Neighborhood-scale reductions in heatwave burden projected under a 30% minimum tree cover scenario

Abstract

Cities pursuing nature-based solutions to mitigate heatwaves need tools to estimate cooling benefits from increased tree canopy cover. This study applied the i-Tree Cool Air model and a heatwave degree day (HW_DD) metric to quantify reductions in heatwave severity if neighborhoods in 10 Italian cities achieved the recommended minimum 30% tree cover. The scenario focused on establishing functional urban forests, with additional canopy placed over permeable surfaces to enhance stormwater infiltration and evapotranspiration-based cooling. Despite dry summer conditions, the 30% tree cover scenario reduced HW_DD by a median of 34% (range: 16–84%), translating into comparable reductions (median 36%) in heatwave-related mortality for those aged 65 + . The tree cover generated new ecosystem service benefits valued at $10 million per city (range: $2–$62 million) through avoided stormwater runoff, air pollution removal, and carbon sequestration. Results consider drought constraints and potential irrigation trade-offs, including exacerbation of humid heat extremes.

Introduction

Heatwaves, already deadly and disabling, are intensifying due to human-caused changes in climate characterized by global warming and changes in land use characterized by urban heat islands^1,2. Urban heat islands occur where human activities generate heat (e.g., waste heat from machinery, such as air conditioning) and soil sealing by impervious structures traps heat and reduces the cooling typically provided by vegetation and soil water³. Heatwaves and urban heat islands can interact in self-reinforcing cycles, exacerbating vulnerabilities in energy supply and human health⁴. Such cycles occur as waste heat emissions from air conditioning systems increase outdoor air temperatures, which in turn elevate indoor temperatures through conduction, ventilation, and infiltration⁵. Luo et al.⁶ simulated how these self-reinforcing cycles produce destabilizing electricity demand surges, potentially leading to blackouts, disruptions of electrical-dependent medical services, and heightened human vulnerability.

Research is urgently needed to help cities understand and sustainably mitigate heatwaves. Nature-based solutions, particularly increasing urban forests, have emerged as strategies to naturalize urban areas, reduce urban heat islands, and mitigate heatwaves across climatic regions^7,8. Urban forests hold significant potential to reshape cities and help them meet the major economic, social, and environmental challenge of heatwaves - provided that sufficient permeable soil or pavement is available to support healthy tree growth and evapotranspiration. Evapotranspiration—the movement of water from Earth’s surface, soils, and vegetation into the atmosphere via—cools the air by diverting net radiation into latent heat flux, as energy is absorbed during the phase change from liquid to vapor, rather than into sensible heat flux that would increase air temperature⁹.

In this study, we introduce a novel modeling framework designed to accurately and efficiently estimate the heatwave mitigation benefits provided by additional tree canopy cover within functional urban forests. In our modeling, the functional urban forest means that all additional canopy was placed over permeable surfaces, enabling the coupled processes of stormwater infiltration and evapotranspiration that sustain tree health and generate green cooling services¹⁰. This study considers green cooling services that arise directly due to evapotranspiration as well as indirectly due to shading from the tree canopy, which reduces radiation heating impervious surfaces and their contribution to urban heat islands.

Cooling from urban green areas can vary widely depending on factors such as climate, weather, geography, scale of analysis, soil and vegetation conditions, and the heat metric used^{8,11,12,13,14}. Estimating the cooling services of urban forests within cities is challenging; while the most reliable measurements come from eddy covariance towers⁹, modeling approaches are often used to balance mechanistic detail with computational efficiency and broad spatial coverage. For example, Iungman et al.¹⁵ and Schwaab et al.¹⁶ leveraged the spatially distributed power of remotely sensed Landsat-8 thermal imagery (LST) to quantify urban forest cooling effects across 93 and 293 European cities, respectively, demonstrating the scalability of thermal remote sensing for large-area analyses. LST observations are limited to cloud-free conditions and represent instantaneous snapshots (e.g., ~10 am overpass time for Landsat in Europe). Moreover, LST measures surface radiative temperature rather than near-surface air temperature, which is more directly linked to human heat exposure and health impacts¹⁶. Iungman et al.¹⁵ estimated summer mean air temperatures through LST-based regressions, but this required aggregating their analysis to 500 m resolution to reduce uncertainty in evapotranspiration rates - and critically, it did not account for dynamic interactions between soil moisture reservoirs, precipitation, and evapotranspiration.

Soil–vegetation–atmosphere transfer (SVAT) models are among the most advanced and validated tools available for simulating the physical processes that govern evapotranspiration in vegetated urban environments¹⁷. For example, Rocha et al.¹⁸ used a SVAT model at pixel resolutions of 10 to 500 m to predict air temperatures and evaporative flux across neighborhoods in 14 European cities, demonstrating that their estimates agreed well with observations from high fidelity eddy covariance towers. The SVAT models couple evapotranspiration flux within the water balance to latent heat flux within the energy balance, constraining estimates to realistic physical bounds¹⁷. However, due to computational demands, Rocha et al.’s¹⁸ simulations across 14 European cities were limited to the hottest hour of a single day. Sinha et al.^19,20 made additional advances in SVAT modeling, simulating six-months of hourly cooling effects of urban forest cover for an entire warm season, at a detailed 30 m pixel resolution, across 10 US cities. To manage computational complexity, their modeling approach relied on simplifications such as simulating only extreme combinations of tree canopy and impervious cover (e.g., 0% or 100%), and interpolating intermediate scenarios. However, capturing the full heatwave mitigation benefits of increased tree canopy requires more than simulating evapotranspiration-based cooling—it also demands estimating how reduced temperatures translate into avoided mortality.

Estimating heat-attributable mortality (deaths beyond baseline rates without extreme heat) and its temperature dependence remains unresolved, with approaches varying by data availability, climate scenarios, and population vulnerability^19,21. One widely used approach for estimating heat-attributable mortality involves exposure response functions (ERFs)²². U-shaped ERFs capture mortality risk due to both low and high temperature extremes, with the temperature threshold at which mortality begins to increase—and the slope of that increase—varying according to geographic location, urban infrastructure, and population acclimatization^20,22. For studies specifically focused on mortality due to heatwaves, J-shaped ERFs may more effectively represent the mortality risk continuum at high temperatures alone In some cases, such as during the 2003 European heatwave, mortality remained stable below a certain threshold temperature and abruptly increased once that threshold was exceeded, justifying the use of a step-function ERF²³. Beyond variation in ERFs, definitions of heatwaves themselves vary considerably, with differences including the use of absolute versus statistically derived temperature thresholds that account for regional acclimatization, criteria based solely on daytime maxima versus those also considering nighttime minima, and durations ranging from one to three consecutive days^24,25.

Reductions in heat-attributable mortality due to urban forest cooling services can be estimated by combining modeled temperature changes, ideally at high spatial and temporal resolution, with ERFs tailored to population vulnerability^19,26,27. This typically involves comparing a base case scenario, which reflects existing tree canopy cover and associated temperatures, with an alternative (or counterfactual) scenario that meets a specified tree canopy cover target^15,28. The difference in modeled temperature and mortality between these two scenarios can be attributed to the cooling benefits of an expanded urban forest^15,26. A tree canopy cover target of 30% has been promoted as an evidence-based guideline associated with improved public health outcomes, including reduced risks of psychological distress, diabetes, hypertension, and cardiovascular disease^15,29. The European Commission has also recommended this 30% minimum to reduce urban heat islands³⁰. Iungman et al.¹⁵ applied this approach by modeling a 30% tree canopy cover scenario during the summer of 2015, estimating a 42% median reduction in heat-attributable mortality. The 2015 summer was considered climatologically typical for Europe, without extreme heatwave conditions, making it a useful baseline for evaluating seasonal impacts¹⁵. However, their analysis focused on seasonal mean temperatures and did not assess tree cooling effects during acute, high-impact heat events.

This study employed an innovative modeling framework to assess how achieving the widely-promoted 30% tree canopy cover within a functional urban forest could mitigate the impacts of extreme heat, offering a nature-based planning tool for urban resilience. We applied this framework to the summer of 2003—an exceptionally severe and well-documented European heatwave estimated to cause 72,210 heatwave-attributed deaths³¹—focusing on Italian cities where its Ministry of Health recorded detailed data on heatwave events and mortality among adults aged 65 and older³². To capture diverse geographic and demographic contexts, we analyzed 10 cities and their neighborhoods: Bari (21), Bologna (90), Bolzano (5), Cagliari (6), Firenze (5), Genova (9), Palermo (25), Roma (155), Torino (94), and Verona (79), totaling 489 neighborhoods. We used a SVAT model, i-Tree Cool Air³³, to simulate hourly air and apparent temperatures and identify heatwave events throughout summer 2003. Each neighborhood was modeled under a base case (median 8.3% canopy cover; range: 2–15.8%) and an alternative 30% tree cover scenario, implemented as a functional urban forest with added permeable surfaces to support evapotranspiration-based cooling. We introduced a metric—heatwave degree days—to represent the cumulative intensity and duration of heatwaves, aligned with observed drivers of heatwave-attributed mortality^23,32.

Results

Heatwave burden reduced by tree canopy

The spatial distribution of heatwave risk across city neighborhoods, expressed in heatwave degree days (HW_DD), was mapped using the i-Tree Cool Air model for the entire summer of 2003. Under the base case scenario, HW_DD values ranged from 0 in some neighborhoods of Bologna, Genova, Palermo, and Roma, to over 186 in hotspots of Torino and Verona (Fig. 1). The scale of risk ranged from 0–2.4 HW_DD (light yellow) to 186.4–270.8 HW_DD (dark red), reflecting increasing cumulative exposure to heatwave conditions.

Fig. 1: Heatwave risk is unevenly distributed across city neighborhoods.

The base case heatwave degree days (°C/day) from heatwaves in 2003 spatially distributed across neighborhoods for each of the 10 Italian cities, where darker red colors correspond with greater estimated occurrence of this risk metric for each neighborhood.

In the 30% tree canopy scenario, most neighborhoods exhibited reductions of at least 25.3 HW_DD, with some neighborhoods in Torino and Verona achieving reductions of 64.8 to 159.7 HW_DD (Fig. 2). Across all city neighborhoods, a linear relationship was observed between increased canopy cover and reduced HW_DD: for each 1% increase in canopy, there was an average decline of 0.81 HW_DD (R² = 0.52) (Fig. 3). This relationship varied spatially, with each 1% increase in canopy leading to heatwave burden reductions ranging from 0.28 HW_DD in a Bari neighborhood to 2.4 HW_DD in a Bolzano neighborhood.

Fig. 2: Reductions in heatwave risk are greatest in the hottest neighborhoods.

The alternative scenario reduction in heatwave degree days (°C/day) from heatwaves in 2003 spatially distributed across neighborhoods for each of the 10 Italian cities, where darker blue colors correspond with greater estimated reductions in this risk metric for each neighborhood.

Fig. 3: Planting more trees achieves larger reductions in heatwave degree days and risk.

Scatterplot of decrease in heatwave degree days (°C/day) with increase in tree cover (%), with trend lines for each Italian city (identified using the first three letters of their name, BOL is Bologna, and BOZ is Bolzano).The line slope, coefficient of determination, and number of points are reported, as well as an insert showing a box and whiskers of the decreased heatwave degree days distribution by city.

Heatwave duration influenced HW_DD values, as seen in Bologna and Bari, where fewer than 10 heatwave events occurred compared to a median of 32.5 HW_DD across cities (Table 1). The base case total across all cities was 779.5 HW_DD, with a median of 76.7 HW_DD and a maximum of 148.7 HW_DD in Cagliari. In the 30% canopy scenario, this total dropped to 526.3 HW_DD, a 32.5% reduction, with the median falling to 35.1 HW_DD, a 54% decrease.

Table 1 The Italian city and its number of neighborhoods (Nh), heatwave maximum apparent temperature threshold (^MATth, deg C), maximum apparent temperature 90th percentile for summer during the 1998 to 2002 period (^MAT₉₀, deg C), the increase in tree canopy cover between the base case and alternative scenario to reach 30% as the average across all neighborhoods (ΔTC, %), and then for base case and alternative (alt) scenario the 2003 heatwave (HW) events, heatwave degree days (HW_DD), and heatwave-attributed mortality

The HW_DD metric effectively quantified cumulative heat exposure, integrating both heatwave duration and intensity—key predictors of heatwave mortality documented in previous epidemiological studies^23,34. For example, in an Italian comparison between the summers of 2003 and 2012, longer heatwave duration helped explain a 12-fold difference in mortality (2704 vs. 225 deaths)³². The flexibility of this HW_DD metric allows it to incorporate alternative heat indicators such as air temperature, wet bulb temperature, or wet-bulb globe temperature, facilitating broader heat-risk assessments and resilience planning.

Reductions in HW_DD were accompanied by notable declines in both heatwave events and estimated mortality. In the base case, the 10 cities experienced 310 total heatwave events (median: 33; max: 47 in Firenze). Under the 30% canopy scenario, this dropped to 198 events, a 36% reduction, with Torino having the highest remaining count (35 events). Bologna, due to its near-threshold MAT values, saw events drop from 8 to just 1 (Table 1). These changes contributed directly to reductions in modeled heatwave-attributed mortality among adults aged 65 + .

Connections between heatwave-attributed mortality and urban heat islands

Spatial patterns of HW_DD closely aligned with patterns of heatwave-attributed mortality across urban neighborhoods. Areas characterized with high impervious surface and low tree canopy cover consistently showed elevated HW_DD and mortality, matching patterns observed during the 2003 European heatwave³⁵. In total, 1962 hospital-recorded deaths among adults aged 65+ were attributed to heatwaves across the 10 Italian cities in 2003. Using simulated HW_DD and census data, we mapped the spatial distribution of this mortality under both scenarios (Figs. 4 and 5). Heatwave-attributed mortality rates among adults aged 65+ ranged from 19.3 per 100,000 in Bari to 288.5 per 100,000 in Torino (SI Table 1). The alternative 30% tree canopy scenario was estimated to prevent 574 deaths (a 29.3% reduction), with rates falling to 9.3 per 100,000 in Bari and 230.5 in Torino. Median rates declined from 108 to 75 per 100,000. Percent reductions ranged from 16% in Cagliari to 87% in Bologna.

Fig. 4: Heatwave mortality varies across neighborhoods by population and heat.

The base case heatwave-attributed mortality in 2003 spatially distributed across neighborhoods for each of the 10 Italian cities, where darker red colors correspond with greater estimated mortality for each neighborhood.

Fig. 5: Potential reductions in mortality are greatest in the hottest, most populated neighborhoods.

The alternative scenario reduction in heatwave-attributed mortality in 2003 spatially distributed across neighborhoods for each of the 10 Italian cities, where darker blue colors correspond with greater estimated reductions in mortality for each neighborhood.

Remapping heatwave-attributed mortality per neighborhood developed areas, i.e., where citizens resided, revealed finer-scale spatial patterns, particularly in compact, high-density neighborhoods of cities like Cagliari, Genova, and Firenze (SI Figs. 1 and 2). These mortality patterns corresponded with local land cover, e.g., water bodies in western Cagliari and forested tracts in Genova and Firenze (SI Fig. 4). Supporting aerial imagery (SI Fig. 3) and neighborhood maps of impervious surface and tree cover (SI Figs. 5 and 6) further contextualized these variations.

Reductions in HW_DD and heatwave-attributed mortality were strongly correlated with decreases in impervious surface and increases in tree canopy (Figs. 6 and 7). High-resolution (30 m) apparent temperature maps for each city’s hottest day in 2003 (Fig. 8) demonstrated clear urban heat islands and localized cooling under increased canopy (Fig. 9). These high-resolution maps revealed distinct urban heat islands and localized cooling from existing and added urban forests. Across the 10 cities, maximum apparent temperatures under the base case had a median of 43.5 °C (range: 38.1–49.2 °C), with the 90th percentile median at 42.7 °C and the 50th at 40.3 °C (Table 2). Under the 30% canopy scenario, maximum apparent temperatures declined by a median of 3.8 °C (range: 1.5–5.6 °C), while the 90th and 50th percentiles fell by 3.5 °C and 2.4 °C, respectively (Fig. 9). This cooling was most pronounced in areas where dense impervious surfaces were converted to tree canopy, underscoring the roles of shading and evapotranspiration.

Fig. 6: New tree canopy covers impervious surfaces, enabling stormwater infiltration.

The alternative scenario impervious cover area without tree cover decrease (%) spatially distributed across neighborhoods for each of the 10 Italian cities, where lighter orange and yellow colors correspond with greater reductions.

Fig. 7: Tree plantings target neighborhoods with the greatest heatwave risk and mortality.

The alternative scenario tree cover area increase (%) spatially distributed across neighborhoods for each of the 10 Italian cities, where darker green colors correspond with greater increases.

Fig. 8: Peak heatwave temperatures are unevenly distributed within neighborhoods.

Spatial distribution of the base case apparent temperature or heat index (°C) for the hottest hour of 2003 in each city, with the map at a 30 m pixel resolution. This map provides a basis for understanding the variation in temperature across the neighborhoods.

Fig. 9: Cooling services from trees target urban heat islands within city neighborhoods.

Spatial distribution of the alternative scenario apparent temperature or heat index (°C) reduction for the hottest hour of 2003 in each city, with the map at a 30 m pixel resolution. This map provides a basis for understanding the variation in temperature across the neighborhoods.

Table 2 The i-Tree Cool Air model estimated heat index, i.e., apparent temperature, (°C) maximum (Max) and percentiles (90th, 75th, 50th) under the base case and the heat index reduction under the alternative scenario

Estimated mortality reductions compared favorably with previous estimates by Iungman et al.¹⁵, despite differences in assessment years and methods. Across six cities studied by both analyses (SI Table 2), median reductions in mortality were similar (36.8% this study vs. 40.7% by Iungman et al.). For all cities assessed, estimated mortality reductions were 36% (2003) and 41% (2015). Differences in absolute mortality reflect methodological choices and the specific lethality of the 2003 heatwave³⁶. Epidemiological studies confirm the elderly experienced the greatest heatwave-related mortality during the 2003 summer in Italy^37,38, particularly in northern cities like Torino and higher-altitude southern cities compared to coastal areas like Bari³⁷. In our analysis, Cagliari had the second-highest heatwave-attributed mortality rate, just behind Torino. While Iungman et al.¹⁵ did not include Cagliari, their study also found Torino among the top three cities for heat-related mortality, supporting the plausibility of our Cagliari estimate. Elevated mortality in Cagliari may reflect uncertainty in input data, absence of a formal heat prevention plan, heat-retaining inland water bodies, and reported energy blackouts^32,39. These concerns are heightened by recent extreme heat records in southern Italy, including the 48.8 °C air temperature in Siracusa, Sicily, in 2021, surpassing the previous European record⁴⁰.

Capturing stormwater to support ecosystem services beyond cooling

Achieving the 30% canopy target required a median increase of 21.8% in tree cover (range: 15–24.1%) across the 10 Italian cities, placed exclusively over permeable surfaces to enhance stormwater infiltration and evapotranspiration (SI Table 3). This vegetation expansion altered neighborhood water balances, reducing annual stormwater runoff by a median of 17% (32 mm). Compared to the base case (median annual interception: 10 mm, infiltration: 432 mm, evapotranspiration: 183 mm, runoff: 187 mm), the 30% tree canopy cover scenario increased interception by 260%, infiltration by 6%, and evapotranspiration by 24%, correspondingly lowering runoff (SI Table 4). Despite these improvements, soil moisture deficits persisted during the exceptionally dry and hot summer of 2003, limiting evapotranspiration-based cooling. Innovative green infrastructure solutions, such as rooftop rainwater harvesting or engineered soils designed for enhanced infiltration and root gas exchange, could further support stormwater management and transform excess runoff into valuable irrigation resources^41,42.

The functional urban forest scenario provided substantial ecosystem service benefits, valued at $132.8 million annually across the 10 cities—up from $47.7 million under the base scenario (Table 3). The largest gains came from avoided stormwater runoff ($71.4 million). Improved air quality contributed an additional $41.6 million annually, driven by reductions in concentrations of carbon monoxide (CO), nitrogen dioxide (NO₂), ozone (O₃), particulate matter ≤10 μm (PM₁₀), particulate matter ≤2.5 μm (PM₂.₅), and sulfur dioxide (SO₂). Enhanced carbon sequestration added another $19.7 million, directly supporting climate mitigation objectives. These estimates do not include the economic value of temperature reduction itself or the many specific ecosystem functions delivered by urban trees within the broad categories of provisioning, regulating, supporting, and cultural services⁴³. Although urban forests involve costs and potential disservices, existing syntheses suggest that net benefits typically outweigh costs two- to fourfold^43,44. Strategic expansion of functional urban forests thus provides a clear pathway toward achieving Sustainable Development Goal 11 (Sustainable Cities and Communities) by delivering critical ecosystem services and enhancing urban climate resilience⁴⁵.

Table 3 The increase in tree cover (TC Delta, m²) and estimated increase in value ($) of tree-based ecosystem services under the alternative scenario compared to the base case, from reduced air pollution by O₃, CO, NO_x, SO_x, PM2.5, and PM10 (Value Air), avoided stormwater (Value Water), and increased carbon sequestration (Value Climate) for the 10 Italian cities of this 2003 heatwave study

Discussion

The cooling effectiveness of urban trees is shaped by both macroclimate and urban microclimates. Mediterranean cities, found throughout Italy, are characterized by hot, dry summers and can exhibit lower urban tree cooling potential than humid northern cities. Schwaab et al.¹⁶, analyzing LST across 293 European cities, found average summertime cooling by urban trees ranged from 8–12 °C in Central Europe to just 0–4 °C in Southern Europe, with Italian cities averaging 7.5 °C cooling between 100% tree and 100% impervious cover. In contrast, Zhao et al.¹¹ observed that in Mediterranean cities, canopy temperature (T_v) often exceeded air temperature (T_a) during summer, and this T_v–T_a gap increased during heatwaves, which they interpreted as evidence of diminished transpiration. While their findings suggest limits to green cooling under Mediterranean drought stress, they relied on 30 m LST observations and air temperatures interpolated at 9 km resolution—scales that may obscure local microclimate dynamics and vertical decoupling within the canopy^11,16.

The microclimate created by land cover can exert strong thermal influence even within regions sharing the same Köppen climate classification. In our analysis of 10 Italian cities—spanning predominantly hot-summer Mediterranean (Csa) and humid subtropical (Cfa) climates (SI Table 4)—we found no consistent relationship between climate type and the magnitude of heatwave degree days or their reduction under increased canopy cover. This aligns with findings by Morabito et al.⁴⁶, who reported that tree cover, rather than climate class, best predicted local surface heat intensity across Italian cities. Urban form plays a pivotal role: street canyons and impervious surfaces intensify warming, while coastal breezes, vegetated interiors, and nearby water bodies can moderate urban heat^14,46,47. Even small-scale interventions such as street trees can meaningfully reduce thermal stress by combining shade and localized evaporative cooling¹². While large green spaces often anchor urban cooling strategies, targeted microclimatic design using linear plantings and strategic placement can substantially amplify cooling potential, even in cities facing strong seasonal water stress⁹.

Concerns that drought and high vapor pressure deficits suppress transpiration-based cooling are well supported by physiological studies showing that stomatal closure typically occurs before thresholds of xylem dysfunction, as trees attempt to avoid embolism^48,49. Yet, a growing body of evidence reveals that even with partial or full stomatal closure, latent heat flux does not entirely cease and there is a physiological decoupling between water loss and carbon assimilation⁴⁹. Guidolotti et al.⁹ reported eddy covariance data of sustained evapotranspiration and cooling (~1 °C lower than in wet years) in a Mediterranean urban green area during drought. Similarly, Tatarinov et al.⁵⁰ observed persistent evapotranspiration despite a large drop in canopy conductance and photosynthesis during a season of Mediterranean heatwaves. These findings caution against relying solely on thermal remote sensing to assess transpiration-driven cooling, especially when evaporative pathways and timing may not correspond with the timing and view of remote sensing. Altogether, they support the selection of species with functional drought tolerance traits such as low turgor loss point and high embolism resistance^51,52.

Tree water dynamics, governed by an integrated hydraulic system, provides a holistic framework for assessing drought stress and recovery potential. Ziegler et al.¹⁰ and Skelton⁵³ describe how trees typically transpire more water during the day than they can take up in real time, tapping into elastic and apoplastic tissue water that causes measurable stem shrinkage. This shrinkage reverses at night under sufficient soil moisture as internal reservoirs refill, allowing stem diameter to recover or even reach new maxima¹⁰. However, when drought persists and turgor recovery remains incomplete, tree water deficit accumulates and hydraulic stress deepens⁴⁸.

Crown dieback and mortality can result when xylem water potential becomes increasingly negative, causing emboli to form and spread within the vascular network. This process progressively blocks water transport and triggers hydraulic dysfunction, initially manifesting as patchy branch dieback and, under prolonged drought stress, leading to full canopy collapse⁴⁸. Marchin et al.⁵² documented such crown dieback and mortality in urban trees during an extreme heatwave, with vulnerable species showing extensive damage that persisted across seasons. These intra-daily and seasonal patterns are increasingly monitored through dendrometers, offering non-destructive, continuous insights into tree water status at micrometer scales¹⁰. Such tools may be essential for guiding species selection, evaluating urban forest resilience, or triggering timely irrigation interventions, particularly in cities where water availability is variable, but overnight stem refilling remains a viable recovery mechanism.

Irrigation presents nuanced trade-offs for mitigating urban heat; while it consistently reduces dry-bulb air temperature, its effects on human heat stress can vary across metrics, time of day, and climate. Notably, studies by Yao et al.⁵⁴, Krakauer et al.⁵⁵, and Chakraborty et al.⁵⁶ show that apparent temperature (e.g., heat index or Humidex) typically does not increase with irrigation and often declines, underscoring its utility as a stable and human-relevant heat metric. Because apparent temperature incorporates both temperature and humidity in a way aligned with thermal comfort and cardiovascular risk, its relative insensitivity to irrigation supports its use in assessing trade-offs in urban cooling strategies.

In contrast to apparent temperature, wet-bulb temperature (T_w)—a physiological limit for evaporative cooling—often increases under irrigation. T_w represents the minimum temperature to which air can be cooled via evaporation; when it exceeds ~35 °C, even healthy individuals struggle to thermoregulate, raising serious health risks^54,55. While irrigation lowers air temperature, it can raise near-surface humidity—sometimes disproportionately—leading to elevated T_w. Yao et al.⁵⁴ found that average T_w often increased more than extremes, with regional mismatches between peak irrigation and humid heat risk complicating interpretation. Chakraborty et al.⁵⁶ further distinguished between natural T_w and psychrometric T_w, finding that urban irrigation reduced the latter but not the former—largely due to differences in wind speed and surface radiation. Such distinctions highlight that T_w, though crucial, may not fully capture the complexities of heat stress without accounting for boundary layer behavior and energy flux dynamics.

Boundary layer dynamics are central to understanding how irrigation influences atmospheric heat metrics, including T_w and wet-bulb globe temperature (WBGT). Studies by Parajuli et al.⁵⁷ and Chakraborty et al.⁵⁶ show that irrigation can suppress daytime boundary layer height and reduce turbulent mixing, trapping moisture near the surface and elevating nighttime Tw and WBGT, particularly in nearby urban or fallow areas via moisture advection. Chakraborty et al.⁵⁶ noted that these dynamics differ from urbanization effects, which tend to amplify both day and night heat stress. These findings suggest that ventilation strategies—whether via street geometry, vegetation height, or mechanical mixing—may help mitigate localized humidity buildup and should be integrated with irrigation planning.

Linkages between boundary layer dynamics, soil moisture retention, and drought feedbacks are particularly important in Mediterranean contexts like the Po Valley of Italy. Valmassoi et al.⁵⁸ used high-resolution simulations of the Po Valley to show that irrigation during heatwaves prevented soil drying, sustained latent heat fluxes, and reduced peak daytime temperatures by up to 2.5–3 °C. Although some nighttime discomfort indices rose due to elevated humidity, the net effect was a dampening of heatwave feedbacks via reduced Bowen ratios and surface flux partitioning. This aligns with findings of Laaidi et al.³⁵, who attributed the 2003 European heatwave’s severity to low soil moisture, and Stott et al.⁵⁹, who found anthropogenic climate change at least doubled its probability. Together, these studies highlight how sustaining soil moisture, even via irrigation, can buffer cities against extreme heat, if managed without exacerbating nocturnal humidity.

Targeted, efficient irrigation remains a promising but context-dependent cooling tool. Broad-spray agricultural irrigation may raise nighttime T_w and WBGT, but drip and subsurface systems—applied at night and regulated by root-zone soil moisture—can sustain urban trees without increasing humidity^60,61. Our own 30% canopy scenario assumes no irrigation, but future designs could incorporate irrigation to support canopy resilience under drought⁶². When tuned to urban microclimates, such strategies may enhance tree-based cooling while avoiding humid discomfort, preserving public health, and optimizing water use in warming Mediterranean cities.

Simulation-based scenario modeling inevitably involves uncertainty, but this study followed established fit-for-purpose environmental modeling practices, using publicly available datasets and simplified physical representations to balance computational feasibility with decision-relevant accuracy⁶³. Simplifications made in our modeling enabled full-summer, city-wide scenario analysis at a neighborhood scale across multiple cities. As Barthelemy⁶⁴ notes, such models play a critical role in urban planning by supporting innovation and facilitating collaboration among scientists, policymakers, and practitioners. Iterative collaboration could refine planting precision, and model spatial resolution, to identify within-neighborhood microclimates best suited for canopy expansion.

Counterfactual scenarios estimating avoided temperatures and mortality inherently face validation challenges; outcomes with proposed intervention cannot be directly observed^19,21. Given that the high rates of observed mortality during the 2003 heatwaves did not follow standard ERFs^23,32, we used a modified approach linked to city-specific and age-specific ERFs²² to model temperature–mortality relationships, accounting for variation in regional acclimation. Definitions of heatwaves themselves vary considerably, affecting risk estimates; differences include the use of absolute or relative thresholds, treatment of nighttime minima, and required event durations^24,25. Despite these uncertainties, our estimated mortality reductions aligned closely with previous assessments¹⁵ and leveraged best-available epidemiological evidence on heat vulnerability among older adults during extreme events^32,36,37,38.

Constraints in data availability created temporal mismatches in the simulations that required assumptions and interpolations, introducing uncertainty into the outputs. Due to differences in data production cycles, land cover and census data could not be matched exactly to the 2003 heatwave year—an issue common in large-scale geospatial analyses⁶⁵. Nevertheless, cross-checks revealed that these mismatches were minor. Population differences between 2003 and 2011 were modest according to regression estimates: in four cities, 2011 census counts appear to have underestimated 2003 populations by an average of 2.9%, while in six cities they appear to have overestimated them by an average of 1.5%. The median difference across all cities was just 0.3% (SI Table 1). These differences are unlikely to have meaningfully altered total heatwave-attributed mortality, though they may have slightly shifted its spatial distribution across neighborhoods. Land cover differences due to data mismatches appear similarly modest. An analysis of CORINE Land Cover (CLC) data, trusted in land cover studies⁶², showed that fewer than 3% of pixels changed categories across the 10 cities between the nearest available years, with a median change of just 1.3% (SI Table 3).

Methods

Soil-water-vegetation-atmosphere transfer model

The i-Tree Cool Air model³³—used in this study to simulate summer 2003 heat metrics—was originally developed as the PASATH (physically based analytical spatial air temperature and humidity) model and is now distributed as part of the i-Tree HydroPlus modeling suite. This analysis used revision 1535 of the model. The model has been previously validated in Napoli, Italy, both at multiple meteorological stations during extreme heat conditions⁶⁶ and against eddy covariance tower data for sensible and latent heat fluxes during both wet (2015) and dry (2017) summer periods⁹. For each timestep, the model assimilates observed 2-m air temperature and humidity from a station-containing pixel to constrain the upper boundary layer conditions. It then propagates these conditions spatially by resolving land cover–dependent energy partitioning (sensible vs. latent heat fluxes) and aerodynamic coupling, generating 2-m air temperature and humidity estimates across the full modeling domain.

In addition to atmosphere–energy partitioning, the i-Tree Cool Air model uses terrain elevation (see SI Fig. 7) to laterally redistribute subsurface water and represent soil wetness based on the topographic index and TOPMODEL theory⁶⁷. It operates in coordination with the i-Tree Hydro model⁶⁸, also included in HydroPlus, which provides a statistically distributed representation of topographic index bins (typically 30) to simulate temporally varying soil moisture deficit (SMD; comparable to groundwater table depth) across the domain using hydrologic similarity⁶⁹. Following Pace et al.⁷⁰, we calibrated the model using data from two full water years (October 1, 2002 – September 30, 2004), which capture the typical annual moisture cycle and allow water budget closure at a common subsurface moisture condition⁷¹. Calibration involved first running i-Tree Hydro to estimate three groundwater parameters, initial groundwater drainage rate, transmissivity at saturation, and hydraulic conductivity decay scaling parameter, based on an average SMD for each city (see SI Table 5). These calibrated values were then passed to i-Tree Cool Air for spatially explicit simulation of water and energy fluxes at the pixel level.

To simulate the 2003 summer period (June 1 – August 31), we initiated the model on January 1, to allow for adequate spin-up and equilibrium of soil moisture values and continued the simulation through December 31 to analyze the annual water budget. To reduce computational demands, energy balance calculations were limited to a single representative pixel for each of the 489 neighborhoods across the 10 cities. Each representative pixel was assigned the neighborhood-average land cover properties, including tree canopy and impervious surface area. Unlike Pace et al.⁷⁰, who coarsened pixel resolution from 30 m to 300 m for some i-Tree Cool Air simulations, we maintained a 30 m pixel resolution for all simulations. To visualize spatial patterns in cooling, we also applied the i-Tree Cool Air model across all city pixels (over 2.5 million in total) for a 24 h period on each city’s hottest day—comparable in scale to the SVAT simulations of Rocha et al.¹⁸. These high-resolution simulations used soil moisture values from the corresponding representative-pixel summer runs, enabling us to map and visualize spatial variability in cooling at the neighborhood scale.

Model scenarios

The study simulated the year 2003 with its extreme summer heatwave events under two scenarios: a base case and an alternative scenario with a 30% minimum tree canopy cover (see SI Table 3). The base case represented existing land cover conditions, including tree canopy cover, impervious cover, and the fraction of tree canopy overlapping impervious and pervious cover. The alternative scenario applied a nature-based solution consistent with the concept of a functional urban forest. In this scenario, additional tree canopy was assumed to be planted over permeable soils or permeable pavement to allow stormwater infiltration and root access to water. Where impervious surfaces without tree canopy were available, the simulation made those surfaces permeable pavement to enable the coupled infiltration and evapotranspiration processes needed to support the new tree cover. If insufficient impervious area was available in each pixel, the remaining canopy needed to reach 30% was assigned to areas with soil and short vegetation cover. No additional tree cover was added to pixels that already met or exceeded 30% canopy in the base case.

Land cover and meteorological data

Elevation and land cover data were obtained from the European Copernicus Land Monitoring Service (CLMS), using 30 m resolution datasets from the 2018 Urban Atlas, Tree Cover Density, and Imperviousness Density products⁷². Although the study focuses on 2003, we used 2018 CLMS land cover data due to the absence and known limitations in the accuracy of earlier datasets, particularly the 2012 products, which underestimated tree canopy and impervious surface coverage. To assess potential land cover changes between 2003 and 2018, we referenced the European CORINE CLC product, available for 2006 and 2018, and trusted for data scarce periods or regions⁶². We mapped these CLC polygons (with five categories of land cover: artificial surfaces, agricultural areas, forests and semi-natural areas, wetlands, and water bodies) to the 30 m resolution domain.

Classes of land cover required by the SVAT model were obtained from the CLMS Urban Atlas and translated into U.S. National Land Cover Database (NLCD) classes using a validated crosswalk method⁶⁶. These converted classes were assigned to the base case. In the alternative scenario, only pixels categorized as developed (e.g., residential, commercial, or industrial) were targeted for land cover modification—by adjusting tree canopy and impervious surface percentages to meet the 30% tree cover target while maintaining realistic urban land characteristics.

Several forest and urban characteristics of the study sites were derived using standard i-Tree methods⁷³. The tree cover over impervious surfaces, which indicates tree canopy extending over roofs, sidewalks, roads, plazas and other impervious features, was estimated using photointerpretation with i-Tree Canopy, achieving a standard error of less than 5%. The proportion of evergreen tree cover was calculated as the ratio of CLMS evergreen tree cover to total tree cover. Leaf-on and leaf-off dates, which represent the average Julian day of 50% leaf out and 50% leaf drop for deciduous trees, respectively, were determined based on frost dates and confirmed through consultation with a professional Italian forester.

The directly connected impervious area (DCIA) represents the portion of total impervious area (TIA) that drains directly to stormwater systems or receiving waters, bypassing infiltration into adjacent permeable surfaces. DCIA was estimated using the empirical relationship DCIA = a(TIA)^b, based on the method of Sutherland³⁷. Here, both DCIA and TIA are expressed as fractions of the total modeled area, and the coefficients a and b are determined by land cover characteristics as described in Rossman et al.³⁶ and elaborated by the U.S. EPA³⁸. For this study, we used a = 0.1 and b = 1.5, representing standard sewer connections and residential rooftops not directly connected to drains.

Meteorological data for each study site were obtained from nearby airports using the U.S. National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information (NCEI) Integrated Surface Database⁷⁴. City-level, spatially averaged precipitation time series were retrieved from NASA’s Giovanni platform using the GPM_3IMERGHH precipitation product⁷⁵.

Heatwave metrics and mortality

Heatwave events were defined using the EuroHEAT project²³ criteria, based on both daytime and nighttime apparent temperatures to capture the full health impacts of sustained heat exposure. Apparent temperature—also known as heat index—combines air temperature and humidity to reflect how hot conditions feel to the human body. According to the EuroHEAT definition, a heatwave occurs when either (1) the maximum apparent temperature (MAT) exceeds the 90th percentile of a city’s historical record for two consecutive days, or (2) both the daily minimum air temperature exceeds its 90th percentile and the MAT exceeds its 50th percentile. These percentile thresholds are calculated using a multi-year reference period, such as 1998–2002 for EuroHEAT.

The heatwave daily MAT threshold, MAT_threshold, was iteratively determined for each city to ensure that the 2003 NOAA weather station time series used for the simulations contained the number of heatwave events reported by the Italian Ministry of Health³² (Table 1). These MAT_threshold values were the 90th percentile daily MAT, computed using the NOAA weather station records for the summer periods (June 1 to August 31) from 1998 through 2002. The same records were used to compute the 50th percentile daily MAT and the 90th percentile daily minimum temperature. Next, the number of heatwave events was counted for the 2003 summer period, each event occurring when the daily MAT reaches MAT_threshold for two consecutive days or both the daily minimum air temperature exceeds its 90th percentile and the MAT exceeds its 50th percentile. If the number of heatwave events did not equal the reported number of events, then the MAT_threshold was increased or decreased until the heatwave event value was equaled. The difference between the MAT_threshold and the 90th percentile MAT was attributed to the Italian Ministry of Health using different weather station records or data value rounding. The same MAT_threshold was used for the base case and alternative scenario.

Heatwave events and associated excess mortality, referred to as heatwave mortality or heatwave-attributed mortality, during the summer of 2003 were reported for 15 Italian cities using a real-time estimation system that accounted for lagged effects between heat exposure and mortality, with totals summed to represent the full summer season³². For Cagliari, only partial data was available, and its number of heatwave events and heatwave-attributed deaths were estimated using the median values reported across the 15 cities. This yielded an estimated heatwave mortality rate in Cagliari of 266 deaths per 100,000 people aged 65 or older, positioned between recorded rates in Torino (288 per 100,000) and Firenze (179 per 100,000). To obtain the Iungman et al.¹⁵ estimated reduction in heat-attributed mortality for the Italian cities (SI Table 2), we used their reported values for preventable age-standardized mortality rate (deaths/100,000 inhabitants) and the 2011 Census data. While this study used the reported heatwave-attributed mortality for the model simulations, as shown in the equations below, we also reported these values as mortality rates using the 2011 Census counts to facilitate comparison with other studies (see SI Table 1).

For each heatwave event, the heatwave degree day, HW_DD, metric was used to quantify the cumulative intensity and duration of the summer heatwave events, defined as,

$${{HW}}_{{DD}}=\mathop{\sum }\limits_{i}^{N}\left({{MAT}}_{i}-{{MAT}}_{{threshold}}\right)$$

(1)

for each day i that contains a heatwave, under the base case and alternative scenario. The HW_DD is computed for each neighborhood, and those values are averaged to obtain a representative HW_DD for each city; standard deviations could also be presented to capture the variation in HW_DD across the city. The alternative to base case HW_DD ratio was used to scale the observed heatwave events, HW_{events_b} used for the base case to an expected number of heatwave events under the alternative scenario, HW_{events_a}, as,

$${{HW}}_{{events}\_a}={{HW}}_{{events}\_b}\left({{HW}}_{{DD}\_a}/{{HW}}_{{DD}\_b}\right)$$

(2)

The summer total heatwave deaths for each city were distributed across neighborhoods for the base case as a function of the neighborhood’s fraction of the city population 65 y and older and its weighted base case HW_DD. This was implemented for the base case using each city’s total observed heatwave mortality, Mort, as,

$${Mort}_{b\_q}=Mort[({HW}_{DD\_b\_q}\,.{Pop65frac}_{q})/\mathop{\sum }\limits_{q}^{Q}({HW}_{DD\_b\_q}\,.{Pop65frac}_{q})]$$

(3)

where ${{HW}}_{{DD\_b\_q}}$ is the neighborhood q ${{HW}}_{{DD}}$ under the base case b, ${{HW}}_{{DD\_b}}$ is the sum of all ${{HW}}_{{DD\_b\_q}}$, and ${{Pop}65{frac}}_{q}$ is the neighborhood q fraction of the city population 65 y and older. The Mort was obtained with a step function ERF, allocating all summer 2003 heatwave-attributed mortality to each city if it reached the observed heatwave events. While the Italian Ministry of Health observed a linear relationship between summer 2003 heatwave mortality and MAT, their attempt to fit various J-shape ERFs did not yield results that matched the reported number of deaths³². Simulated HW_DD for each neighborhood under the alternative scenario were used in relation with the base case values of HW_DD and Mort_{b_q} to estimate the corresponding alternative scenario heatwave deaths, Mort_{a_q} as

$${{Mort}}_{a\_q}={{Mort}}_{b\_q}\left[{{HW}}_{{DD}\_a\_q}/{{HW}}_{{DD}\_b\_q}\right]$$

(4)

Census data

Census data on neighborhood residents aged 65 and older were obtained from the 2011 Population Census conducted by ISTAT, the Italian National Institute of Statistics and Eurostat partner⁷⁶. For the city of Bolzano, these data were provided by the Provincia Autonoma di Bolzano⁷⁷. The census data were delivered in polygon feature files, where the smallest unit was the sezione di censimento (census section), each containing population counts. These sections were spatially aggregated to match the neighborhood polygons used in our study. Although the census was conducted in 2011, it was used to represent the 2003 population because no census data were available for 2003, and topological flaws within the 2001 data polygons prevented aggregation to the neighborhood scale. Additionally, estimated population change between the 2003 and 2011 was relatively small at the city level. To estimate population in 2003, we developed a linear regression based on city-level census counts from 2001 and 2011. Four cities experienced a population decline over this decade, while the other six grew.

Ecosystem services valuation

Ecosystem service values were estimated based on prior i-Tree Eco simulations conducted for all Italian municipalities using 2015 data on air pollution, meteorology, and tree canopy cover. These simulations produced pollutant removal rates for the six pollutants of CO, NO₂, O₃, PM10, PM2.5, and SO₂, expressed in metric ton removed per square meter of canopy cover. The monetary value of air pollution removal was based on pollutant-specific externality values, originally developed for the U.S. context, and adjusted to 2023 values using Italy’s Producer Price Index (PPI). The externality value was obtained from the i-Tree Eco database⁷³ for Italy and set CO at $1,272/tonne/yr, NO₂ at $13,065/tonne/yr, O₃ at $13,065/tonne/yr, PM10 at $4,768/tonne/yr, PM2.5 at $238,215/tonne/yr, and SO₂ at $11,919/tonne/yr.

The valuation methods, described in Nowak et al.⁷⁸, and adjusted for Europe⁷⁹, estimated removal rates and monetary values based on local data, including leaf area index, evergreen tree percentage, weather conditions, pollution concentrations, and population. The temporal mismatch between data required a presumption of comparable removal rates. Carbon sequestration per square meter of canopy cover was determined in the same i-Tree Eco simulations, where storage was calculated by multiplying modeled tree biomass by 0.5, and annual carbon sequestration was estimated by simulating tree growth and taking the difference in biomass between successive years⁷³. The economic value of carbon sequestration is based on the 2016 U.S. Interagency Working Group estimate of the social cost of carbon, PPI adjusted to $330/tonne/yr. The value of runoff avoided was set to $2.36/m³yr, using a representative value for the costs of stormwater control and treatment as reported in sixteen research studies⁷³.