Global field observations of tree die-off reveal hotter-drought fingerprint for Earth's forests

Forests are central to global ecosystem services and human economies, serving as keystone habitats and major drivers of the water and carbon cycles while supporting biodiversity and storing nearly half of terrestrial carbon. Historical forests dominated by large, old-growth trees are especially vital. Anthropogenic change, including wildfires, deforestation, and especially hotter droughts, threatens these systems. Although hotter drought is widely associated with tree mortality events (pulses exceeding background mortality), a quantitative, globally scalable determination of common climate drivers has been lacking due to the absence of a precise record of where and when mortality has occurred. This gap limits validation of models projecting substantial future vulnerability of forests to die-off. Prior extrapolations rely on experiments, single-site or regional analyses, and process-based models that need better representation of mortality processes. Rising temperatures present a triple threat to tree survival: amplification of atmospheric drought (higher vapor pressure deficit, VPD), intensified soil drought, and direct heat stress. Higher temperatures increase VPD, accelerating water loss even when stomata are closed. Warming increases frequency, severity, and intensity of chronic soil droughts, and severe drought and heat can impair plant function and be lethal. Hotter droughts create a trade-off for trees between evaporative cooling and conserving dwindling water reserves, making them especially vulnerable when atmospheric and soil droughts coincide. The frequency of co-occurring extremes of heat and drought has increased over the past century due to anthropogenic warming, altering the character of droughts and potentially exceeding historical climatic limits of local forest adaptation, leading to amplified mortality. The authors hypothesize that such climate exceedance has emerged in recent decades and can be detected via a set of hotter-drought metrics—a global hotter-drought fingerprint—linked to field-observed tree die-off. The study aims to: (1) establish a precisely geo-referenced global database of climate-induced tree mortality; (2) quantify a global hotter-drought fingerprint; and (3) assess how the frequency of climate conditions associated with die-off changes under further warming.

The authors synthesize prior work indicating widespread associations between hotter drought and tree mortality, including experimental manipulations (e.g., rainout shelters, chambers, greenhouse studies), site- to regional-scale climate analyses, and process-based models from physiological to Earth system scales. These studies collectively suggest significant future vulnerability of forests but lack globally validated thresholds tied to documented mortality events. The paper builds on and updates references from four recent reviews of drought and heat-induced mortality, noting persistent uncertainties in representing mortality mechanisms, the role of compound hot-dry events, and biome-specific responses, especially in under-studied boreal and tropical rainforests.

Database construction: The authors compiled a global database of precisely geo-referenced field observations of drought- and/or heat-induced tree mortality (excluding fire) from 154 peer-reviewed studies (1970–2018), yielding 1,303 plots aggregated to 675 unique locations at ~4 km resolution (TerraClimate grid). Inclusion criteria: on-the-ground observations of pulses of mortality (significantly above background), attribution to heat/drought, and precise geolocation (≤1 km or accurately inferable). Geolocation came from published coordinates (n=248) or author data requests (n=1,055). Studies relying solely on remote sensing or broad inventory networks without precise attribution were excluded. Climate data: Monthly climate and hydroclimate data (1958–2019) were obtained from TerraClimate at 1/24° (~4 km) resolution. Six variables were analyzed: monthly average maximum temperature (TMAX), vapor pressure deficit (VPD), climatic water deficit (CWD), soil moisture (SOIL M), monthly total precipitation (PPT), and Palmer Drought Severity Index (PDSI). Mean annual precipitation and temperature (1970–2000) were used to assign Whittaker biomes; elevation was obtained via elevatr. Hotter-drought fingerprint definition: For each site, the climatologically warmest/driest months were identified for each variable (months with highest TMAX, VPD, CWD and lowest SOIL M, PPT, PDSI) over 1958–2019. Anomalies for each variable were computed as deviations from the long-term mean for the same month; anomalies were standardized as z-scores using the 1958–2019 period for cross-site comparability. The mortality year was defined as the onset year reported in source studies. Z-scores were computed for the mortality year and for a nine-year window (±4 years around onset) to assess antecedent and lagging conditions. Linear regressions tested trends in mortality-year anomalies through time. Warming scenarios: Pseudo-global warming scenarios were generated by pattern scaling TerraClimate (1985–2015) to +2 °C and +4 °C above pre-industrial (1850–1879) global mean temperature, with an empirical correction to PET to account for CO2-driven changes in water-use efficiency and surface resistance. For each site and scenario, the frequency (years per decade) when all six variables met or exceeded that site’s mortality-year thresholds during the warmest/driest months was calculated, representing the local hotter-drought fingerprint exceedance frequency. Statistical summaries: Site-level z-score means and 95% CIs were computed globally and by biome (Whittaker classes). Additional analyses included the proportion of sites where 4, 5, or all 6 variables exceeded mortality-year anomalies and temporal trends in TMAX anomalies for mortality years versus all years.

- Global hotter-drought fingerprint: In the mortality year, climate conditions during the warmest/driest months were significantly hotter/drier than the long-term mean across all six variables. Mean z-scores (±SE): TMAX +0.370 ± 0.04, VPD +0.300 ± 0.04, CWD +0.490 ± 0.04; PPT −0.210 ± 0.03, SOIL M −0.390 ± 0.03, PDSI −0.730 ± 0.04 (sign flipped so positive is hot/dry). - Temporal window: The year prior to mortality onset also showed hotter/drier anomalies (weaker than the mortality year), and the year after tended toward hotter/drier conditions. Years outside this 3-year window had smaller departures, indicating episodic hot droughts drive the signal. - Biome coverage: The fingerprint was detected across all biomes individually except tropical rainforest, likely due to limited sample size and climate data uncertainties there. - Site-level concurrence: In the mortality year, all six variables were anomalously hot/dry at 24% of sites; at least five variables at 47% of sites; at least four at 69% of sites. - Trends: Since 1970, mortality-year conditions became warmer/drier at study plots. TMAX anomalies during mortality years increased faster than background warming across all years. - Frequency under warming: Frequency of mortality-year fingerprint conditions increases nonlinearly with warming. Observed climate (1985–2015; +0.7 °C) showed 1.62 ± 0.08 years per decade. Under +2 °C: 1.97 ± 0.07 years per decade (+22%). Under +4 °C: 3.88 ± 0.10 years per decade (+140%). - Dataset scope: 1,303 plots at 675 locations globally across all tree-supporting biomes, spanning >30 °C mean annual temperature and >4 m annual precipitation gradients, elevations from sea level to 3,488 m.

The analysis directly links ground-based, geo-referenced observations of tree die-off to concurrent hot and dry climate anomalies, providing a quantitative global hotter-drought fingerprint. This fingerprint addresses the core question by identifying common climate conditions associated with mortality across diverse biomes, and shows that episodic hot-dry extremes over a 3-year window around mortality onset are key drivers. The consistency across most biomes suggests broad vulnerability of forests to compound heat–drought stress; the absence of a strong signal in tropical rainforests likely reflects data limitations and potentially different climatic drivers (e.g., prolonged dry seasons). The accelerating frequency of fingerprint conditions with warming implies heightened future risk of forest die-off, challenging the reliability of forests as carbon sinks. While mechanistic pathways (e.g., hydraulic failure, carbon metabolism disruption) are not resolved at each site, the empirical approach demonstrates the practical value of linking observed mortality to climate extremes for model validation and risk assessment. The findings underscore that limiting global warming can substantially reduce the frequency of lethal climate conditions threatening forests.

The study establishes the first precisely geo-referenced, global, ground-based database of climate-induced tree mortality and uses it to quantify a global hotter-drought fingerprint. It demonstrates that mortality years are characterized by concurrent hot and dry anomalies and that the frequency of such conditions increases nonlinearly under additional warming. The work provides an empirical foundation for validating Earth system and mortality models and for monitoring global tree mortality. Future research directions include: identifying additional acute and chronic climatic signals (including duration and seasonality), integrating extensive inventory networks, applying remote sensing for broader extent and near-real-time monitoring, benchmarking Earth system models via hindcasting, elucidating potentially unique drivers in tropical rainforests, quantifying how die-off severity scales with warming, and expanding monitoring in under-represented boreal and tropical forests. Limiting warming to +2 °C would keep the frequency of lethal climate conditions to less than half of that projected at +4 °C, with decisive implications for forest persistence and global carbon and biodiversity outcomes.

- Sampling bias: Database is presence-only and biased toward the Northern Hemisphere and regions near well-funded research institutions; boreal and tropical rainforests are under-sampled. - Tropical rainforest signal: Sparse mortality observations and uncertainties in spatially interpolated climate data hinder detection of the fingerprint in tropical rainforests; different climate drivers (e.g., extended dry seasons) may be more relevant there. - Event timing: Onset of mortality can lag environmental drivers, and chronic droughts can predispose mortality with lags, complicating precise temporal attribution. - Variable concurrence criterion: Requiring all six variables to meet/exceed mortality-year thresholds is a stringent filter that likely underestimates frequency of risky conditions. - Temporal resolution: Use of monthly data may miss critical shorter-term (heatwaves) and longer-term (multi-year drought) drivers. - Excluded disturbances: Interacting agents (insects, pathogens, wind, lightning, wildfire) that amplify mortality under warming were excluded, potentially underestimating risks. - Excluded data sources: Remote-sensing-only studies and broad inventory networks lacking precise attribution/geolocation were not included, limiting spatial coverage. - Model scenario simplifications: Warming scenarios use pattern scaling and PET adjustments; while appropriate for broad patterns, they may not capture all local-scale processes or extremes.