A Data-Driven Framework to Identify Tree Planting Potential in Urban Areas: A Case Study from Dortmund, Germany

Abstract

Urban areas increasingly face heat-related climate risks, necessitating targeted, nature-based interventions such as tree planting to improve resilience, livability, and public health. This study presents a data-driven workflow to identify urban tree planting potential (TPP) in the city of Dortmund, Germany. The approach integrates high-resolution spatial datasets capturing land cover, shading, thermal comfort, population density, and critical infrastructure. All variables were harmonized within a 50 m hexagonal grid, normalized, and combined into a composite TPP score using weighting schemes informed by expert judgment and sensitivity testing. Spatial and non-spatial clustering were applied to group urban areas by shared characteristics, and a connectivity analysis evaluated the spatial coherence of high-potential cells and their relationship to existing green infrastructure. The findings demonstrate the potential to strengthen urban green infrastructure and guide coordinated planting strategies while addressing both ecological and social priorities. The presented workflow offers a flexible, transferable tool to support municipalities in prioritizing effective greening interventions and integrating climate adaptation objectives into urban development planning.

1. Introduction

Urban environments are increasingly vulnerable to the impacts of climate change, with rising temperatures, more frequent heatwaves, and deterioration of air quality that pose significant threats to human health and overall urban livability [1,2,3,4]. Sustainable urban development is essential to maintain urban environments as livable spaces [5,6]. Hence, city administrations have committed to achieve defined sustainability goals and to develop resilient urban spaces using dedicated actions [7,8]. Data-driven strategies are implemented to address growing cross-sectorial challenges, such as the transformation of energy and mobility systems [9,10,11,12], environmental education, efficient and resilient supply chain management [13], waste management [14], strengthening of health infrastructure, and community engagement [15,16]. Among strategies to mitigate the impacts of longer and more intense heat waves, increasing urban green infrastructure (UGI) has emerged as a highly effective, nature-based solution [17,18,19].

UGI provides a wide range of ecosystem services that are vital to strengthening urban resilience and improving the livability of urban spaces [17,18,20]. These services include temperature regulation through shading and evapotranspiration, air purification by filtering pollutants, stormwater management through improved infiltration and reduced runoff, and noise reduction [21,22]. In addition, green spaces promote biodiversity by providing habitats for various species [23,24,25]. Furthermore, UGI contributes to the mental and physical well-being of urban residents by offering recreational and health opportunities and enhancing aesthetic value. The integration of UGI into urban planning plays a crucial role in fostering healthier, more livable, and more sustainable cities. Urban trees and forests, as large parts of UGI, contribute significantly to the improvement of ecosystem services in urban environments. In particular, trees and forests contribute to the physical and mental well-being of city residents [26,27,28]. Despite well-documented benefits, maintaining and expanding urban tree cover is a challenging task due to cost and effort for maintenance and the increasing pressure on urban spaces [29]. In addition, changes of temperature and precipitation patterns due to climate change are increasing risk factors for tree health and survival potential [30]. However, the socioeconomic cost of not maintaining and expanding the urban tree population might be even higher for a city administration [31]. Dedicated tree protection laws are effective, although slightly different between countries, regions, and cities [32]. In general, it is legally required that trees that are being removed due to development be replaced. However, finding suitable locations for replanting trees often proves challenging in a dense urban environment. Limited available space, competing land-use demands, and urban surface properties further complicate efforts to restore lost tree cover [29,33]. Hence, a significant challenge lies in the identification of suitable locations for new tree plantings, particularly on the scale of entire city systems. Urban environments are complex, multifactorial, and functional structures where the tree planting potential is influenced by many factors, such as variations in existing green infrastructure, demand for increased mobility space, and socioeconomic disparities. Thus, there is a pressing need for systematic and data-driven methods to evaluate and prioritize potential planting sites in urban landscapes.

A variety of methods have been developed to assess tree planting potential in urban environments. Recent studies have applied GIS-based suitability analyses and decision-support frameworks to identify favorable sites. Examples include remote sensing and distance analyses in Munich [34], thermal hotspot prioritization in Mediterranean cities [35], and combined socio–climatic assessments in Boston [36] and Mexico City [37]. Other approaches extend the scope of decision support by optimizing planting according to solar radiation [38] or developing algorithms to identify suitable locations for green roofs and roadside trees [39]. Earlier contributions laid important groundwork: Wu et al. (2008) introduced a transferable GIS-based method for locating plantable sites in Los Angeles, while Kirnbauer et al. (2009) developed a prototype decision support system integrating spatial placement and feasibility criteria [40,41]. Varol et al. (2019) combined remote sensing and GIS classification to delineate plantable areas in Turkey [42]. Comparing priorization frameworks, Nyelele et al. (2022) found that while these studies demonstrate the value of spatial and decision-support approaches, many are either context-specific, North American in focus, or rely on specialized software not easily transferable to European planning practice [43].

The present study was carried out in the city of Dortmund, the largest city in the densely populated Ruhr Region in Germany. With this status, Dortmund takes on a leading role for cities located in the Ruhr Region in tackling the challenges posed by climate change. Among other measures, a dedicated master plan on climate adaptation action was designed in 2021 [44]. This master plan includes integrated microscale climate analyses with socioeconomic needs and vulnerabilities for all Dortmund city districts and serves as a guideline for a coordinated catalog of measures addressing different spatial segmentation levels to make the city more resilient to climate hazards. As a dedicated authority, the environmental office supervises and coordinates infrastructure development to guide the way to reaching the climate adaptation and mitigation targets set in the regional and national policies and plans. The need for data-driven tools and services to support the development goals defined in the policies is pressing [45,46].

The increase in trees and tree patches is mentioned as a suitable measure to mitigate heat [44]. However, practical planning support to increase the efficiency of tree plantings, including existing green infrastructure and population demand, has not been provided. An increase in momentum for data-driven approaches was gained through the grant of two ICLEI (Local Governments for Sustainability) Action Fund 2.0 (AF2) projects, of which the latter forms the framework of the present study. The CATCH4D project and the Data2Resilience project both aim to improve heat resilience in Dortmund utilizing remote sensing, ground observations, and other data integrated by a broad method spectrum including qualitative measures and GIS. AF2 projects are designed and funded to develop pilot services for urban stakeholders in an approach to bring together state-of-the-art science and urban practice. It also encourages close collaboration with the respective city administration to ensure practicability and sustainability of the services and tools developed.

Within the framework of the Data2Resilience project, the present study introduces a data-driven approach to assess potential tree planting sites in Dortmund. Potential tree planting locations are identified using very high-resolution spatial data of existing green infrastructure, the built-up environment and infrastructure, population estimates, and thermal comfort. We applied weighting and clustering methods to investigate the influence of different parameters on the suitability of a potential planting location. To take into account overarching goals, such as the connection of tree patches for increased ecosystem services, we performed a connectivity analysis that included existing tree patches in Dortmund. We used open data sources to ensure the transferability of our approach to other cities in Germany. The results are recommended to be used as data-driven decision-making support for practitioners in urban administrations.

2. Materials and Methods

2.1. Study Area

The study was carried out for the city of Dortmund (51°30′ N, 7°28′ E). Dortmund is located in western Germany (North Rhine-Westphalia). Once known for its coal mining activities and heavy industry as part of the Ruhr-Region, the city underwent a transformative shift, pushing a greener landscape and evolving into a center for information technology, research, and logistics. However, heavy industry still has left a mark on the city’s layout and neighborhood development. Considering the challenges that climate change poses to urban spaces, this legacy creates an additional level of difficulty in the maintenance and creation of attractive public spaces. Tree coverage in Dortmund is estimated at around one third of the city area, with 33% according to Google’s Environmental Insights Explorer and 37% reported by the Husqvarna Urban Green Space Insights, corresponding to 92 km² and 103 km², respectively [47,48]. In a national comparison of large German cities, Friends of the Earth Germany (BUND) estimates that Dortmund has 178 street trees per square kilometer, ranking 14th among larger cities nationwide [49]. To improve this situation, the 2023 Ecological Forest Management Plan for Dortmund’s City Forest integrates future climate projections, recommends climate-resilient tree species, and introduces monitoring concepts to support both ecological and economic forest management [50]. In addition, targeted planting campaigns, pilot projects on climate-resilient species, and the use of sensor-based irrigation reflect Dortmund’s efforts to address current shortcomings and prepare for future challenges in urban tree management [51,52,53].

2.2. Data

The data used in this study are summarized in

Table 1. Overview of the datasets used for assessing the TPP in Dortmund, Germany.

Variable	Format	Resolution	Year	Source
Building footprints	Vector	-	2019	ALKIS [60]
DSM	Raster	0.5 m	2019	GeoportalNRW [55]
Tree canopy layer	Raster	0.2 m	2020–2023	Google EIE [47]
Population	Vector	-	2018	CLMS [57]
Road network	Vector	-	2018	CLMS [58]
Urban Atlas LULC	Vector	-	2018
Short-term care facilities	Vector	-	2024
Schools	Vector	-	2024
Youth clubs	Vector	-	2024
Playgrounds	Vector	-	2024
Sights	Vector	-	2024	Dortmund Open Data [56]
Museums	Vector	-	2024
Hospitals	Vector	-	2024
Rehabilitation clinics	Vector	-	2024
Day clinics	Vector	-	2024
Train stops	Vector	-	2024
Shadow fractions	Raster	3 m	2019	Generated using UMEP [59]
UTCI	Raster	3 m	2024

and comprise a comprehensive set of variables describing the form, function and outdoor thermal comfort of the Dortmund city area. These include building footprints and heights, tree canopy cover, road network, land use and land cover (LULC), street-level shading, population, the locations of various points of interest (POI), such as public spaces and hospitals, and the summertime Universal Thermal Climate Index (UTCI [54]). Most datasets are publicly available via North Rhine-Westphalia’s geoportal [55], Dortmund’s Open Data portal [56], and the Copernicus Land Monitoring Service (CLMS) [57,58], with the exception of the tree canopy cover, street-level shading, and the UTCI. The former was obtained from the Google Environmental Insights Explorer (EIE) [47], and the latter two were generated using the Urban Multi-scale Environmental Predictor (UMEP) [59], a GIS-based modeling tool for urban climate and planning applications.

To generate the shading data, first for buildings, then buildings and trees combined, we used UMEP and simulated the 3 m street-level shadow patterns across a range of solar azimuth (from 140° to 260°) and altitude (from 25° to 60°) angles that are representative of typical summertime noon and early afternoon (12:00–16:00) conditions in Dortmund. We distinguished between shadow cast by buildings and shadow cast by buildings together with trees in order to weight the temporally stable shading effect of buildings more strongly while still accounting for the more variable shading provided by trees, which fluctuates with season, tree health, and canopy size. The required building and tree canopy heights were derived from a Digital Surface Model (DSM) with a spatial resolution of 3 m, providing average heights above ground. The resulting shadow fractions were then averaged per pixel to create two distinct shading layers: one representing shadows cast by buildings (Figure 1a) and one representing shadows cast by buildings and trees (Figure 1b), thus enabling a more detailed spatial analysis of street-level shading.

The UTCI data were calculated at a 3 m resolution across the entire urban area for a range of summertime (August) weather conditions and noon-time (12:00–16:00) periods, when thermal discomfort tends to peak. To compute the UTCI, we utilized UMEP and the SOlar and LongWave Environmental Irradiance Geometry (SOLWEIG) model [61], using as inputs the building and tree canopy DSMs, the sky-view factor, and meteorological data obtained from numerical weather predictions (NWPs) generated by the ICON-D2 model of the German Weather Service (DWD) [62]. The resulting UTCI rasters were then averaged to a single raster layer.

2.3. Tree Planting Potential Assessment

To assess the tree planting potential (TPP), we employ an overlay analysis [35,36,37,63,64,65]. Overlay analysis is a widely used technique for identifying the most favorable locations for a specific objective [66], which in this case are the locations where planting trees can improve Dortmund’s outdoor thermal comfort the most. The workflow is shown in Figure 2. We start by generating a 50-m hexagonal grid that covers the entire city. Next, using the Urban Atlas (UA) LULC data, we discard all the cells that are unsuitable for tree planting (e.g., airport cells), as well as any isolated cells located in the city outskirts. After this step, we populate the attributes of the remaining cells with information about urban form, function, and outdoor thermal comfort using the data presented in the previous section. Finally, we normalize each attribute to range from 0 to 1 using min–max scaling and calculate the TPP by combining the standardized values using an additive weighted overlay model, which is the most commonly used method in multi-criteria decision-making applications [66].

We selected a hexagonal grid over a rectangular one because hexagonal cells (hexcells) minimize edge effects and maintain consistent neighborhood relationships, which is crucial for spatial analyses involving connectivity, particularly in urban green space planning [67,68]. The cell size of 50 m strikes a practical balance between spatial resolution and computational efficiency. It falls within the generally recommended scale for urban green infrastructure, typically between 10 m and 100 m, enabling meaningful assessments of local connectivity and ecological coherence [69]. This scale is also appropriate for operational integration with planning instruments such as land use plans and urban development frameworks, where green network considerations often operate at similar levels of granularity [44,70].

The UA LULC classes deemed unsuitable for improving outdoor thermal comfort through tree planting are: Airports, Arable land (annual crops), Fast transit roads and associated land, Forests, Green urban areas, Mineral extraction and dump sites, Isolated structures, Pastures, Port areas, Railways and associated land, Sports and leisure facilities, Water, and Wetlands. Any hexcells where the majority LULC class is one of the above were excluded from further analysis. For this step, the UA dataset was used in its original vector format and intersected with the 50 m hexgrid, without rasterization or downscaling. Following this automatic filtering, some isolated rural cells containing scattered structures, such as solitary farmhouses, remained. These were manually reviewed and also excluded, as they were considered unsuitable for tree planting.

To populate the hexgrid attributes, we applied zonal statistics and summarized the raster data within each cell. The added attributes are shown in Figure 3. To calculate the tree cover, built-up, and road fractions, we converted the corresponding datasets into 3 m boolean rasters and calculated for each hexcell the ratio of pixels with a value of 1 to the total number of pixels in each hexcell. The population density classes were obtained from the UA dataset by calculating the density for each UA polygon (population divided by area) and then reclassifying these continuous values into five discrete classes [71], ranging from low to high density, using the natural breaks method [72]. The POI data were aggregated into three thematic categories (

Table 2. Thematic POI groups and the layers assigned to each group.

Group	POI Layer
Culture, Recreation, and Education	Short-term care facilities, Schools, Youth clubs, Playgrounds, Sights, and Museums
Hospitals and Care Facilities	Hospitals, Rehabilitation centers, and Day clinics
Public Transportation Stops	Train stops

) and converted into 10 m heatmaps using a kernel density estimator [73].

Based on expert judgment, influence radii of 1000 m, 3000 m, and 500 m were applied for the respective categories. The mean summertime UTCI was computed per hexcell by averaging only those raster pixels representing outdoor urban environments, such as streets, public spaces, and playgrounds, while excluding built-up areas, following standard practice in outdoor thermal comfort assessments [54]. A similar approach was used to derive three shadow-related attributes: the mean shadow fraction caused by buildings (${\mathrm{Shadow}}_{\mathrm{b},\mathrm{mean}}$), the combined mean shadow fraction from buildings and trees (${\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{mean}}$), and the standard deviation of this combined shadow fraction (${\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{std}}$). Following the calculation of all variables, min–max normalization was applied to scale each attribute between 0 and 1, ensuring comparability across indicators [74].

To avoid assigning overly small weights and to reduce bias caused by highly correlated variables, we reduced the total number of attributes by combining related ones into composite scores [75]. Specifically, we merged the three shadow-related attributes into a single score, defined as

$${\mathrm{Shadow}}_{\mathrm{score}}=2\times \left(1-{\mathrm{Shadow}}_{\mathrm{b},\mathrm{mean}}\right)+2\times \left(1-{\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{mean}}\right)+{\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{StD}}$$

(1)

and the first two POIs of Table 2 into:

$${\mathrm{Social}}_{\mathrm{score}}=2\times {\mathrm{Social}}_{\mathrm{Health}}+{\mathrm{Social}}_{\mathrm{Culture}}$$

(2)

where ${\mathrm{Social}}_{\mathrm{Health}}$ denotes the “Hospitals and Care Facilities” category and ${\mathrm{Social}}_{\mathrm{Culture}}$ refers to “Culture, Recreation, and Education”. In Equation (1), we calculate the complement of the mean shadow fractions (i.e., $1-x$) to reflect the intuition that there is less shade caused by buildings and trees, indicating more open space and, consequently, a greater need and greater potential for tree planting. To match the scale of all other attributes, both ${\mathrm{Shadow}}_{\mathrm{score}}$ and ${\mathrm{Social}}_{\mathrm{score}}$ are normalized using min–max scaling to range between 0 and 1.

The final step of our method involves deriving the TPP for each hexcell by combining the collected attributes into a single composite score using weighted linear combination. This method is widely applied in spatial multicriteria decision-making because it is transparent, easy to interpret, and robust to extreme values [66,75]. The TPP for each hexcell i is computed as

$${\mathrm{TPP}}_i={\displaystyle \frac{{\sum }_{a\in A}{w}_a{x}_{a,i}}{{\sum }_{a\in A}{w}_a}},$$

(3)

where $w_a$ represents the weight assigned to attribute a for all attributes in attribute set A given in

Table 3. Weights assigned to each attribute in the TPP model. Higher values indicate greater influence on the final score.

Attribute	Weight
Tree Cover %	4
BuildCover %	3
RoadCover %	1
Population Density	3
Density of Public Transportation Stops	2
Mean UTCI	4
Socialscore	1
Shadowscore	3

, and $x_{a,i}$ is the corresponding normalized value for cell i. The weights, shown in Table 3, are based on expert judgment and reflect the objective of improving outdoor thermal comfort by prioritizing urban hexcells that have very little vegetation, ample open space, high summertime UTCI values, and are either densely populated or frequently visited. As with shadow-related attributes, the values for tree cover and built-up fractions are transformed to their complements prior to aggregation. This reflects the logic that areas with high existing vegetation or dense development offer limited capacity for tree planting. Finally, the computed TPP values are normalized to range between 0 and 1 and classified into three categories to support interpretation and planning: low potential ($\mathrm{TPP}<0.33$); moderate potential ($0.33\le \mathrm{TPP}<0.66$); and high potential ($\mathrm{TPP}\ge 0.66$).

To evaluate the influence of weighting on TPP, we performed a sensitivity analysis, in which three types of weight configurations were tested: (1) the expert-based weighting scheme of Table 3; (2) an all-equal weights scenario, where each attribute was assigned the same weight; and (3) a series of attribute emphasis runs, in which each attribute was individually assigned a higher weight to assess its isolated impact on TPP. Subsequently, we calculated the cross-correlation between each attribute and its corresponding TPP across the different weight configurations, examining the degree to which each attribute affected the final TPP values. For this sensitivity analysis, we rely on standard procedure to evaluate the robustness of composite indicators in multicriteria decision analysis [74,76]

2.4. Cluster Analysis

To enhance the usability of the study, we applied a cluster analysis to reveal the underlying spatial patterns of TPP and characteristic urban typologies. By identifying zones with similar combinations of environmental, infrastructural, and demographic attributes, clustering provides a structured and scalable basis to compare TPPs in different urban contexts.

We applied both spatial and non-spatial clustering techniques to the normalized attribute set, ensuring consistent distance calculations within a shared feature space. After evaluating several cluster configurations, we selected a solution with $k=4$ clusters, as it offered a meaningful balance between statistical validity and interpretability. This setup yielded a favorable silhouette score, a common metric used to capture the cohesion within and the separation between spatial clusters [77]. The four-cluster solution also aligned well with recognizable city structures, reinforcing its analytical relevance.

The spatial clustering method we used was hierarchical agglomerative clustering with the Ward linkage criterion, which minimizes variance within clusters. The bottom-up Ward approach starts by treating each hexagonal cell as an individual cluster and iteratively merging the most similar pairs until the desired number of clusters is reached [78]. For the non-spatial method, we applied k-means clustering to the same set of attributes to capture similarity in attribute space independent of geographic contiguity [79]. The k-means algorithm initializes k random centroids, assigns data points based on proximity, and iteratively updates the centroids until convergence. Despite the differences in methodology, both clustering approaches offer valuable perspectives on urban structure. Ward clustering, by incorporating spatial proximity, provides a regionally coherent view that mirrors known zoning patterns and contiguous functional areas. This makes it particularly useful for neighborhood-scale planning and district-level interventions. In contrast, k-means clustering prioritizes attribute similarity across space, allowing for the identification of dispersed but functionally similar zones. This is especially useful for city-wide comparisons and thematic planning, such as targeting all areas with high heat exposure or low vegetation cover, regardless of location.

2.5. Green Network Connectivity Analysis

While the previous steps focused on the characterization and prioritization of individual cells based on their intrinsic attributes, the green network connectivity analysis aims to address the possible connectivity of potential planting sites within the broader green infrastructure network. We implemented a neighborhood-based assessment within the 50 m hexagonal grid. For each cell with high TPP, we calculated the number of directly adjacent hexagons that were already covered by trees. To strengthen the role of connectivity in the resulting tree planting potential, we applied a buffer of 100 m around existing green spaces. This step prioritizes cells that not only score high individually but also contribute to the connectivity and strategic expansion of the green infrastructure (GI) network. Considering connectivity is essential, as the integrity of GI networks underpins ecological stability, diversity, and ecosystem service provision [80,81]. The 100 m buffer goes beyond the implicit 50 m created by the hexgrid resolution itself, ensuring that cells which strengthen or bridge existing patches are captured rather than overlooked. By integrating local connectivity into the evaluation, we aim to prioritize plantings that offer synergistic benefits in terms of urban cooling and ecological continuity.

3. Results

3.1. Weighting Evaluation and Attribute Influence

The influence of different variables on the TPP was analyzed by testing three predefined weighting configurations (

Table 4. The weight configurations used for the sensitivity analysis.

Case	Description
1	Expert judgment, weights shown in Table 3
2	All attributes are assigned equal weight 1, $w_a=1,a\in [1,8]$
3–10	Highlight one attribute of attribute set A with a weight of 2, all other attributes are assigned equal weight 1, $w_a=2,w_{b\ne a}=1,a,b\in [1,8]$

). The cross-correlation (

Table 5. Correlations between attributes (rows) and TPP_i for each weighting configuration (columns). Blue shades represent negative correlations, while red shades represent positive correlations (darker colors indicate stronger positive or negative values).

Attribute × Experiment	(1) Expert-Judgment	(2) All-equal	(3) Tree Cover (%)	(4) Build Cover (%)	(5) Road Cover (%)	(6) Population Density	(7) Public Transport	(8) UTCI mean	(9) Social score	(10) Shadow score
TreeCover (%)	$-0.6$	$-0.6$	$-0.6$	$-0.1$	$-0.3$	$-0.2$	$-0.3$	$-0.4$	$-0.2$	$-0.4$
BuildCover (%)	$-0.2$	$-0.2$	$-0.1$	$-0.6$	$-0.2$	$-0.2$	$-0.3$	$-0.3$	$-0.1$	$-0.2$
RoadCover (%)	$0.1$	$0.1$	$0.3$	$0.2$	$0.4$	$0.3$	$0.4$	$0.4$	$0.4$	$0.3$
Population Density	$0.1$	$0.3$	$0.4$	$0.4$	$0.3$	$0.7$	$0.4$	$0.4$	$0.4$	$0.4$
Public Transport	$0.3$	$0.5$	$0.2$	$0.4$	$0.2$	$0.4$	$0.7$	$0.4$	$0.4$	$0.3$
Mean UTCI	$0.7$	$0.3$	$0.5$	$0.3$	$0.2$	$0.1$	$0.2$	$0.3$	$0.2$	$0.0$
SocialHealth	$-0.2$	$-0.3$	$-0.1$	$-0.2$	$-0.2$	$-0.3$	$-0.2$	$-0.3$	$-0.2$	$-0.2$
SocialCulture	$-0.2$	$-0.1$	$0.0$	$-0.1$	$-0.1$	$-0.1$	$0.2$	$0.0$	$-0.1$	$-0.1$
Shadowb,mean	$0.3$	$0.2$	$0.2$	$0.3$	$0.3$	$0.2$	$0.3$	$0.2$	$0.2$	$0.2$
Shadowb+t,mean	$-0.5$	$-0.2$	$-0.5$	$-0.1$	$-0.5$	$-0.4$	$-0.4$	$-0.3$	$-0.2$	$-0.3$
Shadowb+t,std	$-0.2$	$-0.2$	$-0.1$	$0.1$	$-0.1$	$0.1$	$0.1$	$0.1$	$0.0$	$0.0$

) reveals how individual attributes impact the resulting TPP scores within each configuration. The combined scores ${Shadow}_{\mathrm{score}}$ and ${Social}_{\mathrm{score}}$ are not displayed in the figure, as they represent intermediate composites.

The attributes ${\mathrm{Shadow}}_{\mathrm{b},\mathrm{mean}}$, ${\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{mean}}$, Tree Cover (%), and Build Cover (%) were inverted in the scoring Equations (1) and (3), and accordingly show negative correlations for most configurations. For shadow-related attributes, ${\mathrm{Shadow}}_{\mathrm{b},\mathrm{mean}}$ shows slightly positive correlations in several cases, possibly due to its nuanced behavior in areas with varying building densities. In contrast, ${\mathrm{Shadow}}_{\mathrm{b}+\mathrm{t},\mathrm{std}}$ exhibits consistently weak correlations with TPP across all configurations, indicating a limited contribution to the overall variance in TPP and suggesting that it is the least influential attribute in the current model. Other variables, such as Mean UTCI and Population Density, consistently show moderate to strong positive correlations, underscoring their relevance to TPP in heat-affected and highly populated zones. Cultural, health, and public transport indicators also contribute noticeably, but to a lesser extent.

The effect of emphasizing individual parameters in experiments 3–10 is clearly visible: the correlation values are amplified where the target attribute is upweighted (e.g., Tree Cover (%) in its emphasized case). The expert-weighted case most closely resembles experiments 3 and 10 (see Table 4), reflecting its compound emphasis on Build Cover (%), Mean UTCI, and $s_{\mathrm{shade}}$. Comparing Cases 1 and 2 (Table 4) with the others reveals that Case 2 (all-equal weights) is structurally more similar to the highlight experiments, while the expert configuration (Case 1) is distinctly shaped by its targeted focus on thermal comfort and exposure. This confirms that thoughtful weighting can substantially steer model output in line with policy-relevant goals.

3.2. Spatial Distribution of TPP

The categorized TPP scores, representing the relative planting potential across the Dortmund area, are shown in Figure 4. The effects on the spatial distribution of the TPP between the all-equal weights (Figure 4a) and the realized expert judgment (Figure 4b) are visualized. As a result of the attribute selection and weight assignment, the spatial patterns reveal clusters of high-potential cells in areas characterized by low existing tree cover and minimal building or road density (compare with Figure 3). These zones often coincide with elevated Mean UTCI values and a high density of public POIs, indicating both biophysical feasibility and social relevance for targeted tree planting interventions.

The equal-weight scenario concentrates high-potential cells in the urban core, primarily in areas characterized by dense development and infrastructure. In contrast, the expert-weighted configuration results in a more dispersed spatial distribution, capturing peripheral neighborhoods with moderate but strategically significant tree planting potential. According to the Urban Atlas (UA) LULC classification, 33.1% of high-potential cells in the equal-weight case fall within the class ‘Continuous Urban Fabric’, followed by 32.3% in ‘Other Roads and Associated Land’, 17.9% in ‘Discontinuous Dense Urban Fabric’, and 14.5% in ‘Industrial, Commercial, Public, Military, and Private Units’. For the expert-weighted scenario, the distribution shifts notably: 30.3% of high-potential cells are located in ‘Discontinuous Dense Urban Fabric’, and another 30.0% in ‘Other Roads and Associated Land’. This highlights the emphasis of the expert configuration on more moderately dense residential areas and strategic infrastructure corridors outside the city center.

3.3. Combined Use of TPP and Clusters for Decision Support

To further contextualize the TPP and support urban planning applications, we performed clustering analyses to group hexcells with similar characteristics. The hierarchical clustering using Ward’s method identified four main urban typologies (Figure 5a):

• Cluster 0 (purple): Dispersed areas with specialized land use, such as warehouses.
• Cluster 1 (magenta): Dense urban core with limited planting space.
• Cluster 2 (orange): Residential outskirts with moderate development.
• Cluster 3 (yellow): Large-scale infrastructure and industrial zones.

These clusters reflect known urban structural zones and serve as a useful lens to understand where planting strategies may be best tailored. The k-means clustering grouped grid cells based on attribute similarity, regardless of spatial proximity (Figure 5b). This method emphasized the following:

• Cluster 0 (purple): Road network and linear infrastructure features.
• Cluster 1 (magenta): Densely built urban zones.
• Cluster 2 (orange): Compact residential areas.
• Cluster 3 (yellow): Open and transitional urban areas.

Notably, both clustering methods consistently identified the large warehouse area in the north (51.56° N, 7.42° E) as a distinct type, confirming the reliability of the clustering in highlighting unique urban forms.

To support decision-making, we examined the distribution of high TPP scores across spatial and non-spatial cluster typologies. This allowed for analysis of how high-potential planting locations align with broader urban characteristics. In the all-equal weighting configuration, 54% of the 2424 high-potential cells fall within Ward Cluster 1, representing the dense urban core, followed by 39% in Cluster 2 (residential outskirts) and 6% in Cluster 3 (infrastructure zones). Virtually no high-potential cells were located in the warehouse-dominated Cluster 0.

In the k-means clustering for the same configuration, 51% of high-potential cells belong to Cluster 3, which corresponds to open urban areas, while 25% fall within Cluster 0, reflecting road infrastructure. For the expert judgment-weighted scenario, the high-potential cells are more concentrated in Ward Cluster 2 (68%), reflecting its inclusion of moderately populated neighborhoods. Interestingly, Cluster 1 still contributes 16%, a similar absolute number as in the all-equal weights case. In the expert judgment-weighted k-means clustering, the majority of high-potential cells are in Cluster 3 (84%), followed by 7.1% in Cluster 2, 6.9% in Cluster 0, and 1.9% in Cluster 1. These distributions underscore how weighting choices and clustering methods can jointly inform spatial strategies for targeted urban greening.

3.4. Connectivity-Based Prioritization of Planting Sites

To assess whether high-potential tree planting cells contribute to a cohesive urban green infrastructure, we analyzed the spatial connectivity of these sites using two complementary approaches. First, we examined the adjacency of high-potential cells within the 50 m hexagonal grid. Figure 6 displays the distribution of high TPP cells and the number of adjacent neighbors with similarly high scores, comparing expert judgment and all-equal weighting configurations. In both cases, more than 90% of the high-potential cells have at least four neighboring cells with a similarly high TPP. This suggests a strong potential for large-scale coordinated tree planting campaigns that strengthen the existing green structure, especially in residential neighborhoods. Moreover, it becomes visible that the increase in cells with high TPP for the expert case is not evenly scattered throughout the total area but clustered in the local suburban centers, which creates connected TPP on the district and subdistrict level.

Second, we investigated the proximity of high-potential cells to existing green infrastructure. Figure 7 focuses on the expert judgment scenario and highlights all high-potential cells within 100 m of areas classified as ‘Green Urban Areas’ or ‘Forests’ in the UA dataset [58]. A threshold of 100 m was chosen to reflect the typical scale of spatial planning buffers and to ensure relevance for practical integration with land-use plans. Overall, 25% of high-potential cells are located within this buffer zone, suggesting that a significant share of prioritized sites could serve to extend or bridge existing green patches.

4. Discussion

This study presents a five-step data-driven workflow to assess the TPP in Dortmund, integrating environmental, infrastructural, and societal indicators, and identify high-priority tree-planting sites. The approach adds to the growing body of spatial decision support tools for nature-based solutions, particularly in the context of climate adaptation [17,34,39,41]. Our findings align with and extend recent research on urban tree planting potential. Studies such as Reitberger et al. [34] and Zambrano et al. [37] demonstrated the usefulness of remote sensing and distance analyses to identify favorable planting locations, while Wu et al. [40] and Kirnbauer et al. [41] laid important groundwork for transferable GIS-based and decision-support approaches. More specialized frameworks, for example, those that optimize planting according to solar radiation [38] or assess roadside planting potential [39], further underline how methodological choices shape outcomes. Nyelele et al. [43], in particular, emphasized that different prioritization frameworks can yield substantially different recommendations, with important implications for practice. Against this backdrop, our use of a weighted linear combination provides a transparent and reproducible framework in which the relative influence of each factor, urban form, function, and outdoor thermal comfort can be critically assessed and adapted to local priorities. Since the weighting of these factors is explicit and adjustable, our approach allows stakeholders to adapt priorities to local contexts, increasing both the transferability and legitimacy of the results, helping to avoid hidden biases, and contributing to a more consistent basis for urban greening strategies.

While our assessment included factors such as population density and the presence of socially relevant facilities (e.g., care homes, schools, youth clubs, playgrounds, hospitals, and rehabilitation and day clinics), we recognize that these indicators provide only a partial representation of demographic vulnerability. A more comprehensive integration of vulnerability would require finer-scale data on age, income, or health status, which are not consistently available at the urban scale due to privacy and accessibility constraints. This limitation highlights the challenge of incorporating social dimensions into transferable frameworks. Future research should therefore explore ways of embedding more nuanced demographic and socioeconomic data, where available, to strengthen the equity dimension of tree planting prioritization.

Another limitation concerns the resolution of the Urban Atlas data relative to the 50 m hexgrid. Because the UA polygons are coarser, some cells may be excluded entirely if they partially overlap with non-plantable categories. This introduces a conservative bias, as more potential planting sites are likely to be filtered out than would be the case with higher-resolution land cover data. Accordingly, our estimates should be understood as conservative lower bounds of tree planting potential.

The clustering results suggest that the expert-weighted configuration especially supports targeted greening strategies beyond the dense urban core. For example, high-potential cells in Cluster 2 of the spatial Ward clustering analysis (shown in Figure 5a) correspond to peripheral but moderately populated neighborhoods, which might be overlooked in a purely density-driven analysis. This supports a more equitable distribution of urban greening interventions and aligns with the climate justice goals outlined in local adaptation planning [44].

This shift in spatial distribution between the two clustering scenarios reflects differing priorities embedded in the weighting configurations. The equal-weight scenario favors areas with high population and infrastructure density. In contrast, the expert-weighted configuration identifies high-potential cells in more moderately dense, often overlooked peripheral neighborhoods. These areas may not show the highest absolute vulnerability, but their contextual relevance, such as fewer existing green spaces, limited shading, or localized heat exposure, makes them strategically important for achieving a more equitable distribution of urban ecosystem services. The increased representation of ‘Discontinuous Dense Urban Fabric’ in the expert configuration supports the city’s goals for targeted and locally embedded heat adaptation measures. This contrast underscores the value of adjustable weighting strategies to reflect different planning priorities ranging from maximum impact in densely built zones to inclusive green infrastructure development across the urban fabric.

The integration of network connectivity into the assessment reveals that a large portion of high-potential planting sites are located adjacent to existing vegetation or within spatial clusters. These spatial patterns highlight the added value of including connectivity in the TPP assessment. Such patterns are essential for delivering cumulative ecosystem services, including thermal regulation, biodiversity support, and public health benefits [21,82]. By prioritizing cells that both exhibit high individual scores and improve the coherence of the broader green network, planners can increase the ecosystem services provided by urban vegetation. The method provides a scalable decision support mechanism to identify impactful tree planting interventions aligned with integrated green space planning.

This pattern offers planners a choice between two strategic directions: enhancing existing green infrastructure by targeting adjacent cells (connectivity-driven), or expanding green infrastructure in under-served areas where high-potential cells are isolated (coverage-driven). This dual perspective enables cities to balance ecological performance and social equity when allocating resources for urban greening. The clustering results also support spatially coordinated strategies at the neighborhood scale, particularly in areas that combine moderate population density with lower tree coverage. These insights can be directly aligned with district-level heat mitigation plans or quarter-based development frameworks.

However, several practical challenges that are beyond the scope of a data-driven approach must be considered for implementing this tree planting strategy in urban areas. One of the foremost issues is maintaining long-term tree health, which requires the selection of appropriate species, irrigation planning, and protection against urban stressors [83,84,85]. Public participation and stakeholder acceptance also play a critical role, as local engagement is essential for both maintenance and equitable decision-making. Furthermore, better integration with existing tree planting initiatives could help align municipal efforts and improve efficiency [86,87,88]. From a physical planning perspective, underground infrastructure can severely limit feasible planting locations, and in dense urban contexts, increasing vegetation may impact ventilation corridors, potentially affecting urban microclimates. Urban land is also under high pressure from competing uses, which can further constrain implementation [29]. Lastly, while data-driven tools offer valuable planning support, their adoption in administrative practice still requires bridging gaps in capacity, interoperability, and institutional support, challenges that may go beyond the scope of this study but are nonetheless crucial for long-term impact [45,89].

The presented workflow is designed to be transferable to other urban contexts, particularly within regions that have access to comparable spatial datasets. By relying predominantly on publicly available data, open POI sources, and standardized urban climate indicators, the methodology can be adapted with minimal cost and effort. The only exception in this study was the use of tree canopy data from the Google EIE, which could be replaced by local or regional resources available for research. The integration of an open data strategy not only enhances replicability and transparency but also aligns with the growing demand for accessible tools that support data-informed urban governance. With adjustments to local data availability and policy priorities, this workflow offers a scalable foundation to support urban greening strategies, heat adaptation planning, and broader sustainability initiatives in cities worldwide.

Looking ahead, several directions for future research emerge. One important step will be to explore ways of complementing publicly available proxies for vulnerability with more detailed demographic and socioeconomic indicators, where such data can be accessed in order to further strengthen the equity perspective of tree planting prioritization. In addition, extending similar frameworks to incorporate projected climate scenarios and further testing them in other urban contexts would allow for broader transferability and long-term planning relevance. Finally, future work could compare the outcomes of different prioritization models and involve local stakeholders more directly in the weighting of indicators, supporting both scientific robustness and practical uptake in municipal decision-making.

5. Conclusions

This article presents a data-driven and replicable approach to identify urban TPPs that integrate multiple spatial, environmental, and social criteria. The method enables cities to move beyond ad hoc greening measures by providing a structured framework for spatial prioritization and connectivity planning. By applying this workflow to the city of Dortmund, we demonstrate how combining high-resolution datasets, attribute weighting, clustering, and connectivity analysis can reveal actionable insights for climate adaptation and equitable green infrastructure development. The approach highlights the need for integrated planning that considers not only where trees can be planted but also where they will have the most systemic impact. Future applications can adapt this methodology to different urban contexts and policy goals. Building on this foundation, promising research directions include integrating more detailed demographic and socioeconomic indicators to strengthen the equity perspective, incorporating projected climate scenarios for long-term planning, and comparing different prioritization models to better understand the influence of methodological choices. Combined with stakeholder engagement and administrative adoption, such extensions can further inform long-term urban forestry strategies that address both environmental challenges and social vulnerabilities.