Urban greening strategies for mitigating heat exposure and enhancing bird diversity

Abstract

Urbanization impact local climate by replacing green spaces with anthropogenic structures, which can increase heat exposure and diminish biodiversity. Minimizing these threats is required to maintain sustainable cities and communities globally, though the effectiveness of urban greening have not been broadly explored. Here, we analyze data from 1,039 cities worldwide to assess associations between urban greening and bird diversity. Urban greening was associated with 23–34% lower extreme heat exposure for birds when green-space quantity or quality was higher. Expanding green cover by 5–10% of total urban area showed the largest reductions, while higher green-space quality was also associated with lower heat exposure and greater bird diversity. Across cities worldwide, these greening scenarios were associated with 2–3% higher bird diversity. Our findings suggest that adopting urban greening practices would support and enhance urban biodiversity worldwide, a crucial component in the development of sustainable cities and communities.

Research across 1,039 cities worldwide shows that improving the amount, quality and placement of urban green space could reduce birds’ extreme heat exposure by 34 to 39% and increase urban bird diversity by 2 to 4%.

Introduction

Urbanization has altered land use and ecosystem characteristics in urban areas, significantly impacting the natural environment. Natural land cover and vegetation have been replaced with impermeable materials¹ that tend to absorb and emit heat through the urban heat island effect². As a result, heat exposure in urban areas has tripled in recent decades³, profoundly affecting human populations as the climate continues to warm⁴ and is expected to intensify further in the future⁵. The decline of natural habitats and rising urban temperatures threaten the abundance and diversity of species in urban areas, including birds⁶. Birds are among the most widely studied and ecologically informative indicators of environmental change in urban ecosystems. Birds respond rapidly to thermal stress, habitat modification, and vegetation structure^7,8, and their mobility allows them to track spatial heterogeneity in urban climates. Importantly, extreme heat events have been shown to directly affect avian survival⁹, reproductive success¹⁰, and behavior, making birds particularly sensitive to urban heat exposure¹¹. In addition, the availability of global bird occurrence data enables large-scale, cross-city analyses that remain infeasible for many other taxa¹².

Urban greening, which promotes cooling through evapotranspiration and shading provision, is recognized as an effective strategy to mitigate urban heat¹³. Previous studies have explored the impact of specific types of green infrastructure, such as vegetation greenness¹⁴ and urban tree cover^15,16, on the urban heat island effect. However, these studies often lack a comprehensive comparison of the mitigation of cooling capacities of different green infrastructures. Beyond its role in mitigating urban heat, urban greening also significantly supports bird diversity by providing essential habitats, diverse food resources, and migration stopover sites¹⁷. Previous studies have demonstrated that not only the amount, but also the quality and structural characteristics of green cover play a critical role in shaping bird communities¹⁸. Vegetation composition, vertical structure, and habitat quality have been linked to variation in bird species richness, functional traits, and phylogenetic diversity, particularly under climatic stress¹⁹. These findings provide an important foundation for understanding how urban greening may buffer the impacts of extreme heat on bird diversity.

Yet, current urban greening policies primarily focus on benefits to human well-being, such as temperature regulation, air quality improvement, or carbon sequestration, often overlooking the profound impact on urban biodiversity, particularly bird diversity²⁰. A growing body of ecological research has shown that, although birds are among the most frequently monitored urban taxa, assessments of urban bird responses to greening interventions have often relied primarily on abundance-based indicators²¹, which may overlook deeper shifts in community structure, ecological function, and evolutionary history²². Taxonomic diversity captures taxonomic completeness, functional diversity reflects the diversity of ecological strategies and roles present in bird assemblages, while phylogenetic diversity captures the evolutionary breadth and historical uniqueness, which is particularly relevant under climate-related stressors, such as extreme heat²³. Integrating these complementary dimensions is therefore essential for evaluating the ecological effectiveness of urban greening strategies beyond human-centered outcomes. Such an approach helps enable more generalizable assessments across cities and biogeographic contexts and supports the development of urban planning and conservation strategies that explicitly account for biodiversity resilience under ongoing climate warming.

Building on existing work, we analyze data from 1039 cities with a population of 500,000 or more²⁴ to assess the heat exposure of urban birds and to examine how urban greening is associated with bird diversity within urban ecosystems (see Supplementary Fig. 1 for our methodological overview). We first verify the relationship by which green space enhances bird diversity through its role in alleviating heat stress, employing both a spatial lag model (SLM) and a structural equation model (SEM) to quantify the relationships between urban greening indicators and bird diversity. We use three diversity metrics: taxonomic, functional, and phylogenetic diversity. Taxonomic diversity is represented by species richness. Functional diversity is represented by Rao’s quadratic entropy (Rao’s Q, RQ), and phylogenetic diversity is represented by Faith’s phylogenetic diversity (Faith’s PD, FPD). We subsequently apply machine learning techniques to predict urban bird heat exposure and evaluate the effectiveness of three greening strategies: Strategy 1, improving green space quality; Strategy 2, increasing green space quantity; and Strategy 3, optimizing the spatial distribution of green space. Each strategy is tested for its capacity to reduce heat exposure and enhance avian diversity across urban environments.

Results and discussion

Urban air temperature and extreme heat events

Figure 1a illustrates the mean near-surface air temperature (Ta) for 1039 cities worldwide, with each point’s color denoting its long-term average and its size scaled to urban population. Boreal Forest cities (BF, n = 10) cluster at the cold extreme (8–12 °C), reflecting their high-latitude, continental or high-elevation settings. Temperate Forest (TF, n = 363) and Temperate Grassland (TG, n = 94) cities span 16–24 °C, where mid-latitude solar angles, ocean currents, and prevailing westerlies moderate temperatures. Tropical-Subtropical Forest (TSF, n = 332) and Grassland (TSG, n = 74) cities exceed 28 °C year-round under intense, year-round insolation and high atmospheric moisture. Mediterranean (MED, n = 49) and Arid-Semi Desert (ASD, n = 117) cities fall between 20 and 32 °C, shaped by subtropical high-pressure regimes, coastal breezes, or continental dry-air advection.

Fig. 1. Global urban air temperature and extreme heat events.

a Global 1039 cities and their near-surface air temperature. Seven ecoregions: Arid-Semi-Arid Desert (ASD), Temperate Forest (TF), Mediterranean (MED), Temperate Grassland (TG), Tropical Subtropical Forest (TSF), Boreal Forest (BF), and Tropical Subtropical Grassland (TSG). b Frequency of extreme heat events in global cities during the period 2001–2020, classified by ecoregions. The column chart on the left shows the sum of the frequencies of extreme heat events in each city, and on the right is the average frequency of extreme heat events in each grid in the city. n represents the number of study cities within that ecological region. Data are presented as mean values ± SEM.

To characterize extreme heat exposure, we quantified the frequency of days exceeding 35 °C, given most birds will begin to activate physiological mechanisms to cope with heat stress above this temperature (Fig. 1b)^25,26. The average number of extreme heat events reported by cities varied across ecoregions, ranging from 1263 to 14,026. With the exception of BF cities, regional differences in total extreme heat frequency were relatively modest, and high mean Ta did not necessarily correspond to higher extreme heat burden; similarly, decoupling between mean temperature and extreme heat frequency has been reported in previous studies²⁷. When normalized by grid cell area, TSF and TSG cities showed the highest densities of extreme heat events (~60 events per cell). These patterns indicate that urban extreme heat exposure reflects not only background climate but also regional variability and local urban characteristics³.

Urban bird diversity

We created detailed maps illustrating the diversity of bird species across the globe and subsequently extracted urban-specific values using city boundaries (Fig. 2 and Supplementary Fig. 2). Our dataset includes 7135 species, representing approximately 66% of the world’s bird species. The probability of occurrence for these species were generated using citizen science data (Supplementary Fig. 2). To characterize global bird diversity from complementary ecological perspectives, we employed three metrics representing taxonomic, functional, and phylogenetic diversity. Taxonomic diversity was represented by species richness, which captures the expected number of species within each grid cell. Functional diversity was represented by Rao’s Q, which reflects trait dissimilarity and functional differentiation among co-occurring species. Phylogenetic diversity was represented by Faith’s PD, which captures the amount of evolutionary history retained within bird assemblages. These metrics are intended to provide parallel perspectives on how urban bird communities are exposed to and potentially buffered from extreme heat. The spatial distributions of the three diversity dimensions generally align, with cities exhibiting higher bird diversity primarily located in tropical and subtropical regions on each continent (Supplementary Fig. 3). Cities in East Asia, Central Africa, North America, and Central America display relatively high taxonomic diversity, while functional and phylogenetic diversity show partially distinct geographic patterns, reflecting differences in trait composition and evolutionary history among regional bird assemblages.

Fig. 2. Global urban bird diversity map (richness, Rao’s Q and Faith’s PD) and impact of extreme heat events on different bird diversity indicators.

Global distributions of city-level bird diversity across 1039 cities, represented by a species richness, b Rao’s quadratic entropy (Rao’s Q), and c Faith’s phylogenetic diversity (Faith’s PD). Circle size is proportional to urban population, and circle color represents the corresponding bird diversity metric from low to high. Associations between the number of extreme heat events and d species richness, e Rao’s Q, and f Faith’s PD. Each point represents one city (n = 1039), and solid lines indicate ordinary least-squares linear regression fits. Statistical significance of each regression slope was assessed using a two-sided t-test against the null hypothesis that the slope was equal to zero (d.f. = 1037). Shading around each fitted line represents the 95% confidence interval. R² and P values are shown in the corresponding panels. No adjustment was made for multiple comparisons.

Bird diversity was assessed using probability of derived from citizen science data, rather than direct field surveys conducted uniformly across all cities. While this approach enables consistent global coverage and cross-city comparison, it cannot fully capture fine-scale variation in species occurrence or short-term population dynamics. Accordingly, the exposure metrics presented here should be interpreted as relative indicators of community-level heat exposure rather than precise measures of demographic impact. We also acknowledge that altered observer behavior during COVID-19 periods may have introduced residual bias; however, because bird diversity was not based on raw observations, such effects are unlikely to systematically influence the comparative patterns observed among cities.

Relationship exploration

We first sought to uncover the underlying relationships through which urban greening enhances bird diversity by alleviating extreme heat. To this end, we employed a SLM to test whether thermal environments influence urban bird diversity, followed by structural equation modeling (SEM) to identify the causal pathways linking greening to biodiversity through thermal mitigation. Our SLM results reveal a strong and quantifiable relationship among urban greening, extreme heat exposure, and bird diversity metrics (Fig. 3a), with model R² values exceeding 0.9 (Supplementary Fig. 4). Both greening quality (VQI; Models 1–3) and quantity (VCF; Models 4–6) are positively and significantly associated with bird diversity. The larger coefficients for VCF indicate that, on a global scale and across diverse urban contexts, expanding green space tends to yield greater marginal gains in diversity than improving vegetation quality. Previous studies have likewise shown that vegetation cover is an important driver of urban bird diversity, although the relative importance of vegetation quantity and quality may vary across contexts²⁰.

Fig. 3. Relationships of bird diversity with urban environment.

a SLM of various urban greening metrics and bird diversity. We chose vegetation quality index (VQI) to represent green space quality, and vegetation continuous field (VCF) to represent green space quantity. Hotdays denotes the number of extreme heat events; POP, population; QUA, habitat quality; NPP, net primary productivity; PRE, precipitation; and PM₂.₅, particulate matter with an aerodynamic diameter of ≤2.5 µm. Statistical significance of each regression coefficient was evaluated using two-sided asymptotic Wald z tests against the null hypothesis that the coefficient equalled zero. Points represent the estimated regression coefficients, and horizontal bars represent 95% confidence intervals. See methods for additional details. n represents the number of urban grids used for calculations. b SEM to explore the various pathways by which vegetation increases urban bird diversity. In each model, in addition to direct effects, we specified three indirect pathways from greening through heat mitigation, food supply, and habitat enhancement. BD denotes bird diversity; Dw, distance to water bodies; CE, environmental covariates; NPP, net primary productivity; and QUA, habitat quality. Heat event denotes the number of extreme heat events. RI, RQ, and FPD indicate the species-richness, Rao’s Q, and Faith’s PD models, respectively. BD was used as the response indicator and VQI/VCF as the explanatory variable in this model. Heat event (refers to Hotdays), NPP, and QUA were used as mediating variables in the three pathways. The path where the heat event is located represents the indirect effect path of heat mitigation. The path where the NPP is located represents the indirect effect path of food supply. The path where the QUA is located represents the indirect effect path of habitat improvement. The different colored paths point to the results of the corresponding bird diversity metrics.

However, the consistent significance of VQI highlights the importance of vegetation health and structural complexity, as represented by Rao’s Q and Faith’s PD. High-quality vegetation, such as multi-layered tree and shrub canopies, provides enhanced thermal buffering and habitat heterogeneity, which are critical for maintaining bird diversity under extreme heat conditions. In contrast, the number of extreme heat events (Hotdays) consistently exhibits strong, negative associations with bird diversity across all six models. The results indicate that hot days above the 35 °C threshold is associated with a significant reduction in diversity, consistent with documented ecological effects of extreme heat on birds, including increased mortality risk, reduced reproductive success, and heat-induced behavioral constraints reported in previous studies^28,29. Greening serves as a viable intervention to counter these effects by improving microclimatic conditions^30,31, enhancing insect-mediated food resources^32,33, and providing shaded, thermally buffered nesting and foraging habitats^34,35.

We further applied SEM to assess whether these mechanisms operate independently and robustly at the global scale for each combination of diversity and greening metrics. The SEM results (Fig. 3b) clarify that direct greening effects explain approximately 52–62% of the total influence on bird diversity, while indirect effects account for the remaining 38–48%. Among the indirect effects, the thermal mitigation pathway emerges as the most influential, contributing 38–54% of the total indirect effect. Both VQI and VCF significantly reduce extreme heat exposure, reaffirming that temperature buffering is the primary pathway through which greening benefits biodiversity. Collectively, our findings demonstrate that the biodiversity benefits of urban greening arise from an interplay of mechanisms rather than a singular effect. These include direct habitat provision and a suite of indirect ecological improvements that synergistically enhance species richness and community resilience in increasingly thermally stressed urban environments, consistent with previous studies³⁶.

Greening strategies to mitigate heat exposure

Effective mitigation of urban heat exposure requires consideration not only of the total amount and area of urban greening but also the type of green infrastructure employed²⁰. Specifically, by quantifying the cooling potential of increased green space metrics, we aimed to determine the degree to which such interventions can dampen the extreme heat exposure (Fig. 4a and Supplementary Fig. 5). We analyzed the impact of three different urban greening strategies (Strategy 1, improving green space quality; Strategy 2, increasing green space quantity; and Strategy 3, optimizing the spatial distribution of green space) on reducing the various exposure of urban birds (BE1-3) to extreme Ta, where BE1, BE2, and BE3 correspond to richness, Rao’s Q, and Faith’s PD respectively. For each strategy, we established 5 optimization increments in average greening indicators, ranging from small to large (Supplementary Table 1). Our findings indicate that spatial optimization is particularly effective at smaller optimization increments, especially Strategy 3b, which achieves 5.62, 12.56, and 7.07% reductions in the three types of bird heat exposure by optimizing space and increasing the greening indicator by just 1% (Fig. 4b). This suggests that even very small, strategically targeted increases in high-performance green space can deliver outsized reductions in avian heat risk. By optimizing placement and choosing the right green space location, we can unlock the greatest cooling, therefore the greatest reduction in bird heat exposure, with minimal additional green area.

Fig. 4. Bird exposure reductions under greening strategies.

a Results of different bird exposure values in 1039 cities. We propose bird exposure values corresponding to different bird diversity. BE1 corresponds to richness-based taxonomic diversity exposure values, BE2 corresponds to Rao’s Q-based functional diversity exposure values, and BE3 corresponds to Faith’s PD-based phylogenetic diversity values. b Expected effects of different optimization strategies on bird exposure reduction. Strategy 1 increases the quality of green space according to the increment, Strategy 2 increases the quantity of green space according to the increment, and Strategy 3 optimizes the green space distribution. In Strategy 3, the top 30% of heat-exposed areas in each city are regarded as hot spots, and their corresponding green space indicators are upgraded to the urban green space average. After optimization, Strategy 3a increases the quality of green space according to the increment in the hotspot area, and Strategy 3b increases the quantity of green space according to the increment in the hotspot area. View Supplementary Fig. 5 for more information. n represents the number of urban grids used for calculations. Data are presented as mean values ±95% confidence interval.

An increase in greening quantity generally results in more vegetation cover, which provides effective shading and reduces heat exposure³⁷. As the optimization increment increases, strategies that enhance the quality or quantity of green spaces globally yield better results. At the 20% optimization increment, Strategy 1 demonstrates that globally improving green space quality leads to 26.22, 29.61, and 32.58% reductions in bird heat exposure, which are 8.91–19.03% greater than those achieved through spatial optimization alone, and are particularly pronounced for richness and Faith’s PD. An overall increase in trees and vegetation can offer extensive shade, significantly reducing direct sunlight on surfaces, thereby aiding in temperature regulation and mitigating the urban heat island effect. Additionally, green space contributes to cooling through transpiration, which helps lower the temperature of the surrounding environment, a crucial process mentioned for regulating heat in urban areas³⁸.

Our analysis also revealed that, in a global optimization strategy, improving the quality of green space is more effective than merely increasing its quantity in mitigating heat exposure for birds, especially at low implementation levels. A 5% quality increment reduces richness, Rao’s Q, and Faith’s PD exposure by 6.22%, 8.48%, and 7.94%, respectively, compared with 6.10%, 8.06%, and 6.46% under quantity expansion. As the increment intensifies beyond 10%, quantity gains begin to match or exceed quality gains for Rao’s Q, reaching a 34.49% reduction at the 20% increment compared with 29.61% under quality improvement. However, quality improvement still leads for richness and Faith’s PD at higher increments, with reductions of 26.22% and 32.58%, respectively. All these dynamics indicate that when land is scarce, targeting physiological and structural vegetation quality maximizes cooling efficiency per unit effort, whereas substantial area increases become more advantageous only at higher intervention levels. An optimal mitigation strategy should therefore marry high-performance vegetation location with judicious expansion of green cover to achieve the greatest overall reduction in bird heat exposure.

The bird heat exposure metrics we used are exposure-based indicators, not mechanistic models of physiological or demographic responses. They combine the frequency of extreme heat events with community-level bird diversity to capture where and how often bird assemblages coincide with high temperatures, rather than implying a direct causal process. This approach allows consistent comparison of greening strategies across cities and shows that spatially targeted interventions can meaningfully reduce bird exposure to extreme heat, even with relatively small increases in green infrastructure. Future work should look to develop mechanistic models that corroborate our results here. Moreover, this study focused exclusively on heat exposure and did not explicitly consider other interacting urban stressors, such as air pollution, noise, habitat fragmentation, predation pressure, or resource availability. Integrating multiple stressors with species-specific physiological, behavioral, and demographic data in future research would further refine our understanding of how urban greening can enhance biodiversity resilience under ongoing climate warming.

Mitigating heat exposure for bird diversity

Using data on greening indicators modified by various strategies, combined with the original bird diversity model, we regenerated global urban bird diversity maps to reveal how different greening strategies and optimization increments enhance urban bird diversity (Fig. 5). Overall, the impact of these greening strategies on biodiversity improvement aligns closely with their effects on mitigating heat exposure. These strategies are expected to increase global urban bird abundance by up to 2.6–2.9%. Among them, strategies that focus on overall increases in both the quality and quantity of green space show the most significant improvements in bird diversity. At the maximum optimization increment, global greening strategies increase richness by 2.50–2.56%, Rao’s Q by approximately 2.72%, and Faith’s PD by 2.73–2.87%. Spatial optimization that combines increased green space quantity with overall improvements in green space quality offers distinct advantages depending on the optimization increment.

Fig. 5. Effects of different greening strategies on bird diversity.

Selection frequency histogram of the optimal strategy for improving urban bird diversity under different optimization increments. The ordinate is the frequency, and the abscissa is the corresponding diversity improvement value. n represents the number of study cities. Data are presented as mean, minima, maxima, and center values.

Contrary to our initial hypothesis that strategies increasing green space quantity would associate enhancing bird diversity, the results reveal a more nuanced pattern. Our findings are consistent with previous work showing that vegetation quality and structural characteristics can be as important as, or more important than, sheer vegetation extent in shaping avian responses to urban environments³⁹. At low optimization increments, spatial optimization delivers stronger biodiversity gains, especially Strategy 3b, which increases richness, Rao’s Q, and Faith’s PD by 0.93%, 1.57%, and 1.10%, respectively, at only a 1% increment. As increments increase to 10–20%, global greening strategies become more effective, Strategy 2 produces the largest improvements in richness and Rao’s Q, while Strategy 1 produces the largest improvement in Faith’s PD at the 20% increment. This finding underscores the critical role those high-quality habitats play in supporting urban bird diversity. Healthy, multilayered plantings of diverse, native species supply a broad suite of ecosystem services (abundant food resources, secure nesting sites, heat refuges and structural complexity) that underpin both abundance and functional diversity⁴⁰. By contrast, large-scale increases in vegetation cover alone may add canopy area but lack the compositional and architectural variety birds need, yielding smaller returns per unit effort¹⁹. Moreover, quality-rich green spaces create a dynamic mosaic of micro-habitats, enabling species to exploit distinct ecological niches across seasons and weather extremes, thus bolstering resilience and speeding recovery after disturbances⁴¹.

As indicated by previous results, high-quality greening strategies more effectively mitigate heat exposure, improve soil moisture, and enhance air quality, thereby making habitats more suitable for birds^42,43. This microclimate regulation effect is especially significant in large-scale optimization, contributing to the overall improvement of the urban environment⁴⁴. When selecting greening strategies with different levels of optimization, the most effective approaches consider both heat exposure mitigation and bird diversity enhancement. With minor adjustments, increased greening can provide crucial refuge and buffer zones for birds, thus protecting greater bird diversity⁴⁵.

Functional diversity was represented by Rao’s Q and phylogenetic diversity by Faith’s PD, which summarize community-level differences in functional traits and evolutionary history but do not directly capture species-specific physiological tolerance or vulnerability to heat stress. As a result, changes in these diversity metrics reflect shifts in community structure rather than mechanistic responses of individual species or traits.

Regional heterogeneity

We also observed considerable variation in the effectiveness of different green space optimization strategies for mitigating heat exposure across various ecoregions and bird diversity measures (Fig. 6). When selecting the optimal strategy, BE1, BE2, and BE3 showed different overall transition patterns across optimization increments. For BE1, related to richness, Strategy 3a dominated at low increments of 1–2%, whereas Strategy 1 became the dominant strategy from the 5% increment onward. For BE2, related to Rao’s Q, Strategy 3b was most frequently selected at low to moderate increments of 1–5%, while the optimal strategy became more dispersed at higher increments, with Strategy 3a slightly leading at 10% and Strategy 1 at 20%. For BE3, related to Faith’s PD, Strategy 3b dominated at 1–2%, Strategy 2 became more important at 5–10%, and Strategy 1 became dominant at the 20% increment. Yet regional departures emerged: ASD and MED cities tended to benefit more from global increases in green space quantity at larger increments, whereas TF, TSF, and TSG cities more often shifted toward global green space quality enhancement for BE1 and BE3. BF cities showed a strong preference for spatial optimization combined with green space quantity at low to moderate increments, but shifted toward global strategies at the largest increment.

Fig. 6. Optimal greening strategies for mitigating bird heat exposure across global ecoregions.

a Results for bird exposure based on species richness (BE₁). b Results for bird exposure based on Rao’s quadratic entropy (BE₂). c Results for bird exposure based on Faith’s phylogenetic diversity (BE₃). Within each panel, the left chart shows the optimal greening strategy for reducing bird heat exposure in each ecoregion at different optimization increments (1, 2, 5, 10, and 20%), including the total sample (Tot). The right Sankey diagram summarizes how the optimal strategy shifts across increments when all cities are considered together under the corresponding bird-exposure metric. Colors indicate the optimal strategy: Strategy 1, global improvement in green-space quality; Strategy 2, global increase in green-space quantity; Strategy 3a, spatially targeted improvement in green-space quality; and Strategy 3b, spatially targeted increase in green-space quantity. ASD, Arid-Semi-Arid Desert; BF, Boreal Forest; MED, Mediterranean; TF, Temperate Forest; TG, Temperate Grassland; TSF, Tropical-Subtropical Forest; and TSG, Tropical-Subtropical Grassland. n indicates the number of study cities within each ecoregion. n represents the number of study cities within that ecological region.

The varying responses of different ecoregions to greening strategies highlight how unique vegetation structures and ecological characteristics shape the urban environment. In TF and TSF cities, there was a stronger preference for strategies that enhance green space quantity, especially for richness and Faith’s PD, suggesting that the existing vegetation structure in these areas may already be sufficient to support adjustments in the urban microclimate. In these ecoregions, forests typically have dense vegetation⁴⁶. This dense vegetation effectively absorbs and reflects solar radiation, reducing surface temperatures. Additionally, soils in these regions are often rich in organic matter and have excellent water retention properties⁴⁷, which helps to store water and slow the rate of evaporation, further contributing to lower surface temperatures.

ASD and MED cities share a common response. At low increments, spatial optimization strategy yields the greatest species-richness boosts, while at higher increments broad, strategy that enhances green space quality globally takes over. In these mid- and low-latitude zones, initial hotspot greening maximizes per-unit cooling in the most heat-stressed micro-habitats, but only larger-scale canopy coverage can uniformly lower urban temperatures and open new niches for a broader suite of bird species⁴⁸. TG and TSG cities showed more indicator-specific shifts; richness and phylogenetic exposure tended to favor global quality improvement at higher increments, whereas some functional-diversity responses benefited more from quantity expansion. In this semi-arid biome, water scarcity and high evaporative demand constrain plant performance. Improving vegetation physiological health via tolerant, high-transpiration species delivers the greatest local cooling benefits and habitat suitability, supporting richer bird assemblages even under modest greening⁴⁹. In BF cities, the tendency to increase the amount of greening is likely due to the dense tree cover, which readily creates a tall and thick vegetation structure^50,51.

For BE2, related to Rao’s Q, regional heterogeneity was even more pronounced. Spatial optimization combined with increased green space quantity was generally more effective at low increments, indicating that targeted canopy-cover improvement can rapidly reduce heat exposure for functionally diverse bird communities. As the optimization increment increased, however, the optimal strategy differed among ecoregions: ASD, MED, and TF cities more often favored global quantity increases, while TG, TSF, and TSG cities showed stronger benefits from global quality improvement at the highest increment. Thus, increasing vegetation area not only directly alleviates heat exposure through shading and transpiration, but its effectiveness also depends on whether the limiting factor in each ecoregion is canopy extent, vegetation quality, or the spatial distribution of cooling resources.

This study provides a global, comparative assessment of how urban greening can mitigate bird heat exposure by integrating near-ground air temperature, multiple dimensions of bird diversity, and spatially explicit greening strategies across 1039 cities. We show that extreme heat exposure of urban birds is widespread but uneven, and that greening interventions can substantially reduce this exposure. Importantly, the effectiveness of greening depends not only on the amount of green space added, but also on its quality and spatial configuration. Targeted, spatially optimized interventions yield disproportionately large reductions in bird heat exposure at low implementation levels, while broader increases in green cover become increasingly effective at higher levels of intervention. Together, these findings position urban greening as a viable climate-adaptation strategy for biodiversity.

Methods

Data collection

Bird data

We used data from the eBird Basic Dataset (EBD), which is global in extent and updated monthly. For our analysis, we used the EBD, version Nov-2022¹², from which we extracted data from March 2001 to February 2021, which was the period with abundant bird observation records and complete environmental variable data. To ensure these data represent the “best quality” data, we employed a set of filtering, aiming to remove potential “outliers” which could bias our dataset⁵². Checklists were only included if (i) users indicated that they recorded all birds detected during the survey, i.e., “complete” checklists from which we could obtain both detected presence and pseudo-absence; (ii) the protocol type was either “stationary” or “traveling”; (iii) duration was >5 min and ≤240 min; (iv) distance traveled was ≤5 km; and (v) the number of observers was ≤10. We defined an equal area hexagonal grid across the study region, and then subsampled detections and non-detections separately to ensure that we don’t lose too many detections. In our example, we sampled one detection and one non-detection checklist from each grid cell for each week. With hexagonal grids, we can be sure all the cells are of equal area, which is particularly important if we have a large region for modeling⁵². Then we randomly subsampled the non-detections so there were equal numbers of detections and non-detections, which can address the issue of class imbalance. We only selected the species which had more than 50 valid checklists with complete environmental data⁵³. Our dataset consisted of 7135 species, representing ca. 66% of the world’s bird species⁵⁴.

The eBird observations from 2010 to 2021 were used to estimate species-specific response probability curves rather than to represent bird distributions in any single year. The use of multi-year observations allowed us to characterize each species’ response to environmental gradients across a broad range of conditions and to improve the stability of the estimated species-environment relationships. These fitted response functions were then projected onto the environmental predictor layers for 2020 to generate simulated species occurrence probability surfaces for 2020. Therefore, the reference year of the final bird diversity maps and heat exposure analyses is 2020, while the 2010–2021 eBird records were used only for model training. This approach does not assume that bird observations from all years directly match the environmental conditions in 2020; rather, the multi-year observations were used to infer general species-environment response relationships, which were subsequently applied to the 2020 environmental conditions. At the community level, bird diversity was characterized using three complementary metrics: species richness, Rao’s Q, and Faith’s PD.

Species richness was defined as the expected number of species per grid cell, calculated by summing species-specific probabilities of occurrence within each 0.1° cell. We employed the working principle of Naive Bayes to calculate the probability of each species being observed, essential for considering the nuanced response patterns of birds to various environmental variables (Supplementary Fig. 6). These models describe species-environment response relationships and were combined with environmental data for 2020 to generate spatially explicit predictions in 2020, from which urban areas were subsequently extracted (see Supplementary Information S.2 for model details). The probability of a bird species being observed and not being observed in a spatial pixel can be calculated as:

$$P\left(B_k/C\right)=\frac{P\left(B_k\right)\times P\left(C/B_k\right)}{{\sum }_k\left(P\left(B_k\right)\times P\left(C/B_k\right)\right)}$$ 1

where C represents the geographical environment of the spatial pixel, and $B_k$ represents whether the species has been observed. $P\left(B_k\right)$ can be calculated by the ratio of the number of records in which the species was observed or the number of records in which the species was not observed to the total number of records. $P\left(C,\mid ,B_k\right)$ can be calculated as:

$$P\left(B_k/C\right)={\prod }_{i=1}^{m}P\left(A_i/B_k\right)$$ 2

where $A_i$ represents the characteristics of an environmental variable of a spatial pixel. For categorical environmental variables, birds’ responses to them often show a multi-polar pattern, and the proportion of each category in the statistical observation records is expressed as $P\left(A_i,\mid ,B_k\right)$. For continuous environmental variables, birds may exhibit a stabilized response, an itemized response, or a randomized response mode. We used four probability distribution models of normal distribution, exponential distribution, Weibull distribution, and random distribution to fit the distribution of a bird species in the environmental variable. The best-fitting distribution model would be the best fitting distribution model would be selected according to the AIC information criterion selected according to the AIC information criterion. After calculating the probability of each species being observed, we added the probabilities of all species to obtain the global spatial pattern of bird richness. Expected species richness was calculated by summing species-specific probabilities of occurrence within each grid cell. No thresholding was applied to convert probabilities to binary presence-absence. By summarizing species specific response patterns to these covariates, the modeling framework reduced the influence of spatial clustering of observations and uneven observer effort (S.2 in Supplementary Information). As a result, predicted distributions reflect ecologically plausible occurrence patterns rather than raw sampling intensity.

Following the calculation of species richness, we quantified two additional dimensions of bird diversity from the same species-specific occurrence probabilities. Rao’s Q was used to represent functional diversity, defined as the expected functional dissimilarity between two species drawn from a probability-weighted assemblage^55,56. Faith’s phylogenetic diversity Faith’s PD was used to represent phylogenetic diversity, defined as the amount of evolutionary history represented by the assemblage, measured as the total branch length connecting the species expected to occur in a pixel⁵⁷.

Rao’s Q was used to represent functional diversity by combining the relative contribution of each species with the functional dissimilarity among species⁵⁵. Rao’s Q provides a flexible framework for integrating bird traits with different data structures, including continuous morphological traits, binary traits, and categorical ecological traits. By converting these heterogeneous trait variables into a unified functional dissimilarity matrix, Rao’s Q enables a consistent comparison of assemblage-level functional differentiation across global urban grid cells. In this study, functional diversity was used to assess how urban heat exposure and greening scenarios influence the overall functional structure of bird communities, providing a complementary perspective to species richness and relative abundance. We acquired bird functional trait data from Pigot⁵⁸ and Wilman⁵⁹. We examined four traits: morphology, nesting preference, trophic ecological niche, and nocturnality (Supplementary Fig. 2). Continuous traits were standardized, and categorical traits were converted to dummy variables. A pairwise functional dissimilarity matrix was then generated for all matched species.

$$R{Q}_j={\sum }_i{\sum }_k{w}_{i,j}{w}_{k,j}{d}_{i,k}^F$$ 3

where $R{Q}_j$ is the Rao’s Q value of grid cell j; ${\text{w}}_{\text{k},\text{j}}=\text{P}\left({\text{B}}_{\text{k}}\text{∣}{\text{C}}_{\text{j}}\right)/{\text{Sigma}}_{\text{l}}\text{P}\left({\text{B}}_{\text{l}}\text{∣}{\text{C}}_{\text{j}}\right)$, $\text{P}\left({\text{B}}_{\text{k}}\text{∣}{\text{C}}_{\text{j}}\right)$ is the species occurrence probability estimated from the species distribution model described above; $w_{\mathrm{k},\mathrm{j}}$ is the normalized probability weight of species k in grid cell j; $d_{i,k}$ is the functional dissimilarity between species i and species k; between species i and species k; and the summations are taken over all species retained after matching the probability surfaces with the trait table. A higher $R{Q}_j$ indicates that a cell contains species with more distinct functional traits or a more even probability distribution across functionally different species. Grid cells with zero summed occurrence probability were treated as missing values.

Faith’s PD was used to quantify the amount of evolutionary history represented by the probability-weighted urban bird assemblage⁵⁷. We used a dated global avian phylogeny derived from the BirdTree framework and pruned each tree to the species retained in the probability dataset⁶⁰. Because the input community data were continuous probability surfaces rather than observed local species lists, we calculated an expected version of Faith’s PD. This approach preserves the branch-length interpretation of Faith’s PD while avoiding an arbitrary probability threshold for defining species presence.

$$FP{D}_j={\Sigma }_b{L}_b\left[1-{\Pi }_{k\in D_b}\left(1-P\left(B_k\mid C_j\right)\right)\right]$$ 4

In this formula, $FP{D}_j$ is the expected Faith’s PD of grid cell j; b denotes a branch in the phylogenetic tree; $L_b$ is the length of branch b; $D_b$ is the set of descendant species subtended by branch b; and $P\left(B_{\mathrm{k}}\mid C_{\mathrm{j}}\right)$ is the same occurrence probability used to calculate richness. The term $\left[1-{\Pi }_{k\in D_b}\left(1-P\left(B_k\mid C_j\right)\right)\right]$ represents the probability that at least one descendant species of branch b occurs in grid cell j. Branches with high descendant occurrence probability and long evolutionary length therefore contribute more strongly to $FP{D}_j$. We used 50 multiple trees and then averaged each tree’s value to reduce the influence of phylogenetic uncertainty.

To account for seasonal variation in bird communities, bird diversity was estimated separately for two seasonal periods. Season A corresponds to the Northern Hemisphere breeding period from March to August, and Season B corresponds to the non-breeding period from September to February of the following year. Species distribution probabilities and community-level diversity metrics were generated independently for each season.

Near-surface air temperature and heat thresholds

Since the living environment of birds is usually above the ground, the heat conditions there are very different from the surface temperature measured by satellite land surface temperature (LST) data; we used a dataset that describes the global near-ground air temperature (Ta) at a 1 km spatial resolution in 2020⁶¹. This dataset integrates satellite land surface temperature, elevation, and in situ meteorological observations through a spatial modeling framework, providing globally consistent estimates of near-surface air temperature suitable for ecological and urban-scale analyses. Combined with the research on bird heat tolerance, we chose 35 °C as the threshold for extreme heat exposure temperature because most birds will begin to activate physiological mechanisms to cope with heat stress above this temperature^25,26,62. Subsequently, we quantified the number of days surpassing this threshold during the three hottest months of each year in each city, providing a quantitative measure of the frequency of extreme heat events during in 2020.

For each city, the local hot season was identified based on near-surface air temperature (Ta). Cities were then matched to the corresponding seasonal bird diversity dataset according to whether their heat season overlapped with Season A or Season B. This procedure ensures that bird diversity used in heat exposure calculations reflects the assemblages that are seasonally present during periods of elevated temperature, rather than assuming year-round species presence. Additional details are provided in the Supplementary Information.

Urban greening variables

To quantify the impact of vegetation cover and its dynamics on extreme hot days in the study area, we incorporated the vegetation quality index (VQI), referring to greening quality, and vegetation continuum field (VCF), referring to greening quantity, into our analysis.

Giving that the morphological structure and community density of vegetation will significantly affect its cooling efficiency, we selected three indicators: normalized difference vegetation index (NDVI), canopy height, and above ground biomass. These variables were selected a priori based on established evidence linking vegetation structure and condition to urban cooling processes, including shading¹³, evapotranspiration⁶³, and surface energy balance regulation⁶⁴. Vegetation quality was summarized using a VQI constructed through principal component analysis (PCA). PCA was employed to integrate multiple vegetation-related variables into a single composite indicator by reducing collinearity and deriving data-driven weights. The first principal component was retained as VQI because it captured the dominant shared variance among vegetation attributes and represented a gradient of vegetation structural and functional quality. Importantly, PCA was used here as a statistical aggregation method rather than as a means of directly validating cooling effects. NDVI was obtained from the MODIS MOD13A3⁶⁵ product (NASA LP DAAC), providing monthly NDVI composites at a spatial resolution of 1 km. Canopy height data were derived from the ETH Global Canopy Height 2020⁶⁶ dataset, which provides global forest canopy height estimates at 10 m resolution based on spaceborne lidar observations. Aboveground biomass was obtained from the ESA Climate Change Initiative (CCI) Biomass product⁶⁷, providing global aboveground biomass density estimates at 100 m resolution. All datasets were aggregated to a common 1 km grid for subsequent analyses. Although VQI was not calibrated against direct temperature reduction measurements, higher VQI values were consistently associated with lower bird heat exposure under multiple greening scenarios, supporting its interpretation as an indicator of relative vegetation cooling potential.

The VCF characterizes the percentage of canopy cover on the ground surface. It reflects the proportion of space occupied by canopy in a given area. The estimation of VCF relies on data from the MODIS product MOD44B v006, offering annual estimates of canopy cover on a global scale⁶⁸. The computation of VCF can be summarized as follows:

$$VCF=\frac{A_{vegetation}}{A_{pixel}}\times 100\%$$ 5

where A_vegetation is the actual area covered by canopy within a given pixel, and A_pixel represents the total area of the pixel. This approach allows us to provide an estimate of the canopy cover for each pixel, ranging from 0% (no vegetation) to 100% (complete coverage).

Distance to water bodies

We estimated distance to water bodies using the global surface water distribution data from the European Commission Joint Research Center (https://global-surface-water.appspot.com/download), at a 30-m spatial resolution⁶⁹. Water occurrence is a measure of the frequency of water presence, expressed as the percentage of available observations identified as water over time. The variable used in our model is defined as follows:

$$d_w=\frac{1}{{D_w}^2}$$ 6

where D_w is the distance from each pixel to water. In this model, we normalized D_w between 0 and 1 to maintain this variable within the same range as the other predictor variables. We estimated the distance from all non-water body pixels to water bodies worldwide using Python 3.10.

Data aggregation

Based on data from the United Nations Population Division, we selected 1039 large and medium-sized cities around the world with a population of 500,000 or more²⁴. For each selected city, we delineated the urban extent using the 2018 Global Artificial Impervious Area (GAIA) dataset at a spatial resolution of 30  m⁷⁰. The resulting urban boundary polygons were designed to capture the main built-up areas and the majority of urban residents. Compared with administrative boundaries, this built-up-area-based delineation better represents the densely populated urban spaces where cooling interventions are most relevant. We converted the 1039 city polygons into 1 × 1 km rasters. All the requisite information required to run our models (i.e., the variables described above) and conduct our analyses were then spatially integrated and measured across the grid cells.

Input datasets were originally available at different native spatial resolutions. Bird diversity was first modeled on a global 0.1° grid using species distribution probability surfaces. To harmonize the bird diversity layers with other environmental variables, these 0.1° outputs were resampled to the common 1 km analytical grid using bilinear interpolation prior to any urban-scale analysis. Near-surface air temperature was derived from global climate products at approximately 1 km resolution and therefore required no further aggregation. Urban greening indicators were derived from finer-resolution remote sensing products and were aggregated to 1 km using area-weighted averaging to preserve subgrid information while ensuring compatibility with the analytical framework. For distance to water bodies (D_w), we calculated the distance from the center of the grid cell to the nearest water body on the same 1 km grid.

All the further core analyses were also conducted at a spatial resolution of 1 km. This scale is widely used in global urban climate and heat exposure studies, allowing direct comparison with existing assessments of extreme heat patterns^3,71. It is also ecologically appropriate for representing neighborhood-scale variation in urban bird communities and their thermal environments¹⁸, while maintaining global consistency and computational feasibility across the 1,039 cities included in this study.

Relationship exploration

Heat exposure of urban birds

Previous studies have characterized population heat exposure as the count of days per year when air temperature³ or LST¹⁴ surpasses a specified threshold (exposure threshold) multiplied by the population size. Bird heat exposure (BE1-3) was calculated using an exposure-based index that multiplies the number of extreme heat days by community-level biodiversity metrics; BE1-3 refers to taxonomic, functional, and phylogenetic diversity. This index reflects cumulative exposure of bird assemblages to heat events, rather than a direct biological response mechanism. The number of days considered corresponded to the average of the thresholds surpassed between daytime and nighttime values. The daytime exposure threshold was established as 35 °C. For each pixel in the city grid, we computed the number of Ta values surpassing the exposure threshold in 2020 year.

$${BE}_i={BD}_i\times Hotdays$$ 7

where ${BE}_i$ represents bird heat exposure; ${BD}_i$ represents a measure of bird diversity, i 1–3 refers to richness, Rao’s Q and Faith’s PD; Hotdays represents the number of days above the exposure threshold.

Model selection and analysis

Studies have demonstrated that increased urban greening promotes enhanced bird diversity, but the impact of increased bird diversity due to extreme air temperature mitigation by vegetation has not yet received attention. In addition to heat conditions, bird diversity is influenced by a variety of factors. Vegetation within cities provides habitats for birds, providing them with resources, such as food and shelter. At the same time, human activities within cities bring numerous pressures to birds. We constructed a SLM of bird diversity incorporating five covariates that account for a range of climatic and environmental factors: population (POP), net primary productivity (NPP), habitat quality (QUA), air quality (PM_2.5), precipitation (PRE). POP reflects direct human pressure on birds; NPP measures the abundance of food resources⁷²; QUA reflects the influence of the surrounding environment and represents the quality of the habitat that vegetation can provide⁷³; PM_2.5 concentration reflects the potential pressure of air pollution on bird health and behavior⁷⁴; PRE indirectly regulates the habitat of birds by affecting vegetation growth and water availability⁷⁵.

$$BD={\beta }_{1}UrbanGreening+{\beta }_{2}Hotdays+{\beta }_{3}CE+{\rho }_2{W}_2\times BD+\epsilon$$ 8

where BD (Bird diversity) is the dependent variable, representing a measure of bird diversity, including abundance, species richness, and CWM; β₁ is the coefficient of the direct influence of Urban Greening on BD; Urban Greening represents index to describe vegetation’s quality or quantity; β₂ is the coefficient of the direct influence of Hotdays on BD; Hotdays represents the number of days above the exposure threshold; β₃ is the coefficient of the direct influence of environmental covariates (CE) on BD; CE refers to covariates we have introduced before (POP, NPP, QUA, PM 2.5, PRE); ρ₂ is the coefficient of the spatial lag of BD, indicating the influence of spatially neighboring regions’ BD values; W₂ describes the weight matrix of spatial relationships for BD, used to quantify spatial interactions among different regions; and ε is the error term of the model. Population data (POP) were obtained from the WorldPop global population dataset⁷⁶, providing gridded estimates of human population counts at a spatial resolution of 1 km for the year 2020. These data were used to characterize human population density within urban areas. Net primary productivity (NPP) was derived from the MODIS MOD17A3 product⁷⁷, which provides annual estimates of terrestrial net primary productivity at a spatial resolution of 1 km. Habitat quality (QUA) was estimated using the InVEST Habitat Quality model, which integrates land-use and land-cover information from the MODIS MCD12Q1 product⁷⁸ with anthropogenic threat layers to produce a relative index of habitat condition, as described in Supplementary Information S.3. Air quality was represented by annual mean fine particulate matter (PM_2.5) concentrations, obtained from a global PM2.5 dataset⁷⁹. This dataset combines satellite observations, chemical transport models, and ground-based measurements to provide globally consistent estimates of ambient air pollution levels. Precipitation data (PRE) were obtained from a global gridded precipitation dataset⁸⁰, providing annual total precipitation for the study year. All spatial products were aggregated or resampled to a common 1 km grid to ensure spatial consistency across analyses.

All analyses were conducted in R (version 4.4.3). Spatial data processing and weights matrix construction were performed using the packages sf and spdep. During the model selection process, we developed 32 models and employed an AIC-based selection method to identify the most suitable model for assessing the relationship between urban greening metrics and bird diversity metrics (Supplementary Fig. 7). Models with a ΔAIC of less than 2 were deemed to exhibit similar support, and the model with the lowest AIC value was selected as the optimal specification⁸¹. Detailed AIC values and model selection results are provided in Supplementary Information S.5. Model predictive performance was subsequently evaluated using mean absolute error and R², as shown in Supplementary Fig. 8.

After proving that the heat environment is closely related to urban bird diversity, we used SEM to explore the various pathways by which vegetation increases urban bird diversity⁸². SEMs were fitted using the lavaan package in R, and standardized path coefficients were used to compare direct and indirect effects. We used the variables that appeared in the model with the lowest AIC values to construct SEMs for different green space and diversity indicators. In addition to the direct effect, we also explored the indirect effects of vegetation through reducing extreme heat, increasing food, and improving habitat quality (Fig. 3). BD was used as the response indicator and Urban Greening as the explanatory variable in this model. Hotdays, NPP, and QUA were used as mediating variables in the three pathways. Finally, we compared the direct and indirect effects of urban greening on bird diversity and identified the relative contributions of heat mitigation, food supply, and habitat improvement in the indirect pathway to better explain the coupling mechanism between vegetation ecological functions and bird diversity.

Mitigation strategy simulation

Spatial lag model and cross-validation

The former results support the hypothesis that urban greening improves bird habitats by reducing local heat stress. We now focus on the potential for increased urban vegetation to alleviate heat stress, that is, how extreme heat events will change under different green space improvement scenarios. We analyzed the impact of three different urban greening strategies on reducing the exposure of urban birds to extreme Ta, calculating this value for each 1-km pixel within each city. Considering the autocorrelation between the variables, we employed a SLM to describe the relationship between the number of extreme heat events and the green space index^83,84:

$$Y=\beta X+\rho W_{y}Y+\epsilon$$ 9

where Y = βX + ε corresponds to a standard linear regression model with a spatial weighting matrix applied to the observed variables and spatial autoregression parameters reflecting the spatial dependence inherent in the sample data, measuring the average effect of neighboring observations on the observed values. In our SLM, we normalized the inter-weighting matrix so that its rows summed to 1, while we employed a k-nearest neighbor approach with k = 8, where all neighbors are equally weighted.

We used the model in combination with a cross-validated machine learning approach to predict the number of days above the threshold for each city. Where the predictor variable X1 is the Urban Greening variable (one of VQI or VCF), X₂ is the near-surface air temperature (Ta), and X₃ is the distance to water body d_w. We conduct cross-validation for each green space variable: X_NDVI + Ta + d_w～Hotdays; X_VCF + Ta + d_w～Hotdays, a total of two models. To validate the model, we used a k-fold cross-validation scheme, which consists of three main phases: training, validation, and testing. In our dataset, we performed k-fold cross-validation using k = 5, and we evaluated the performance of the model in the training and validation subsets. We chose the values of the model parameters that optimize the R² values in the validation setup and tested the models using the relative test subset. We repeated the same analysis five times for all test subsets, and we ended up with 25 R² values for the training and validation steps and five values for the test set. We then perform a mitigation strategy analysis using the average value of the parameter β.

Mitigation strategies for heat exposure of birds

To evaluate how different urban greening strategies may reduce bird exposure to extreme heat, we developed a spatially explicit optimization framework based on the SLM results. Coefficient estimates obtained during the cross-validation stage were used to quantify the relationship between changes in urban greening indicators and bird heat exposure, while holding distance to water bodies constant at observed values.

Based on this framework, we defined three urban greening optimization strategies. Strategy 1 and Strategy 2 represent global greening approaches, in which the VQI and VCF were increased, respectively, across the entire urban area. Strategy 3 represents a targeted greening approach, in which greening interventions were applied only within heat hotspot areas identified as grid cells exceeding the 70th percentile of extreme heat event frequency. Strategy 3 was further divided into Strategy 3a targeting VQI and Strategy 3b targeting VCF.

For each strategy, five incremental greening scenarios were implemented, corresponding to increases of 1, 2, 5, 10, and 20% in the average greening indicator within the urban boundary. The optimization was conducted using a spatially explicit approach based on the global bird diversity grid intersected with urban areas. At each increment, a fixed proportion of eligible grid cells was modified according to predefined priority rules, described in the Supplementary Information. Following each incremental modification, urban temperature fields were updated, and bird heat exposure metrics were recalculated using the season-matched bird community representation.

Bird heat exposure was quantified as the product of bird diversity and the number of extreme heat days, and exposure reduction was calculated as the relative change in exposure before and after each greening increment. These procedures were applied consistently across all 1039 cities included in the analysis.

As an illustrative example, we present the application of the optimization framework for Beijing, China (Supplementary Fig. 9). This example demonstrates how global and hotspot-targeted greening increments were implemented at the grid-cell level and how corresponding changes in bird heat exposure were derived. We applied this increment to all pixels with urban values <0.85 observations and calculated the exposure reduction:

$${BE}_T={BD}_T\times Hotdays$$ 10

$$\mathrm{\Delta }{BE}_T\left(\%\right)=\frac{{BE}_T^1-{BE}_T^0}{{BE}_T^1}$$ 11

where ${BE}_T$ represents bird heat exposure; ${BD}_T$ represents a measure of bird diversity; Hotdays represents the number of days above the exposure threshold; $\mathrm{\Delta }{BE}_T\left(\%\right)$ is the exposure reduction, ${BE}_T^0$ is the total exposure value before the Value increment, and ${BE}_T^1$ is the total exposure value after the value increment.

The Beijing case is intended solely to illustrate the methodological workflow, while all comparative analyses were conducted across the full global city dataset. In subsequent analyses, we conducted incremental urban greening analyses for all 1039 cities considered in this study and performed robustness tests according to different ecoregions based on the Global Ecoregion partitioning⁸⁵. We aimed to ascertain the efficacy of all three urban greening optimization strategies in mitigating the heat exposure of birds across diverse ecoregions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

B.G. and J.W. conceived the study. Z.Y. performed modeling of bird diversity. R.C. and Z.C. performed visual analysis. J.W., R.C., Z.C., and B.G. helped perform the analysis with constructive discussions. J.W., R.C., and Z.C. wrote the first draft. B.G., C.C., and F.A.L. revised the paper.

Peer review

Peer review information

Nature Communications thanks Juan Antonio Hernández-Agüero and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

J.W. discloses support for the research of this work from the National Science Foundation of China (Nos. 32271935, 32471932).

Data availability

The bird occurrence data used in this study are available in the eBird Basic Dataset database under dataset version Nov-2022 at https://ebird.org/data/download. The avian morphological trait data used in this study are available from the Tobias Lab data portal at https://www.tobiaslab.net/data/. The nesting, trophic, and nocturnality trait data used in this study are available in the Wiley Figshare database under accession code https://doi.org/10.6084/m9.figshare.c.3306933.v1 at https://doi.org/10.6084/m9.figshare.c.3306933.v1. The global avian phylogenetic trees used in this study are available in the BirdTree database at https://birdtree.org/downloads/. The daily maximum and minimum near-surface air-temperature data used in this study are available in the Iowa State University Research Data repository under accession code 10.25380/iastate.c.6005185 at https://doi.org/10.25380/iastate.c.6005185. The normalized difference vegetation index data used in this study are available in the NASA Land Processes Distributed Active Archive Center under product code MOD13A3.006 at https://doi.org/10.5067/MODIS/MOD13A3.006. The vegetation continuous field data used in this study are available in the NASA Land Processes Distributed Active Archive Center under product code MOD44B.006 at https://doi.org/10.5067/MODIS/MOD44B.006. The net primary productivity data used in this study are available in the NASA Land Processes Distributed Active Archive Center under product code MOD17A3HGF.061 at https://doi.org/10.5067/MODIS/MOD17A3HGF.061. The land-cover data used in this study are available in the NASA Land Processes Distributed Active Archive Center under product code MCD12Q1.061 at https://doi.org/10.5067/MODIS/MCD12Q1.061. The global canopy-height data used in this study are available from the ETH Global Canopy Height 2020 database at https://langnico.github.io/globalcanopyheight/. The global above-ground biomass data used in this study are available in the Centre for Environmental Data Analysis database at https://doi.org/10.5285/02e1b18071ad45a19b4d3e8adafa2817. The surface-water data used in this study are available in the European Commission Joint Research Centre Global Surface Water database at https://global-surface-water.appspot.com/download. The gridded human-population data used in this study are available in the WorldPop database at https://www.worldpop.org/datacatalog/. The urban-boundary data used in this study are available in the iEarth DataHub under the Global Artificial Impervious Area dataset at https://data-starcloud.pcl.ac.cn/iearthdata/13. The global ground-level PM_₂.₅ data used in this study are available in the Zenodo database under accession code 10.5281/zenodo.10800980 at https://doi.org/10.5281/zenodo.10800980. The precipitation data used in this study are available in the NASA Goddard Earth Sciences Data and Information Services Center under product code GPM_3IMERGM_07 at https://disc.gsfc.nasa.gov/datasets/GPM_3IMERGM_07/summary.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77596-9.