and Performance Evaluation

The main elements of the urban form included streets, parcels, buildings, and open spaces. A 20 m-wide green corridor was created alongside the canal to be a recreational and ecological space on the waterfront. A total of 3 building types were created based on the height and program: Building Type 1 for office and mix-used with a height range of 175–250 m, Building Type 2 for residential used with a height range of 75–150 m, and Building Type 3 for retail and entertainment with a height range of 25–50 m. In order to reduce the pressure of high-rise buildings on the street level, and to create rooftop spaces for activities, all three building types have podiums, with a height range of 24–40 m for Building Type 1, and 8–24 m for Building Types 2 and 3. The floor height was set to be 4 m across all the buildings.

The urban form generation process included nine major steps (Figure 3): (a) adding movable points on the site boundary; (b) connecting the points with offset to create streets; (c) creating parcels and allocating green spaces (occupancy range: 1–3 parcels); (d) allocating building types, with Type 1 in red (occupancy range: 1–3 parcels), Type 2 in blue (occupancy range: 1–5 parcels), and Type 3 in yellow (occupancy range: remaining parcels); (e) randomly selecting one out of the four corner points of each parcel, and scaling down the parcel to be the footprint of the main building; (f) scaling up the parcel based on the point selected to create the footprint of the associated podium; (g) extruding the footprint of the main buildings; (h) extruding the footprint of the podiums; And (i) identifying extra green spaces in addition to the green spaces on the ground allocated at Step (c), including unbuilt lands within the parcels if the area is greater than 1000 m² (otherwise turned into a plaza) as well as the podium roofs if the area is greater than 500 m² (otherwise turned into a roof deck).

As introduced in the Research Methodology section, four performance objectives were integrated into the evolutionary algorithm (Figure 4), through the following calculation and measurement: (a) The Floor Area Ratio (FAR) = the total floor area/total site area. (b) The Green Area Ratio (GAR) = the total area of the green spaces/total site area. The green spaces included the green corridor (5997 m²) alongside the canal, the dedicated parcels for green spaces, as well as the areas of the podium roofs (if >500 m²). (c) Daylight Potential is measured by the ratio of the building floors with direct sun exposure through the facades to all the building floors. The analysis ran from 9 a.m. to 5 p.m. on the Winter Solstice in Guangzhou, with 2 h of minimal direct sun exposure. An algorithm was developed to contour the buildings every two floors and then connect the midpoint on each exploded segment of the buildings to the points in the sky presenting the sun’s locations during the analysis period. (d) The construction cost, measured by the total construction cost of all the buildings in USD in millions. The price per square meter for each building was determined based on the building height and the quadratic regression model (y = 0.0809x² − 16.13x + 2062.6, R² = 0.32), and then was multiplied by the GFA of each respective building to estimate the total construction cost.

5. Analysis and Selection

5.1. Result Analysis

The algorithm ran a total population of 7500 (design solutions) via Wallacei comprised of 100 generations with 75 solutions each. Each solution included four performance objective values, so there were 30,000 values in total. Each performance objective (fitness function) was analyzed separately across four key metrics: the standard deviation, fitness values (performance values), the standard deviation trendline, and the mean value trendline (Figure 5). The simulation was proven to be successful in improving the mean values for Performance Objective 1 (FAR) and Performance Objective 2 (GAR) based on the mean value trendline. The variation of solutions fluctuated throughout and did not converge toward an optimal result for Performance Objective 3 (Daylight Potential) and Performance Objective 4 (construction cost).

5.2. Evaluation and Selection

As introduced in the Research Methodology section, a two-round selection process was conducted. In the first round, 809 pareto front solutions out of the entire population (e.g., 7500 iterations) were identified. Then, these selected solutions were grouped into 20 clusters via a K-means clustering algorithm. Specifically, the solution closest to the center of each cluster was picked, so a total of 20 solutions were selected. In addition, the optimal solution for each performance objective, e.g., the highest Floor Area Ratio (FAR), the highest Green Area Ratio (GAR), the greatest Daylight Potential, and the lowest construction cost, were selected, totaling fou solutions. Finally, two more special solutions, the “Average Solution” and “Ideal Solution”, were selected. The “Ideal Solu-tion” happened to be the same as the optimal solution for the Daylight Potential, so, the number of the selected solutions was reduced to 25 from 7500 (Figure 6).

The result of each performance objective for each solution, including the absolute value and its ranking among all the 7500 solutions, was exported, analyzed, and visualized via Wallacei (Figure 7). Lower values represent better performance on Wallacei. “Solution (Gen. 68|Ind. 43)” was found to be the “Average Solution” (with the closet equal weighting between the four performance objectives), presenting 5.33 FAR (ranking # 3741/7500), 32% Green Area Ratio (ranking # 4229/7500), 39% Daylight Potential (ranking# 3802/7500), and a construction cost of USD 714.48 million (ranking # 4171/7500). “Solution (Gen. 53|Ind. 2)” was found to be the “Ideal Solution” (the closest optimal result for the four performance objectives), presenting 3.75 FAR (ranking # 6517/7500), 36% Green Area Ratio (ranking # 2551/7500), 55% Daylight Potential (ranking# 1/7500), and a construction cost of USD 193.17 million (ranking # 764/7500).

5.3. Second-Round Evaluation

In order to offer a more comprehensive understanding of the overall performance of the design proposals produced by the algorithms, an additional four performance objectives were added to further evaluate the 25 selected solutions (Figure 8) through the following calculations and measurements: (a) Heat Stress was measured by the ratio of the number of “heat stress” hours to the total hours reflecting the Universal Thermal Climate Index (UTCI) on one of the hottest days in Guangzhou. Six categories are defined to reflect the UTCI conditions: No Thermal Stress (9 ≤ UTCI < 26), Slight Heat Stress (26 ≤ UTCI < 28), Moderate Heat Stress (28 ≤ UTCI < 32), Strong Heat Stress (32 ≤ UTCI < 38), Very Strong Heat Stress (38 ≤ UTCI < 46), and Extreme Heat Stress (46 < UTCI). The “heat stress” defined in this study included the three strongest Heat Stress levels: Strong Heat Stress, Very Strong Heat Stress, and Extreme Heat Stress. 26 August to 1 September was found to be the hottest week in Guangzhou. The study ran the UTCI simulation on 26 August for the full day and calculated the “heat stress” hours. (b) Green Space Accessibility (GSA) was measured by the walking distances between the green spaces and the surrounding neighborhoods. Two corner points on the north boundary of the site were set as the starting points for walking towards the green spaces within the site. The average length of the pedestrian paths from the starting points to the middle points of the designated green spaces was used for comparison. (c) The View Interest was measured by the ratio of the building floors with a direct visual connection with the canal to all the building floors. The visual connection to the river was excluded in the quantification of the View Interest. The river is in the south, so optimizing the view to the river would lead to the production of similar solutions to optimizing the Daylight Potential. It is important to set up competing objectives in the multi-objective evolutionary algorithm. An algorithm was developed to contour the buildings every two floors and to connect the midpoint on each exploded segment of the buildings to the middle line of the portion of the canal adjacent to the site. (d) The Shadow Impact was measured by the change of the direct sun hours on the facades of the surrounding buildings immediately outside the site boundary with and without the proposed urban development.

The study conducted simulations on these four additional performance objectives for the 25 selected solutions through various Rhino/Grasshopper plugins. Then, the simulation results were combined with the results of the previous four performance objectives for the final evaluation. A ranking criterion was developed to evaluate the overall performance of each solution. Specifically, the average value of each performance objective was set as a threshold, and if the performance value was better than the average, one point will be gained. Performing better than the average did not mean that the value was greater than the average. The higher value the better for the performance objectives FAR, GAR, Daylight, and View Interest; and the lower value the better for the performance objectives construction cost, Heat Stress (UTCI), Green Space Accessibility, and Shadow Impact. For example, the average FAR among all the 25 solutions was 5.43, so a solution with a FAR greater than 5.43 received one point, otherwise it received no point. The final ranking of the solutions was generated based on the total points (Figure 9).

Following this approach, no solution received eight or seven points. The top-ranked solutions included one solution with six points and four solutions with five points each (Figure 10). “Solution (Gen. 41|Ind. 60)” was found to have the best overall performance, with six objectives performing better than the averages, including the Green Space Ratio (GAR), the Daylight Potential, the construction cost, Green Space Accessibility (GSA), View Interest, and Shadow Impact.

Linear relationships between some performance indicators were found as predicted, e.g., the higher the FAR the greater the construction cost, but some relationships were dynamic, which offered unique insights for urban designers. For example, “Solution (Gen. 20|Ind. 66)” had the fourth highest FAR (6.91) with an above-average GAR (36%), meaning that we could still obtain large green spaces in an ultra-dense urban development; “Solution (Gen. 66|Ind. 02)” had a below-average GAR (29%) with the third greatest GSA (153.8), meaning that we could still make the green space very much accessible even though the size of the green space is limited.

6. Discussion

This study introduced a generative urban design approach to address complex and interdisciplinary challenges in the context of high-rise urban development. Specifically, multi-objective evolutionary algorithms were used to generate and evaluate how morphological variation evolved in response to conflicting performance objectives across environmental, social, and economic aspects. One of the major limitations of multi-objective evolutionary algorithms in urban design is the very limited number of objectives that can be applied in the algorithms due to the capacity of the tools that are currently available, as well as the simulation time; therefore, many studies have used subjective indicators to expand the scope of the performance matrix. The two-round, quantitative evaluation process with two different sets of performance objectives (totaling eight performance objectives) presented in this study offers an innovative approach to expanding the number of quantifiable performance objectives, filtering a large solution set (totaling 7500 urban forms with 30,000 performance values), and understanding each solution’s performance environmentally, socially, and economically. Particularly, integrating construction cost into the algorithms as one performance objective is a unique contribution to the early stage of urban design. There are other important indicators involved in urban design that are hard to quantify, e.g., cultural and aesthetic considerations. How to integrate such indicators into a generative urban design workflow, either through algorithms or other analytical approaches, should be explored in future research.

It is important to highlight that the generative urban design workflow developed in this study was not intended to give a definitive answer on which solution is the best. As discussed above, the urban design process is complex, so it requires input from a variety of stakeholders who often represent conflicting requirements and interests. The ranking criterion proposed in the study treated each of the eight performance objectives equally, but future research may apply different weights to the metrics representing the priorities of certain stakeholders. Using the performance objectives in this study as an example, developers might give more weight to the construction cost, urban designers might give more weight to Shadow Impact, and landscape architects might give more weight to the Green Area Ratio and accessibility. Also, different weights might be given based on the project’s location and climatic environment. In this study, for example, more weight could be given to Heat Stress and Shadow Impact, as Guangzhou has a subtropical monsoon climate with high temperatures and humidity in summer.