Intelligent extraction of urban ventilation corridors based on region growth algorithm with restricted direction

Abstract

Rapid urban developments have resulted in numerous environmental challenges, including severe air quality and heat island effect problems. The construction of urban ventilation corridors provides an effective solution for enhancing air circulation and reducing the heat island effect. The present study focuses on Nanhu District, Jiaxing City, Zhejiang Province, employing Unmanned Aerial Vehicle tilt photography to obtain high-resolution urban remote sensing data and create a three-dimensional digital model of the city. Wind direction and speed over the past decade were analyzed in the proposed study area. Dominant wind directions were identified for both summer and winter seasons to calculate frontal area density under dominant wind directions, while an area growth method was applied with limited direction to determine urban ventilation corridors. The experimental results show two ventilation corridors under the winter-dominant wind direction and four under the summer-dominant wind direction. The results were validated against fieldwork conducted by three urban planning experts, with an error rate of 4.91% and 4.57% on a digital map. Eventually, this study will provide a tangible scientific foundation and technical support for the planning and construction of urban ventilation corridors to mitigate the heat island effect and improve air quality.

Introduction

As socio-economic development continues, the global urbanization process is accelerating, yet this rapid urbanization has introduced several potential challenges. For instance, the expansion of transport and industrial activities has significantly increased the emissions of air pollutants^1,2. Additionally, the rise in urban building density has heightened surface roughness, weakening natural wind strength and limiting air circulation, thereby reducing pollutant dispersion efficiency^3,4. Furthermore, the rapid growth in urban population and improving living standards have led to an increase in solid waste production. This waste’s incineration, landfill processes, and motor vehicle emissions threaten air quality^5,6. These factors collectively contribute to frequent haze events⁷, the deterioration of the microclimate environment⁸, the exacerbation of the heat island effect^9,10, and even severe ecological damage¹¹.

Research indicates that constructing urban ventilation corridors influences city air circulation¹². These corridors significantly improve airflow, facilitating the diffusion and dilution of pollutants¹³, resulting in a fresher urban environment. Additionally, ventilation corridors play a crucial role in mitigating the urban heat island effect^14,15,16, thereby enhancing the comfort of urban living and working environments through improved air quality¹⁷ while also strengthening the city’s ecological environment¹⁸. According to the principles of local circulation, Kress proposed that urban ventilation systems consist of three primary components: the Wirkungsraum (action space), the Ausgleichsraum (compensation space), and the Luftleitbahn (air guiding channel)¹⁹. The Luftleitbahn, functioning as a ventilation corridor that connects the Ausgleichsraum and Wirkungsraum, is characterized by sparse building distribution, low surface roughness, and high permeability. This theoretical framework establishes a systematic foundation for the construction of urban ventilation corridors, offering significant practical implications, particularly in mitigating urban heat island effects and air pollution. Consequently, the scientific and rational planning and construction of urban ventilation corridors is a key strategy for achieving sustainable urban development and enhancing residents’ quality of life.

There are various methods for constructing urban ventilation corridors, which generally encompass both macro and micro scales²⁰. At the macro scale, spatial pattern analysis based on Geographic Information System (GIS) is widely used. Most studies construct and quantitatively assess urban ventilation corridors by integrating multi-source data and utilizing spatial analysis methods like the Least Cost Path (LCP) algorithm.

Suder, Arkadiusz et al. utilized LiDAR scanning to obtain elevation data and applied morphometric methods to calculate roughness parameters. These were combined with GIS analysis techniques to identify potential ventilation paths in a Polish city. However, the study did not account for the influence of factors such as vegetation on ventilation pathways, suggesting that more detailed spatial data are required to enhance the accuracy of the analysis²¹. Liu Weiling, et al. integrated multi-source data and utilized ECMWF and Landsat 8 data as proxies to calculate the Universal Thermal Climate Index (UTCI). They employed GIS-based algorithms, including the Frontal Area Index (FAI) and Least Cost Path (LCP), to construct ventilation corridors in the Pearl River Delta region. This study quantitatively revealed the impact of urban ventilation corridors on thermal comfort and identified key influencing factors. However, the analysis did not fully account for the influence of topographic factors on the accuracy of FAI and highlighted the need to consider finer spatial scales²².

Liu, Dandan, et al. analyzed the thermal environment, urban subsurface features (e.g., building density, road length, water body length, and vegetation cover), and the urban wind environment of Kaifeng City. This analysis helped determine the compensation and action spaces, deriving the integrated ventilation potential for each region. Using meteorological analyses alongside RS and GIS technologies, two ventilation corridors were constructed based on temperature difference analysis and ventilation corridor zoning. However, the study is limited by a lack of in-depth analysis of micro-level and seasonal climatic factors²³. Shen Ying et al. proposed a GIS-based analysis method using the city center of Guangzhou as a case study. The method combines the Comprehensive Ventilation Cost Index (CVCI) with the Least Accumulated Cost (LAC) approach to analyze ventilation resistance across different building forms in a multi-level analysis. The study also dynamically observed and evaluated the city’s ventilation performance under various wind directions to identify potential ventilation corridors. However, the analysis did not consider the impact of terrain elevation on CVCI calculations, and the results focused more on ventilation probability rather than actual wind speed²⁴. Li et al. developed an urban ventilation refinement assessment (RAV) model using GIS evaluation methods. Taking Xi’an, an inland city in China, as a case study, the model integrated the heat load index with six factors related to urban spatial morphology. The least-cost path method, combined with urban wind environment and heat load maps, was used to establish six primary and four secondary corridors. However, the complexity of urban spaces limits the accuracy of microclimate simulations. To achieve a more realistic urban wind environment, parameter settings should be optimized and simulation durations extended²⁵.

At the micro-scale, ventilation corridor construction techniques have evolved from early direct observations of wind environments to the use of Computational Fluid Dynamics (CFD) simulations to model gas flow paths. Current research emphasizes a more integrated approach, combining GIS technology’s spatial analysis capabilities with simulation techniques for comprehensive analysis and optimal design.

Chang, Shouzhi et al. used CFD simulations to refine the wind environment assessment in Changchun City. They employed GIS technology for spatial analysis and integrated both approaches to establish five ventilation corridors at the urban scale. However, further research is required at the neighborhood and building scales to create a cohesive system with ventilation corridors as individual units²⁶. Gu, Kangkang et al. developed a Ventilation Corridor Planning (VCP) model for the comprehensive assessment and planning of urban ventilation corridors in Bozhou City, Anhui Province, China. The model utilizes a multi-criteria urban climatology evaluation method combined with GIS technology. It effectively integrates the macro-scale urban atmospheric boundary layer and the micro-scale urban canopy.

However, further empirical studies are needed to verify the model’s applicability across various urban environments and climatic conditions²⁷. Zhu et al. studied Hangzhou’s Qiantang River waterfront area, using GIS technology and specialized software (e.g., ENVI, PHOENICS, CADNA/A) to simulate wind, thermal, and acoustic environments quantitatively. The study identified physical environmental issues such as urban heat islands, cold islands, calm wind zones, weak wind zones, and noise pollution areas. It proposed a methodology for urban spatial form layout and multi-scale optimization based on these simulations, developing a framework that includes physical environment status simulation, coupled optimization, and design enhancement. However, further evaluation is needed to assess the long-term effectiveness of this approach and its applicability across different urban environments²⁸.

In conclusion, current methods for constructing urban ventilation corridors exhibit certain limitations. Although GIS technology is widely utilized on a macro scale, its accuracy is often limited by data source precision, data collection methods, and algorithm reliability. These limitations hinder the effective visualization of the three-dimensional morphology and detailed characteristics of urban landscapes. On a micro-scale, while CFD simulation technology offers advantages such as high accuracy, speed, and cost-effectiveness compared to traditional field surveys and wind tunnel experiments, it is still undergoing refinement and development. The accuracy and applicability of CFD in real-world scenarios require further verification and assessment.

This study employs oblique photogrammetry, which, unlike conventional GIS techniques, captures multi-angle imagery to generate high-resolution, three-dimensional remote sensing data at both street and building scales. After preprocessing, a Digital Elevation Model (DEM) and a Digital Surface Model (DSM) were constructed, with elevation data extracted from their difference. A wind rose diagram was derived from a decade-long analysis of wind direction frequencies within the study area, enabling the identification of dominant wind directions. By incorporating the dominant wind directions and adopting the frontal area density (FAD) of buildings as a critical indicator, this study introduces a direction-constrained region-growing algorithm to construct urban ventilation corridors.

Compared to previous methods, this study introduces several key innovations. To begin with, it employs UAV oblique photogrammetry instead of traditional Geographic Information System (GIS) techniques, enabling the cost-effective and rapid acquisition of multi-angle imagery. This method provides high-precision, three-dimensional data at both architectural and street scales. Following digitization, a Digital Elevation Model (DEM) and a Digital Surface Model (DSM) of the city are constructed, from which urban elevation data are derived based on the difference between the DEM and DSM. In addition, the proposed method incorporates a direction-constrained mechanism, which sets it apart from conventional region-growing algorithms. By calculating wind direction frequencies, a wind rose diagram is generated to determine the city’s dominant wind directions. The direction-constrained region-growing algorithm is then applied to extract urban ventilation corridors along these directions, effectively overcoming the limitations of low accuracy and insufficient scientific rigor found in previous studies. Moreover, this study integrates multi-source data—including remote sensing imagery, meteorological records, and building height data—to facilitate the automatic extraction of ventilation corridors through an efficient algorithm. This comprehensive approach significantly enhances both the practicality and precision of the corridor excavation process.

This method effectively resolves the issues of low precision and limited applicability prevalent in existing research. Empirical validation conducted in Nanhu District, Jiaxing City, Zhejiang Province, demonstrates the robustness and reliability of the proposed approach.

The proposed methodology

UAV inclined photography and urban elevation data

Drone tilt-photography technique

This study focuses on the excavation of urban ventilation corridors in the Nanhu District of Jiaxing City. High-resolution Digital Surface Models (DSMs) and Digital Elevation Models (DEMs) were generated using datasets derived from oblique photogrammetry obtained through advanced multi-angle image processing techniques.

The oblique imagery data was acquired using an AMC5100 airborne oblique camera system mounted on an Unmanned Aerial Vehicle (UAV). The AMC5100 is equipped with five high-resolution cameras, each featuring a 100-megapixel sensor capable of simultaneously capturing multi-directional oblique images. In contrast to conventional vertical photography, this technology captures imagery from both a vertical perspective and four oblique angles, enabling the generation of geospatial data with enhanced realism and three-dimensionality. The flight altitude determines the ground sampling distance (GSD) in oblique photogrammetry and can be calculated using the following formula:

$$GSD = \frac{h \cdot a}{f}$$

(1)

where $h$ represents the flight altitude, $a$ denotes the pixel size, and $f$ is the focal length of the camera. As illustrated in Fig. 1, the GSD exhibits a linear relationship with flight altitude, with lower altitudes yielding higher spatial resolution. This multi-angle imaging approach significantly enriches the dimensionality of map data while also providing more precise and comprehensive spatial information for applications such as terrain analysis, urban planning, and three-dimensional modeling²⁹.

Fig. 1

Schematic diagram of image resolution calculation.

Digital elevation model and digital surface model

The DEM is a digital representation based on terrain elevation data and serves as a fundamental branch of the broader Digital Terrain Model (DTM). A DTM is typically defined as a comprehensive model that incorporates elevation data and other geomorphological features, such as slope, aspect, and the rate of slope change. In contrast, a DEM is a first-order, simple digital model focusing solely on elevation, providing the essential groundwork for more complex geomorphological analyses. Various geomorphological parameters can be derived by extensive mining and expanding DEM data, offering enhanced information for terrain analysis and related applications³⁰.

The Digital Surface Model (DSM) is a ground elevation model that includes the heights of surface features such as buildings, bridges, and trees. Unlike the Digital Elevation Model (DEM), which contains only the elevation information of the terrain itself, the DSM extends beyond this by incorporating the elevation data of all surface features above the ground.

In this study, DSM and DEM models are constructed using UAV tilt camera technology, following specific operational steps. First, a suitable UAV and tilt camera are selected, ensuring their performance and parameters meet the requirements for aerial photography. Detailed route planning is then conducted to define a flight path and an aerial photography area of the UAV. After thorough calibration, the aerial photography mission is carried out, maintaining the drone’s stability during flight and adhering to the pre-established route to capture a large volume of high-definition, high-precision image data. Once acquired, the images undergo pre-processing, including geographic correction, radiometric correction, border removal, aberration correction, and image chunking. Images captured from various angles are aligned to enhance the efficiency and accuracy of subsequent processing. The 3D model is then constructed from the image data and refined through texture mapping and surface smoothing to improve the model’s quality and visual appeal. Finally, the city’s DSM and DEM are imported into Global Mapper software, where elevation information for buildings, vegetation, and other features is extracted through the difference between DSM and DEM. The workflow for generating urban DSM and DEM models using UAV tilt photography technology is illustrated in Fig. 2 below.

Fig. 2

Workflow for generating urban DSM and DEM Models Using UAV Tilt Photography Technology.

Following the abovementioned procedures, DSM and DEM images for Nanhu District in Jiaxing City were generated, as illustrated in Fig. 3.

Fig. 3

Urban digital model of Nanhu District, Jiaxing City. Generated from UAV oblique photography (DJI Phantom 4 RTK) using Pix4D mapper (v4.4.12; https://www.pix4d.com).

Building height data

By analyzing the elevation data of various features, the results in Fig. 4 show that 98.47% of urban buildings in Nanhu District, Jiaxing, are 25 m or less.

Fig. 4

Height distribution of urban buildings in Nanhu District, Jiaxing City.

Analysis of Urban wind environment

Overview of the target study area

This study focuses on Nanhu District, Jiaxing City, Zhejiang Province, as the research area. Nanhu District is located in the northeastern part of Zhejiang Province, at the core of the Hangzhou–Jiaxing–Huzhou Plain within the Yangtze River Delta. It is geographically bounded by Shanghai to the east, Hangzhou to the west, Suzhou to the north, and Hangzhou Bay to the south. The district features low-lying, flat terrain with an average elevation of 3.7 m. The region, located at the southern edge of the northern subtropical zone, experiences an East Asian monsoon climate characterized by distinct seasonal shifts between winter and summer monsoons. The climate is marked by moderate temperatures, abundant rainfall, and ample sunshine, with an average annual temperature of 15.7 °C, an average annual precipitation of 1179 mm, and an annual evaporation rate of 1,301 mm. The district encompasses a total land area of 426 km², including 22,773 hectares of arable land. It has a resident population of approximately 878,000 and an urbanization rate of 88%³¹.

Wind direction frequency map and dominant wind direction

Wind frequency refers to the proportion of occurrences of a specific wind direction relative to the total number of observations (including calm periods) within a given time period, expressed as a percentage. The formula for calculating wind frequency is:

$${\text{g}}_{n} = \frac{{f_{n} }}{{(c + \sum {_{n = 1}^{16} f_{n} } )}}$$

(2)

Equation (2) ${\text{g}}_{n}$ represents the wind frequency in the $n$-direction, $f_{n}$ denotes the number of observations of winds in the $n$-direction during the statistical period, and $c$ indicates the number of observations of calm winds during the same period. A wind frequency chart, also known as a wind rose, is a meteorological statistical diagram that quantifies the frequency of various wind directions in a specific area over a defined period. The chart uses a scaling factor (i.e., the radius value in the polar coordinate system) based on the frequency of winds from each direction. Larger radius values indicate higher frequencies for those directions. These data are plotted on a polar coordinate chart with eight or sixteen azimuths. The endpoints of adjacent directions are connected with straight lines, creating a closed, rose-like shape known as a wind rose chart. The wind direction and speed data employed in this study were sourced from multiple observational stations operated by the Jiaxing Meteorological Bureau, comprising 24-h daily observations over the past decade. Statistical analysis of these records enabled the generation of wind rose diagrams for 16 azimuths, as depicted in Fig. 5a and b.

Fig. 5

The wind rose diagram of Nanhu District, Jiaxing City.

The dominant wind direction refers to the prevailing wind orientation within a specific area and time period. It typically covers a continuous range of approximately 45° and should account for at least 30% of the total occurrences to be considered a significant component of the overall wind distribution. As shown in Fig. 5a, the dominant wind direction in winter in Nanhu District, Jiaxing City, is around 135°. Accordingly, this study will focus on subsequent analyses within the range of 135° ± 32°. Similarly, Fig. 5b indicates that the dominant wind direction in summer is primarily centered around − 4°, and therefore, more detailed analysis and discussion will be conducted within the range of − 4° ± 31.9°.

Calculate frontal area density

Frontal Area Density (FAD) of urban buildings refers to the ratio of a building’s windward area in a specific wind direction to the area it occupies at a given height increment. This metric is commonly used to evaluate the ventilation potential of an area³². The formula for calculating FAD is as follows³³:

$$\lambda_{f(z,\theta )} = \frac{{A(\theta )_{proj(z)} }}{{A_{T} }}$$

(3)

In Eq. (3), $\lambda_{f(z,\theta )}$ represents the FAD in direction $\theta$, $A(\theta )_{proj(z)}$ denotes the projected area of the building’s windward side in direction $\theta$, $A_{T}$ is the area of the site where the building is located, and $z$ indicates the height range used for calculating the projected area. The mathematical analysis shows that ventilation resistance decreases in regions with a relatively low index, resulting in higher ventilation potential.

This study employs a raster-based computational model to calculate the FAD of urban buildings³⁴. The process begins with standard image processing algorithms to obtain the unit normal vector for each pixel in the model’s surface data. The surrounding triangular surface structure is constructed using the octant subdivision method. The normal vectors of these triangles are then averaged outward using the pixel as a reference point and weighted by their respective areas to determine the normal vector at the pixel’s Centre. Next, the dot product between the pixel’s normal vector and a horizontal unit vector in a specified wind direction is calculated, revealing the pixel’s projection in that direction. Given the unique characteristics of uneven terrain, the windward condition is typically identified when the angle between these two vectors exceeds 90°. Consequently, the windward area density for a particular direction is obtained by summing all pixels with negative dot product results. The windward area density distribution for Nanhu District in Jiaxing City during winter and summer has been calculated, as illustrated in Fig. 6.

Fig. 6

Windward area density of Nanhu District, Jiaxing City. Processed using Global Mapper 16.0.5 (https://www.bluemarblegeo.com) for DSM-DEM elevation extraction and MATLAB R2019b (https://www.mathworks.com) for building height calculation and FAD computation. Analysis performed with grid resolutions of 50–100 m based on dominant wind directions from local wind rose diagrams.

Research has demonstrated a clear negative correlation between windward area density and wind speed ratio within the vertical height range of 0 to 15 m. Specifically, the wind speed ratio decreases as windward area density increases, reducing ventilation potential. When the windward area density exceeds 0.6, the wind speed ratio drops below 0.1, indicating poor natural ventilation conditions in urban areas. Conversely, when the windward area density is below 0.35, the wind speed ratio rises above 0.2, suggesting a more favorable natural ventilation environment³⁵.

Definition of urban ventilation corridors

Urban ventilation corridors are airflow pathways composed of urban open spaces with low aerodynamic roughness and minimal airflow resistance. The construction standards for urban ventilation corridors are still under discussion and refinement. Kress was the first to propose these standards, which include a surface roughness of less than 50 cm, a wind path length of more than 1000 m, and a width greater than 50 m; the total area of obstacles within the corridor should not exceed 10% of the corridor’s width, with a height limit of 10 m; and strict regulations should be enforced to prohibit the construction of buildings or tall vegetation within the corridor¹⁹. The China Meteorological Administration (CMA) has established guidelines for the identification and classification of urban ventilation corridors, including standards for the primary and secondary corridors³⁶:

In urban planning, the design of ventilation corridors should prioritize enhancing the city’s ventilation efficiency and mitigating the urban heat island effect. Primary ventilation corridors should span the entire urban area, strategically aligned with zones characterized by low surface roughness and high ventilation potential, connecting the city center with suburban areas, particularly those with higher ventilation capacity and elevated heat island intensity. Urban infrastructure, such as major thoroughfares, rivers, and parks, should be fully integrated into land use planning as the structural framework for these corridors. The design of primary corridors should align with the prevailing regional wind direction, maintaining an angle deviation of no more than 30°, with a minimum width of 200 m and a ventilation potential index of at least 3. Ideally, these corridors should exceed 500 m in width and have a ventilation potential index of at least 4 to ensure effective connectivity between high-ventilation zones and areas with intense heat islands.

Secondary ventilation corridors are strategically planned supplementary structures in regions with high ventilation potential. These corridors connect green spaces with densely built-up areas and adjacent zones that exhibit significant differences in ventilation capacity. The design of secondary corridors should enhance the effectiveness of the primary corridors, thereby improving overall ventilation efficiency. For optimal performance, secondary corridors should maintain an angle of less than 45° relative to the local prevailing wind direction, with a minimum width of 80 m, and the width of obstacles should not exceed 10% of the corridor width. These corridors should extend over 2000 m and have a ventilation potential index of at least 3. In a suboptimal scenario, the secondary corridors should have a minimum width of 50 m, with obstacles covering no more than 20% of the corridor width, extending over 1000 m with a ventilation potential index of at least 2. This study integrates established practices for urban ventilation corridors³⁷ with the specific characteristics of Nanhu District in Jiaxing City. Based on this integration, the following empirical standards have been proposed for urban ventilation corridors tailored to local conditions:

• (1) FAD ≤ 0.35.

• (2) Corridor length ≥ 1500 m.

• (3) Corridor width ≥ 50 m.

Ventilation corridor extraction methods

Region growing algorithm

The region-growing algorithm is an image segmentation technique based on the principle of image homogeneity^38,39. The core idea involves starting with a set of seed points within a region $R_{i} ,i = 1,...,M - 1$. Pixels similar to these seed points are merged into the same region, creating new ones. This process continues until no further pixels can be merged with the region. Ultimately, this results in forming distinct, non-overlapping regions or objects.

$$S = \bigcup\limits_{i = 1}^{M} {R_{i} } ,{\text{ where}}\;R_{i} \cap R_{j} = \emptyset \;{\text{for}}\;i \ne j$$

(4)

The basic steps of the region-growing algorithm are as follows^40,41:

Based on the characteristics of the existing image, select one or more initial seed points that meet the following criteria:

$$A = \{ a_{1} ,...,a_{M - 1} \} \subset I$$

(5)

• (1) Define a growth criterion $RG(\psi )$ for the segmentation process, such as the FAD value of the pixel. Neighboring pixels are only incorporated into the current region if their FAD values fall within a specified range.

• (2) From the seed points, examine the characteristics of adjacent pixels. If these neighboring pixels meet the growth criterion, they are added to the current region, and the newly added pixels become new seed points.

• (3) Repeat step (2) until no adjacent pixels around any seed points meet the growth criterion or until a predetermined number of iterations is reached and the algorithm terminates.

• (4) The resulting segmented image can be represented as:

  $$S(I,RG(\psi ),\varsigma ) = \bigcup\limits_{i = 1}^{M} {R_{i} }$$

  (6)

In Eq. (6), $I$ represents the image to be processed, $RG(\psi )$ denotes the region growing algorithm with a specific growth criterion $\psi$, and $\varsigma$ refers to the initial seed point for the growth algorithm.

Urban ventilation corridor extraction algorithm

When designing the urban ventilation corridor mining algorithm, it is essential to incorporate a mechanism that restricts the direction of growth, which is not typically present in conventional region-growing algorithms. Additionally, the algorithm should be optimized to align with the city’s dominant seasonal wind direction. The process of the urban ventilation corridor mining algorithm based on directional region growing is outlined as follows:

• (1) First, select the initial seed point with coordinates $(x_{0} ,y_{0} )$ and define the growth range, including the base growth angle and the offset angle.

• (2) Begin the traversal from points with coordinates differing by 1 unit in the horizontal or vertical direction from the initial seed point. For each candidate growth point with coordinates $(x_{i} ,y_{i} )\left( {i = 1,2,...,n} \right)$, check if the line formed between the new growth point and the initial seed point falls within the defined growth direction range. If so, the candidate point is eligible for growth.

• (3) Compare the FAD values of the candidate growth point. If the FAD value is below the growth threshold K, merge the point into the region: $f(x,y) < K$, and update the seed point to include the newly merged pixel.

• (4) Repeat step (3) until all candidate growth points have FAD values below the threshold K or until the algorithm reaches the predetermined number of iterations.

An example of region growth in a limited direction is shown in Fig. 7. Figure 7a shows the original image, with numbers indicating the FAD values of the pixels. The growth criterion is that the FAD value of the point to be grown is ≤ 3.5. The pixel point f(3,2), whose FAD value is within the threshold range, is selected as the initial growth point. Figure 7b shows the first stage of region growth. Among the seed points with a difference of 1 in horizontal/vertical coordinates from the initial point, only f(3,3) is merged, as its FAD value satisfies the threshold condition and is within the angular range of the limited direction. Figure 7c illustrates the second stage of region growth, where f(3,4) and f(4,4) are merged. Figure 7d depicts the third stage of region growth, with f(3,5) and f(4,5) merged. At this point, no more pixel points satisfy the growth criterion, and the growth process stops.

Fig. 7

Urban ventilation corridor mining algorithm process.

Methods for automated urban ventilation corridor extraction

In this paper, the tilt photography technique is utilized to obtain the difference data between DSM and DEM, which is then used to calculate the height of urban buildings. The dominant wind direction in summer and winter is determined by analyzing the urban wind direction data, and the windward area density under the dominant wind direction is calculated to create an FAD map. Subsequently, the urban ventilation corridors are identified using the area growth method with limited direction.

• (1) Set the length and width of the corridor with a growth threshold of ≤ 0.35. Define the growth range as the dominant wind direction angle ± offset angle, denoted as $\alpha \pm \theta$

• (2) Using the FAD map, perform preliminary excavation of the corridors along the dominant wind direction with the direction-constrained region growing method, which yields $w$ corridor clusters. Subsequently, clustering techniques based on characteristics such as the inclination angle of the corridor clusters are applied to group these $w$ clusters $m_{1} ,m_{2} ,...,m_{n}$, ultimately resulting in $n$ corridors (usually $w > > n$).

• (3) Individually excavate each of the $n$ corridors. Select the starting coordinates $\left( {x_{1} ,y_{1} } \right),\left( {x_{2} ,y_{2} } \right),...,\left( {x_{n} ,y_{n} } \right)$ for each regional growth and perform direction-constrained region growing based on both the winter and summer dominant wind directions.

• (4) Merge the $n$ excavation results to obtain the complete urban ventilation corridors. The detailed method is illustrated in Fig. 8.

Fig. 8

Automated method for excavating urban ventilation corridors.

Results and discussion

Excavation results of ventilation corridors

This paper uses tilt photography to capture aerial data for Nanhu District, Jiaxing City. DSM and DEM maps are then generated, and the difference between DSM and DEM is calculated to determine the city’s elevation data. The wind rose diagrams based on 16 wind frequency directions are produced using the wind direction and speed data from the past 10 years in Jiaxing City, allowing for the analysis of dominant wind directions in summer and winter. Subsequently, the FAD map under the dominant wind direction is derived and used to identify and analyze ventilation corridors.

For the winter’s dominant wind direction, ventilation corridors with lengths ranging from 1500 to 5000 m were pre-excavated. The corridors were designed with a minimum width of 50 m and an FAD value ≤ 0.35, and the excavation was conducted within the dominant wind direction range of 135° ± 34°. The results of this pre-excavation are illustrated in Fig. 9.

Fig. 9

Pre-excavation results of 1500–5000 m ventilation corridors. Identified using direction-constrained region growing algorithm implemented in MATLAB R2019b (https://www.mathworks.com).

A 4000 m corridor with a more distinct distribution was selected for further excavation by comparing pre-excavation images of corridors of various lengths. The 2124 corridor clusters from the preliminary excavation were then clustered, resulting in the identification of 2 ventilation corridors, as depicted in Fig. 10.

Fig. 10

Excavation results of the 4000-m corridor. The identified ventilation corridors (length ≥ 4000 m) were visualized by MATLAB R2019b and overlaid on DSM color maps. Data processing and visualization performed using MATLAB R2019b (https://www.mathworks.com).

Finally, the excavated ventilation corridors were overlaid onto the original city’s DSM map, as illustrated in Fig. 11.

Fig. 11

Overlay of ventilation corridors and DSM. Processed using MATLAB R2019b with Gaussian smoothing and overlaid on DSM data. Shows distribution of primary ventilation corridors (length ≥ 4000 m). Software: MATLAB R2019b (https://www.mathworks.com).

Analysis of ventilation corridor excavation results

Based on the excavation results, two urban ventilation corridors, each measuring 4000 m in length and 50 m in width, have been identified under the dominant winter wind direction, as shown in Fig. 12.

Fig. 12

Excavation results of ventilation corridors under winter dominant wind directions. Processed using MATLAB R2019b with corridor smoothing and validated by field expert evaluation. Shows ≥ 4000 m ventilation corridors under winter prevailing wind conditions. Software: MATLAB R2019b (https://www.mathworks.com).

The analysis reveals the following: (1) The first corridor extends along the Beijing-Hangzhou Canal, shifts to follow the Ring City River, and continues southeast along the Changzhong Harbour after passing Nanhu Lake. (2) The second corridor begins at Wangwumen, extends eastward, and then shifts southeast after passing Jiaxing Station, continuing in that direction.

Under the dominant summer wind conditions, four urban ventilation corridors, each satisfying the criteria of a minimum length of 2500 m and a minimum width of 50 m, were excavated, as illustrated in Fig. 13.

Fig. 13

Excavation results of ventilation corridors under summer dominant wind directions. Processed using MATLAB R2019b with corridor smoothing and validated by field expert evaluation. Shows four identified ventilation corridors (length ≥ 2500 m) under summer prevailing wind conditions. Software: MATLAB R2019b (https://www.mathworks.com).

From the excavation results, it can be analyzed that:

• (1) The first wind corridor starts at People’s Park, follows the Ring City River, passes through Wangwumen to the northwest, and extends to the Yuehe Historical Quarter.

• (2) The second wind corridor begins at Jiaxing Zicheng Ruins Park. The northern section extends westward through the Beijing-Hangzhou Canal and then north to Jieyan Xincun, while the southern section continues westward through Mingyue Park to Jiyang Park.

• (3) The third wind corridor starts at Fan Li Lake Park. The northern section crosses the Beijing-Hangzhou Canal, extending northwest to Mingyue Park, while the southern section runs westward along Hangzhou Tang, reaching the vicinity of Jiaxing College.

• (4) The fourth wind corridor begins north of the Jiaxing Municipal Government. The northern section extends northwest through Southwest Lake, passes Fan Li Lake Park, and reaches the vicinity of Mingyue Park, while the southern section passes through Southwest Lake Eco-Park, continues through the southern part of Southwest Lake and Fuanghe Zhou, and extends southwest to the vicinity of the southern area of Jiaxing Nanhu College. Three urban and rural planning experts at Zhejiang University of Technology were invited to manually sketch the ventilation corridors based on specific conditions without knowledge of the systematic excavation results. The results demonstrated an average overlap rate of nearly 95%, as shown in Table 1, demonstrating the effectiveness of the proposed research method.

Table 1 Comparison of system excavation results and experts’ recommendations.

Complexity analysis of the ventilation corridor extraction algorithm

This study introduces a novel algorithm for extracting urban ventilation corridors based on a direction-constrained region-growing method. The algorithm comprises two core steps.First, the algorithm grows regions along specific directions. The time complexity of this process is primarily determined by the number of growth points ($k$) and the neighborhood checks for each point ($m$), expressed as O($k \cdot m$). Second, the initially extracted ventilation corridor clusters are further clustered and merged to generate the final ventilation corridor network. The time complexity of this step is O($p\log p$), where $p$ represents the number of ventilation corridor clusters. The overall computational complexity of the algorithm is O($k \cdot m + p\log p$). To assess its practical efficiency, experiments were conducted on datasets of varying scales. The results are summarized in Table 2.

Table 2 Computational performance of the ventilation corridor extraction algorithm on datasets of varying scales.

As can be seen from the table, even on large-scale datasets, the algorithms are still able to complete the computation in a relatively short time period and meet the needs of practical applications.

Discussions

Remote sensing data

The remote sensing data for this study were obtained from UAV tilt photography. Unlike traditional GIS techniques, tilt photography captures images from multiple angles, providing more precise and three-dimensional map data at street and building scales. By digitizing these multi-angle images, accurate DEM and DSM of the city are created, offering reliable elevation data crucial for developing urban ventilation corridors.

Improvement in urban ventilation environment

This paper introduces an efficient and accurate technique for excavating urban ventilation corridors by integrating remote sensing technology with a direction-constrained region-growing algorithm. The increasing building density and surface roughness during urbanization reduce natural wind strength and restrict air circulation, which impedes the efficient dispersion of pollutants. In order to address this challenge, this study constructs urban ventilation corridors to enhance airflow within the city by effectively facilitating the dispersion and dilution of air pollutants. The proposed approach has significant scientific and practical implications for mitigating the urban heat island effect and improving air quality. Besides, this research advances the precision of ventilation corridor planning and enriches the theoretical framework of urban climatology and urban planning. Henceforth, the findings can be directly applied to urban planning and environmental management, providing a scientific basis and technical support for creating more livable urban environments.

Implementation strategies and technical challenges of urban ventilation corridors

This study identified two urban ventilation corridors under dominant winter wind conditions and four corridors under dominant summer wind conditions within the study area. The findings of this research have been submitted to the relevant departments of Jiaxing City as a key reference for urban planning. It is recommended that these ventilation corridors be integrated into the city’s master plan to enhance the urban ventilation environment and mitigate the heat island effect and air pollution.

In the actual planning of urban ventilation corridors, a multi-objective, synergistic planning approach can be adopted. The construction of these corridors should align with the goals of mitigating the heat island effect, improving air quality, and developing green infrastructure networks. For instance, integrating ventilation corridors with urban greenways and water networks can maximize ecological and environmental benefits.

In addressing potential challenges during implementation, such as land-use changes due to evolving urban planning or the obstruction of existing ventilation corridors caused by increased building density, the current plan may face issues related to rigidity and an inability to adapt to long-term climatic variations or shifts in urban demand. Therefore, it is essential to incorporate the dynamic nature of ventilation corridors into the planning process. Green spaces, water bodies, or low-density development areas should be reserved within these corridors to meet ventilation needs while also allowing for flexibility in adapting to future functional changes.

Conclusion

This paper addresses the growing challenges of the urban heat island effect and air quality issues associated with urbanization. High-precision urban remote sensing data from Nanhu District, Jiaxing City, Zhejiang Province, were obtained using UAV oblique photography. A Digital Elevation Model (DEM) and Digital Surface Model (DSM) of the city were established, with urban elevation data derived from the difference between the DEM and DSM. A wind rose map was created by calculating wind direction frequencies to identify the city’s dominant wind directions. A direction-constrained area growth method was proposed to extract urban ventilation corridors utilizing the density of the windward area of buildings as an indicator. This method successfully identified two winter and four summer corridors within the study area. The findings provide a scientific foundation and technical support for the planning and construction of urban ventilation corridors, which are crucial for mitigating the urban heat island effect and improving air quality.

Limitations and future work

In this study, the windward area density of buildings was used to assess the ventilation potential of the area. However, the influence of vegetation and other subsurface factors on the wind environment was not sufficiently incorporated into the calculations, which may cause discrepancies between the actual performance of the ventilation corridors and the simulation results.

Future research will further explore the following aspects. First, a more thorough consideration of the impact of vegetation, surface roughness, and other subsurface factors on ventilation corridors to enhance model accuracy. Second, the integration of additional field measurements to calibrate and validate the model, ensuring its applicability across various urban environments. Finally, the integration of CFD simulation technology, utilizing fluid dynamics principles and numerical methods to solve the governing equations of gas flow, enables the prediction and analysis of gas flow parameters and characteristics, which helps to validate and optimize ventilation corridor designs.