Continental-scale impact of bomb radiocarbon affects historical fossil fuel carbon dioxide reconstruction

Abstract

Historical reconstructions of fossil fuel carbon dioxide emissions are vital to setting emission mitigation strategies, and tree-ring radiocarbon analysis provides a reliable approach. However, residual radiocarbon from past nuclear tests can distort the signal, and regional variations beyond the well-known latitudinal gradient remain underexplored. Here we analyzed tree-ring radiocarbon records from Nanling (1921–2020, this study) and published datasets to identify spatial patterns of bomb-derived fallout in China from 1965 to 1985. Compared to latitudinal compilations and European backgrounds, radiocarbon enhancements extended up to 3000 km from test sites, highlighting the dominant role of source characteristics. We further demonstrate that re-release of bomb radiocarbon from terrestrial ecosystems prolongs its regional atmospheric residence time to 5–20 years, introducing spatial variability that biases fossil fuel carbon dioxide reconstructions up to 5.4 ppm. These findings underscore the need to integrate source-specific fallout and atmospheric transport into regional radiocarbon interpretation frameworks.

Introduction

Fossil fuel carbon dioxide emissions (CO_2ff) are a primary cause of global warming¹. Accurate historical CO_2ff are crucial for understanding the global carbon cycle and informing climate policy². This information provides a baseline for measuring current and future emission reduction efforts, offers insights into the effectiveness of past policies and industrial practices in reducing emissions, and can be used to refine climate models that predict future climate change scenarios^3,4. Tree-ring radiocarbon (¹⁴C) is a valuable proxy for reconstructing historical CO_2ff levels, because it reflects the carbon isotope composition influenced by fossil fuel combustion^5,6,7,8. However, nuclear weapons testing in the mid-20th century introduced a substantial amount of ¹⁴C into the atmosphere, creating a distinct bomb spike that is evident in tree-ring records worldwide. This complicates the interpretation of tree-ring data^{9,10,11,12,13,14,15}.

Most of the bomb ¹⁴C was directly injected into the lower stratosphere, and then moved into the troposphere after approximately two years. This resulted in the creation of noticeable latitudinal gradients of tropospheric ¹⁴CO₂^10,16,17. Latitudinal and hemispheric baselines are widely used to isolate the bomb ¹⁴C signal^7,18,19,20. For example, Hua et al.¹⁰ divided the Northern Hemisphere into three zones, low-to-middle latitude (NH-3), middle latitude (NH-2), and mid-to-high latitude (NH-1), based on the ¹⁴C values of atmospheric samples and tree rings collected at clean-air sites. Each zone is assumed relatively uniform ¹⁴C levels, with tropospheric bomb ¹⁴C levels decreasing progressively from north to south¹⁰. However, there may be a continental-scale impact of bomb ¹⁴C due to the complex composition of explosion fallout. Low-yield bomb tests (<1 Mt) could contribute to the tropospheric ¹⁴CO₂ and influence its levels in local and regional air masses^9,10. Additionally, explosions produced by fusion (such as hydrogen bombs) generate markedly more ¹⁴C than those produced by fission (like atomic bombs)²¹. Numerous low-yield fission and fusion tests were conducted at sites such as Novaya Zemlya, Semipalatinsk, Nevada, and Lop Nor, and their documented total partitioned fission yields in the troposphere, local and regional air were 1.1, 3.1, 1.4, and 0.9 Mt, respectively²². Therefore, we hypothesize that post-bomb ¹⁴C has continental-scale impacts and could vary regional ¹⁴C levels, and subsequently influencing on the estimation of historical CO_2ff.

To verify this hypothesis, we analysed the ¹⁴C fluctuations in East Asia and Europe, i.e., the continents where most aboveground detonations occurred. Our analysis was based on tree-ring data from Nanling (113.05°E, 24.94°N) spanning the period 1921−2020, which was collected specifically for this study, as well as data reported in the literature. We demonstrated the impact of these fluctuations on historical CO_2ff reconstructions through observations from Shenzhen (114.07°E, 22.56°N), the Hengshan station (112.73°E, 27.31°N), and data reported in previous studies. Through this multi-regional and multi-decadal assessment, we provide evidence for the long-range and persistent influence of nuclear fallout on atmospheric radiocarbon records, highlighting the dominant role of source characteristics in shaping regional radiocarbon patterns.

Results and discussion

Continental-scale impact of bomb ¹⁴C

We identified higher Δ¹⁴C levels at background sites in China from 1965 to 1985 compared to the latitudinal zone compilations defined by Hua et al¹⁰. (ΔΔ¹⁴C_zone; Fig. 1a) and European background site (ΔΔ¹⁴C_eu; Fig. 1b), including Vermunt (47.07°N, 9.57°E; 1959–1983; Levin et al.²³) and Jungfraujoch (1986–2015; Hammer and Levin)²⁴. The background sites in China were distributed across the country, including Huangzhong in the northwest (36.27°N, 101.67°E; Xiong et al.²⁵), Shangdianzi in the north (40.65°N, 117.12°E; Niu et al.¹¹), Dailing in the northeast (47.50°N, 129.27°E; Dai et al.¹²), Mingyin in the southeast (27.21°N, 100.33°E; Dai and Fan¹⁴) and Nanling in the south (this study; Supplementary Fig. 1)²⁶. The ΔΔ¹⁴C_zone values for these sites were 28.3 ± 14.8‰, 24.5 ± 18.4‰, 9.9 ± 32.3‰, 67.6 ± 22.8‰ (limited to 1965–1967), and 5.3 ± 12.9‰, respectively. The ΔΔ¹⁴C_eu values for these sites were 26.6 ± 8.2‰, 30.6 ± 23.9‰, 21.2 ± 27.5‰, 26.8 ± 29.2‰ (limited to 1965–1967 available), and 3.6 ± 13.8‰ (8.1 ± 8.4‰ during 1968-1985), respectively. Before 1968, Nanling’s ΔΔ¹⁴C_eu exhibited negative values, likely due to latitudinal gradients between Nanling (24.94°N; within NH-3 zone) and Vermunt (47.07°N; within NH-1 zone), potentially masking regional radiocarbon enhancement (Fig. 1b). The stations in China relied on tree-ring sample data, whereas European stations used atmospheric observations, and the latitudinal compilation data combined smoothed tree-ring and atmospheric measurements. For the comparison, we specifically selected atmospheric and compilation data corresponding to the tree-ring growth period, thereby minimizing the influence of temporal variations. A known challenge in interpreting tree-ring ¹⁴C records is the physiological effect caused by the remobilization of stored non-structural carbohydrates during earlywood formation, which can introduce offsets in the measured Δ¹⁴C values, especially during the bomb spike period²⁷. However, this effect is species-dependent and substantially reduced in evergreen conifers, such as pine and spruce, whose intra-annual ¹⁴C signals remain well-aligned with atmospheric trends²⁷. Accordingly, all Chinese records used in this study are based on evergreen conifers: spruce at Huangzhong, Minyin, and Dailing; Chinese red pine (Pinus tabuliformis) at Shangdianzi; and Guangdong pine (Pinus kwangtungensis Chun ex Tsiang) at Nanling.

Fig. 1: Continental-scale impact of bomb ¹⁴C.

The impact of bomb ¹⁴C from Lop Nor: ΔΔ¹⁴C values at different Chinese background stations relative to either latitudinal zone compilations (a) or European atmospheric backgrounds (b). Specifically, black squares indicate Huangzhong (~1300 km from Lop Nor)²⁵; pruple diamonds represent Mingyin (~2000km)¹⁴; red circles correspond to Shangdianzi (~2400 km)¹¹; blue upward triangles denote Nanling (~3000 km, this study); and green downward triangles indicate Dailing (~3200 km)¹². Vertical bars indicate the annual yields of nuclear bomb tests²², with orange representing fission tests and green representing fusion tests conducted at the Semipalatinsk (left light blue shaded region) and Lop Nor (right lilac shaded region) test sites. Solid lines represent linear trends in ΔΔ¹⁴C for Huangzhong (black) and Shangdianzi (red), fitted from 1965 onward. The horizontal gray shaded band denotes a ± 10‰ range in ΔΔ¹⁴C, with the central gray solid line indicating 0‰. Vertical dashed lines mark the year when each fitted trend line (black for Huangzhong, red for Shangdianzi) crosses the +10‰ threshold. The latitudinal compilation data includes NH-3, NH-2, and NH-1 data from Hua et al.¹⁰. The European backgrounds include Vermunt data (1959–1983) and the Jungfraujoch data (JFJ, 1986–2015)^23,24. All the Chinese stations use tree-ring samples, with comparisons conducted during the tree-ring growth periods: March–October at Nanling and Mingyin, May–August at Huangzhong, and May–September at Shangdianzi and Dailing. c The impact of bomb ¹⁴C from Novaya Zemlya: ΔΔ¹⁴C at different European background stations relative to either latitudinal zone compilations or European atmospheric backgrounds. Black squares represent Fruholmen (air, ~1100 km from Novaya Zemlya)²⁸; red circles denote Trondheim (air, ~2100 km)²⁸; blue upward triangles show Trøndelag (tree-ring, ~2100 km)³⁷; green downward triangles represent Lindesnes (air, ~2650 km)²⁸; purple diamonds indicate Eastern Jutland (tree-ring, ~2700 km)³⁸; and goldenrod left-pointing triangles denote Niepołomice (tree-ring, ~3100 km)³⁶, each compared to the corresponding latitudinal zone compilations. Relative to the Vermunt/JFJ European background: cyan circles represent Fruholmen; dark brown hexagons, Trondheim; dark khaki stars, Trøndelag; orange pentagons, Lindesnes; cronflower blue diamonds, Eastern Jutland; and green right-pointing triangles, Niepołomice. Annual air sampling data are shown as means with standard error bars, while tree-ring data are based on species-specific growing seasons (May–August for Trøndelag and Niepołomice, and June–August for latewood in Eastern Jutland). Vertical bars indicate fission (orange) and fusion (green) bomb test yields at Novaya Zemlya (light blue shaded region)²². Cyan solid line represents linear fit of Fruholmen ΔΔ¹⁴C (relative to Vermunt/JFJ). Vertical gray dashed lines mark the year when the fitted trend line crosses the +10‰ threshold. Gray shading and other annotations follow the conventions of (a).

To assess the statistical significance of the differences, we performed T-tests or Wilcoxon signed-rank tests depending on the results of Shapiro–Wilk normality tests (Supplementary Table 1). The results indicated that the Δ¹⁴C levels at Huangzhong and Shangdianzi were significantly elevated compared to both the latitudinal and European baselines (adjusted p < 0.05). In contrast, Dailing exhibited a significant increase only relative to European baselines but not latitudinal compilations. Mingyin exhibited no statistically significant differences against either baseline after multiple comparison adjustments, likely due to limited data (n = 3). For Nanling versus NH-3, the unadjusted p-value indicated nominal significance (p = 0.0099), but the association showed only borderline significance after correction (adjusted p = 0.0993). However, Δ¹⁴C levels in Nanling were significantly elevated relative to the European baseline during 1968–1985 (adjusted p = 0.0021), supporting the presence of a persistent historical radiocarbon signal. To evaluate whether the smoothing effects inherent in the latitudinal compilations were responsible for the observed Δ¹⁴C elevation at Chinese stations, we conducted a case study comparing the Nanling data with individual site measurements from the three stations that define the NH-3 baseline: Doi Inthanon (Thailand; Pinus kesiya tree-ring samples; Hua et al.¹⁵), Debre Zeit (Ethiopia; air samples; Nydal and Lövseth²⁸), and Mandla (India; Tectona grandis tree-ring samples; Murphy et al.²⁹). Although the overall comparison between Nanling and the NH-3 composite yielded only borderline significance after correction, pairwise comparisons with individual NH-3 sites revealed statistically significant Δ¹⁴C enhancements at Nanling relative to both Doi Inthanon (adjusted p = 0.0036) and Debre Zeit (adjusted p = 0.0179), with corresponding mean differences of 17.5 ± 24.3‰ and 25.4 ± 8.4‰, respectively (Supplementary Table 2). The comparison with Mandla was not statistically significant (adjusted p = 0.5461), likely due to the limited sample size (n = 4) and potential signal distortion from remobilized stored carbohydrates in Tectona grandis tissue²⁷. These results indicate that the observed Δ¹⁴C elevation at Nanling cannot be fully explained by smoothing effects in the latitudinal NH-3 baseline and reflect a distinct regional signal.

Global post-bomb Δ¹⁴C values typically follow latitudinal gradients, with similar values within the same latitudinal zone and lower Δ¹⁴C in southern regions compared to northern baselines such as Vermunt (47.07°N) and Jungfraujoch (46.55°N). However, our study reveals notably deviations from this expected pattern in China during 1965–1985, primarily driven by nuclear tests at Lop Nor, China (40°N, 90°E) from 1964 to 1980. Over this period, Lop Nor hosted 22 nuclear tests, including six high-yield detonations that injected substantial ¹⁴C into the stratosphere and 16 low-yield tests that released large amount of ¹⁴C into the troposphere²². Notably, six low-yield tests were conducted on the land surface, with fallout from their fireballs dispersing ¹⁴C into local and regional tropospheric air masses²². Atmospheric trajectory simulations confirmed that tropospheric air masses from Lop Nor could reach continental China, influencing regional Δ¹⁴C levels (Fig. 2a–c)³⁰. Our conclusions are supported by Niu et al.¹¹ and Dai et al.¹², who attributed the post-1964 ¹⁴C enhancements at Shangdianzi and Dailing to Lop Nor’s nuclear tests.

Fig. 2: Trajectory frequencies of air masses during major nuclear test periods.

a–c Trajectory frequencies at Huangzhong (HZ), Shangdianzi (SDZ), and Naling (NL) from 27 October to 26 November 1966, following the Lop Nor bomb test on 27 October 1966. d, e Trajectory frequencies at Huangzhong and Naling from 22 August to 19 November 1957, during the Semipalatinsk bomb tests conducted between 22 August and 26 September 1957. f Trajectory frequencies at Fruholmen from 5 August 1962 to 1 January 1963, corresponding to the Novaya Zemlya tests carried out from 5 August to 25 December 1962. In these maps, red five-pointed stars represent nuclear test sites, including Lop Nor, Semipalatinsk, and Novaya Zemlya, while red triangles denote the locations of sampling stations. The background shading represents the frequency of air mass trajectories, with light cyan indicating >0.1%, light green >1%, blue >10%, and orange >100% (not present in this figure but included for legend consistency). The trajectory frequencies were calculated using HYSPLIT version 5.3.0, based on meteorological data from the National Oceanic and Atmospheric Administration (ftp://arlftp.arlhq.noaa.gov/archives/reanalysis). The maps are drawn by MeteoInfo version 3.9 (http://www.meteothink.org/). The boundary of China was sourced from map products provided by the National Geomatics Center of China (https://www.webmap.cn/). Other country boundaries were derived from the Natural Earth public-domain vector dataset (ne_10m_admin_0_countries.shp, https://www.naturalearthdata.com/).

Elevated Δ¹⁴C values after 1964 at sites closer to the Lop Nor nuclear test site, such as Huangzhong (~1300 km), compared to more distant sites like Nanling (~3000 km), demonstrate the regional impact of these tests (Fig. 1a, b). This spatial pattern, with higher Δ¹⁴C near the test site, reflects the direct input of bomb-derived ¹⁴C, amplified by proximity to the source. No other aboveground nuclear tests occurred globally during this period, ruling out additional bomb-related ¹⁴C sources. Although atmospheric transport processes influence Δ¹⁴C distribution, our study demonstrates that the characteristics of Lop Nor nuclear tests—location, yield, and type—primarily govern regional Δ¹⁴C patterns, as evidenced by spatial gradients and atmospheric trajectory simulations^10,31.

To confirm that Lop Nor’s influence overshadows other factors, we evaluated alternative drivers of Δ¹⁴C anomalies, including anthropogenic sources (CO_2ff and nuclear power plants) and natural processes (cosmogenic ¹⁴C, oceanic exchange and biogenic respiration). Based on data from the Emissions Database for Global Atmospheric Research, China’s CO_2ff emissions (910 to 1900 Mt CO₂; 1970−1990) were considerably lower than Europe’s (3500 to 3900 Mt CO₂; mean ratio = 2.7 ± 0.8)³². If CO_2ff emissions were a dominant factor, China should exhibit slower decay (longer ecological half-lives). However, the half-lives were nearly identical: 11.2 ± 0.1 years (China, averaged over Nanling, Huangzhong, and Shangdianzi) and 11.0 years (Europe; May–September; Supplementary Table 3). Within China, the spatial variability in ΔΔ¹⁴C is not driven by CO_2ff emissions. Despite similar CO_2ff levels, Nanling exhibits lower ΔΔ¹⁴C than Huangzhong and Shangdianzi (Fig. 1a, b), indicating that CO_2ff-driven dilution does not account for regional differences. More critically, while CO_2ff dilution typically acts uniformly across broad regions, ΔΔ¹⁴C displays a pronounced spatial gradient, decreasing with distance from Lop Nor: Huangzhong (~1300 km; ΔΔ¹⁴Czone = 28.3 ± 14.8‰, ΔΔ¹⁴Ceu = 26.6 ± 8.2‰), Shangdianzi (~2400 km; 24.5 ± 18.4‰, 30.6 ± 23.9‰), Dailing (~3200 km; 9.9 ± 32.3‰, 21.2 ± 27.5‰), and Nanling (~3000 km; 5.3 ± 12.9‰, 8.1 ± 8.4‰). This gradient correlates strongly with distance from Lop Nor rather than fossil fuel intensity, providing compelling evidence that a regional fallout source is the primary driver of ΔΔ¹⁴C variability. Similarly, no operational nuclear power plants existed near the observation sites before 2000, ruling out nuclear facilities as a ¹⁴C source. Cosmogenic ¹⁴C produces a minor north-to-south Δ¹⁴C gradient (<10‰) in the Northern Hemisphere, insufficient to explain the elevated Δ¹⁴C at Chinese sites located at lower latitudes than European baselines³³. The ocean acted as a ¹⁴C sink before the 2000s, and while reduced oceanic exchange in inland regions could slightly elevate Δ¹⁴C, this effect is minimal^34,35. For instance, Shangdianzi (~200 km from the Bohai Sea) shows higher Δ¹⁴C than European sites like Vermunt (~250 km from the Mediterranean), despite comparable distances from the sea. Biogenic ¹⁴C, derived from atmospheric ¹⁴C, is re-emitted through ecosystem respiration but typically has a minimal impact on global atmospheric Δ¹⁴C levels³⁴. Absent regional ¹⁴C sources like nuclear fallout, biospheric fluxes alone cannot account for the elevated Δ¹⁴C values observed at Chinese sites compared to global reference curves (e.g., NH-1, NH-2). These factors collectively fail to account for the observed Δ¹⁴C anomalies, underscoring that Lop Nor’s source characteristics predominantly shape regional Δ¹⁴C patterns.

The statistically significant Δ¹⁴C elevations at Chinese sites, coupled with their spatial gradient relative to Lop Nor, confirm that regional nuclear fallout from the Lop Nor tests was the dominant driver of the observed anomalies. By demonstrating that source-specific factors outweigh atmospheric transport and alternative factors, our study establishes a first-principle framework: the source of ¹⁴C primarily governs its regional distribution. This approach enhances our understanding of Δ¹⁴C variability in China during the post-bomb era and provides a robust model for interpreting regional Δ¹⁴C patterns globally.

The elevated Δ¹⁴C levels in China confirms the continental-scale impact of bomb ¹⁴C, extending beyond the expected latitudinal gradient. This influence is evident not only from Lop Nor but also from other bomb testing sites. Xiong et al.²⁵ reported that nuclear fallout from the Semipalatinsk (50.1°N, 78.7°E; 1949–1962) enriched Δ¹⁴C levels at Huangzhong (~2400 km away) during 1957-1964, with a ΔΔ¹⁴C_zone of 28.1 ± 19.7‰ and adjusted p-value of 0.0098 (Fig. 1a, Supplementary Table 1). A similar pattern was observed at Nanling (~4000 km away from Semipalatinsk) during 1954−1964 with a ΔΔ¹⁴C_zone = 12.2 ± 19.9‰ (raw p = 0.034; adjusted p = 0.1371). In contrast, ΔΔ¹⁴C_zone values at Nanling remained within 5.0‰ prior to 1954 (Fig. 1a). Atmospheric trajectory analysis indicates that the air mass from Semipalatinsk dispersed to these sites (Fig. 2d, e). A similar pattern is observed in Europe, where nuclear tests at Novaya Zemlya (54.75°N, 73.38°E; 1955–1962) had a broad continental impact (Fig. 2f)²². To evaluate this influence, we analyzed clean-air datasets from NH-1 zone background stations, as filtered by Hua et al.¹⁰, located in rural areas 1100 km (Fruholmen)²⁸ to 3100 km (Niepołomice)³⁶ from Novaya Zemlya, with minimal interference from local CO_2ff or nuclear facilities. The comparison reveals substantial Δ¹⁴C enrichment, with enhancements reaching up to 40‰ in annual records and 110‰ in monthly records during 1961–1965 for sites within 2700 km of Novaya (Fig. 1c). Statistical comparisons show significant Δ¹⁴C enhancements at Fruholmen²⁸ and Lindesnes²⁸ relative to both NH-1¹⁰, and Vermunt²³ references (adjusted p < 0.05), supporting the presence of regional nuclear fallout (Supplementary Table 4). Among the datasets, three sites were based on tree-ring records: two used Pinus sylvestris (Scots pine) from Trøndelag³⁷ and Niepołomice³⁶, while the third analyzed latewood of Quercus (oak) from Eastern Jutland³⁸, which is minimally influenced by the remobilization of stored carbohydrates²⁷.

Despite these regional Δ¹⁴C elevations, we found that ΔΔ¹⁴C values are not significantly correlated with annual bomb test yields, due to the complexity of ¹⁴C sources and measurement limitations. Tree-ring Δ¹⁴C data, integrating atmospheric signals over months, smooth out seasonal peaks. High-yield tests (>1 Mt) injected ¹⁴C into the stratosphere, causing delayed tropospheric mixing (~2 years), while low-yield detonations (<1 Mt) enhanced regional signals. Variations in ¹⁴C production between fusion and fission tests, combined with atmospheric transport, further modulate these signals, leading to sustained regional ¹⁴C elevations rather than immediate spikes.

Residence time of bomb ¹⁴C in the regional troposphere

Our study reveals that bomb ¹⁴C remains in the regional troposphere for 5−20 years, driven by proximity to test sites, detonation intensity and type, atmospheric transport, and ecological conditions. We found longer ¹⁴C tropospheric residence times at sites closer to Lop Nor, with Huangzhong (~1300 km), Shangdianzi (~2400 km), Nanling (~3000 km), and Dailing (~3200 km) showing distinct Δ¹⁴C patterns after 1964 (Fig. 1a, b). Linear regression analyses of ΔΔ¹⁴C at these sites revealed:

• Huangzhong: Elevated ¹⁴C levels persisted for 20 years after Lop Nor tests ceased, until 2000 (r < –0.6, p < 0.001), with ΔΔ¹⁴C_zone = 20.1 ± 13.2‰ and ΔΔ¹⁴C_eu = 17.9 ± 10.3‰, stabilizing within ±10‰ thereafter²⁵.

• Shangdianzi: Positive ΔΔ¹⁴C values lasted until 1990 (r < –0.8, p < 0.001; ΔΔ¹⁴C_zone = 21.5 ± 15.7‰, ΔΔ¹⁴C_eu = 24.6 ± 20.8‰), dropping to –6.9‰ in 1996 due to CO_2ff dilution from the Beijing-Tianjin-Hebei region, suggesting prolonged positive values absent this effect¹¹.

• Nanling and Dailing: Located farther from Lop Nor, ΔΔ¹⁴C levels remained within ±10‰ before 1985, indicating minimal bomb ¹⁴C influence.

During 1991–2000, the effective half-lives of Δ¹⁴C decline were 14.1 years at Huangzhong, 8.4 years at Shangdianzi, and 13.9 years at the European background sites. The shorter half-lives at Shangdianzi reflects stronger CO_2ff influence from regional industrialization, whereas the similarity between Huangzhong and European values suggests a relatively limited CO_2ff impact. Between 2000 and 2015, Δ¹⁴C declined linearly at rates of 5.2‰ year⁻¹ at Huangzhong, 5.7‰ year⁻¹ at Shangdianzi, and 4.6‰ year⁻¹in Europe. The faster declines observed at the Chinese sites indicate growing CO_2ff emissions and their increasing influence in the post-2000 period.

It should be noted that tree-ring samples at these Chinese sites were collected from forested areas, where locally respired CO₂ trapped beneath the canopy may have biased Δ¹⁴C measurements relative to the free tropospheric background. Although this does not alter the overarching conclusion regarding the continental-scale radiocarbon impact of nuclear testing, it may partly contribute to the apparent extension of half-lives inferred at these sites.

Higher Δ¹⁴C levels at Fruholmen (~1100 km from Novaya Zemlya) compared to Vermunt/Jungfraujoch (~3100 km) persisted until 1982 (r < –0.7, p < 0.001; Fig. 1c). These findings are unaffected by canopy effects, as the stations were specifically designed for atmospheric monitoring and are located in open environments, avoiding localized biases. We propose that the residence time of bomb ¹⁴C in regional ecosystems depends on the intensity of the nuclear explosions, the distance from the bomb test site, and the local carbon cycling dynamics. The long residence times observed at Huanghzong (~20 years) and Fruholmen (~20 years) can be partly attributed to their proximity to test sites. Biosphere carbon turnover also plays a critical role. It has been demonstrated that most state-of-the-art vegetation models underestimated post-bomb ¹⁴C accumulation in vegetation biomass, particularly in short-lived non-woody plants³⁹. These plants, with their shorter carbon storage duration and rapid carbon turnover, release bomb ¹⁴C back into the atmosphere earlier and more substantially than previously estimated, thereby prolonging regional tropospheric bomb ¹⁴C residence time³⁹. Furthermore, ecological characteristics such as vegetation cover, temperature, and humidity influence the timing and periodicity of ¹⁴C re-release, affecting regional ¹⁴C dynamics. In temperate and tropical ecosystems, ¹⁴C re-emission occurs earlier than in high-latitude regions due to higher temperatures, longer growing seasons, and faster litter decomposition and soil respiration³¹. In addition to its farther distance from the bomb test site, the subtropical station Nanling exhibited a shorter residence time for ¹⁴C compared to the boreal ecosystem station Huangzhong (Fig. 1a). Similarly, at Fruholmen (71.1°N, 23.98°E), the ¹⁴C residence time extended to approximately 1982, largely due to the cold environmental conditions, which delayed carbon cycling and prolonged the residence time of ¹⁴C, even 20 years after testing ceased (Fig. 1c).

Our study reveals that bomb-derived ¹⁴C persists in the regional atmosphere and biosphere for 5–20 years, as evidenced by measurable Δ¹⁴C enhancements like Huangzhong and Fruholmen relative to global background baselines. This prolonged residence time primarily reflects proximity to test sites, such as Lop Nor and Novaya Zemlya, with secondary contributions from biospheric re-emission and atmospheric transport, rather than isolated air masses. Given the complex factors affecting regional tropospheric ¹⁴C residence time during the post-bomb period, future research should focus on observing and simulating bomb ¹⁴C cycling with a higher spatial resolution. The increasing global CO_2ff emissions after 2000 have rapidly diluted tropospheric ¹⁴C, by adding ¹⁴C-free CO₂, lowering the atmospheric ¹⁴C/C ratio. This enhances the ¹⁴C concentration gradient between the atmosphere and terrestrial/oceanic reservoirs, accelerating ¹⁴C release through soil respiration, plant respiration, and ocean-atmosphere CO₂ exchange⁴⁰.

Impact of post-bomb ¹⁴C and background selection on historical CO_2ff reconstruction

Considering the continental-scale impact and residence time of bomb ¹⁴C, selecting an appropriate background is crucial for accurately reconstructing urban and regional CO_2ff history. While CO_2ff levels are evaluated as the enhancement relative to a specified background, using a well-chosen baseline minimizes biases and ensures the reconstructed values more accurately reflect actual CO_2ff, providing a closer approximation to the true signal. However, due to a lack of regional Δ¹⁴C measurements, European backgrounds and zonal compilations have been widely used, such as in China, Europe and Japan^7,18,19. To quantify the influence of background selection, we calculated the historical CO_2ff at four Chinese stations using three different backgrounds: zonal compilations (NH-1, NH-2, NH-3), European backgrounds (EU, from Vermunt and Jungfraujoch), and Chinese regional backgrounds (Nanling, Huangzhong, and Shangdianzi)^{10,11,23,24,25,41}. The four stations included three urban locations (Shenzhen at 22.56°N (this study), Xi’an at 33.70°N¹⁸, Beijing at 39.73°N⁴²) and one regional station (Hengshan Mountain at 27.31°N, Hunan province; this study).

We found that historical CO_2ff would be underestimated without using regional backgrounds (Fig. 3), and closer to the nuclear explosion site a greater bias is shown. The bias decreased over time. Correspondingly, the rate of historical growth of CO_2ff would be overrated (Fig. 3). For Xi’an, CO_2ff was underestimated by 4.5−5.4 ppm during 1991−2000 without regional backgrounds, but the bias decreased to 2.2−3.4 ppm after 2000 (Fig. 3a, Supplementary Table 5). In Beijing, CO_2ff was underestimated by 4.3−5.4 ppm before 1990 without using Shangdianzi as the regional background, but then decreased to 2.2−3.3 ppm during 1991–1995 (Fig. 3b, Supplementary Table 5). Shangdianzi became polluted by regional CO_2ff after 1995, potentially due to the relatively low sampling height (287 m above sea level). Therefore, the comparison was limited to data collected up to 1995.

Fig. 3: Comparison of CO_2ff estimations using various background selections.

a–d CO_2ff estimations for Xi’an¹⁸, Beijing⁴², Shenzhen (this study), and Hengshan (this study), respectively, based on regional (Huangzhong (HZ)²⁵, Shangdianzi (SDZ)¹¹, Nanling (NL, this study)), latitudinal (NH-1, NH-2, NH-3)¹⁰, and European (EU)^23,24 background references. Each box plot represents the distribution of CO_2ff values for a specific time period and background type. The colored boxes indicate the interquartile range (25th–75th percentile), with whiskers extending to 1.5 times the interquartile range (IQR). Horizontal black lines within the boxes represent the median values, black squares denote the mean values, and black stars indicate outliers. The blue dashed lines indicate the fluctuations in mean values, while the pink dashed lines depict the changes in median values. Time periods vary by panel and reflect available data ranges for each location and background. Each time segment is represented by a distinct box color: pink for the earliest period, blue for the second period, yellow for the third period, and green for the most recent period.

We find that there was a small bias (<0.8 ppm) in Shenzhen during 1984–1990, which was due to the low enrichment and short residence time of bomb ¹⁴C at Nanling. However, this situation subsequently changed due to the rapid increase in CO_2ff in China. Because the EU and zonal backgrounds do not account for the continental effect of anthropogenic activities in China, the estimated CO_2ff in southern China would have been overestimated and this discrepancy worsened over time (Fig. 3c, d). During 2011−2020, the calculated CO_2ff levels in Shenzhen (Fig. 3c) and Hengshan (Fig. 3d) were 7.6 ± 1.8 and 4.4 ± 2.4 ppm, respectively, when using Nanling as the regional baseline, which were 3.7−3.9 ppm lower than estimates based on the EU and NH-3 baselines (Supplementary Table 5). This difference is comparable to the total emissions of a medium-sized city such as Indianapolis, United States of America⁴³. Furthermore, the growth rates of CO_2ff are overestimated (Fig. 3), leading to a misunderstanding of urban and regional carbon emission trends and, consequently, resulting in misjudgements regarding carbon mitigation measures.

Overall, the use of regional backgrounds is strongly recommended when reconstructing historical CO_2ff, because it reduces the biases introduced by inappropriate background choices and produces estimates that better reflect true CO_2ff. In this study, we quantified the impact of background selection at sites in China, where post-bomb ¹⁴C dynamics were uniquely influenced by regional nuclear tests and rapid growth in CO_2ff. Future studies should include additional sites from other continents or countries, enabling diverse geographical locations to be considered.

Conclusions

Our study offers important insights into reconstructing CO_2ff emissions using tree-ring Δ¹⁴C data, particularly by addressing the post-bomb ¹⁴C effect from nuclear weapons testing. We identified key challenges in isolating the fossil fuel-derived ¹⁴C signal from the bomb-derived ¹⁴C signal, a crucial step for accurate historical CO_2ff reconstructions. Our analysis revealed pronounced regional variations in Δ¹⁴C levels, especially near nuclear test sites like Lop Nor, underscoring the dominant influence of source characteristics, such as location, strength, and type, in shaping regional Δ¹⁴C patterns. These findings emphasize that bomb ¹⁴C distribution is shaped not only by atmospheric transport processes but also source-specific factors. We also confirmed the continental-scale impact of bomb detonation, consistent with global bomb-affected areas. Furthermore, the re-emission of bomb ¹⁴C from terrestrial ecosystems prolongs its tropospheric residence time, contributing to region-specific Δ¹⁴C signatures. Consequently, applying inappropriate regional baselines can bias historical CO_2ff estimates. To enhance the robustness of CO_2ff reconstructions, we recommend generating additional region-specific Δ¹⁴C curves globally.

By integrating the continental-scale impact of bomb ¹⁴C, source-specific characteristics, and precise regional baselines, we can enhance the accurate of CO_2ff reconstructions. This ensures climate policies and emission reduction strategies rest on robust scientific foundations. Our study marks a crucial step in refining the historical understanding of anthropogenic carbon emissions, supporting more effective global efforts to combat climate change.

Methods

Sample collection

We collected tree-ring samples from three locations: the regional background station Nanling, the urban station in Shenzhen, and the regional station in Hengshan.

In August 2020, samples from well-grown Pinus kwangtungensis Chun ex Tsiang (Guangdong pine) were taken at an elevation of 1200 meters above sea level in the south-central part of Nanling National Forest Park (113.05°E, 24.94°N). Nanling National Forest Park is located in the northern part of Guangdong Province, covering an area of 273 km². This region falls within the subtropical monsoon humid climate zone. According to weather data recorded from 1965 to 2019 at a nearby meteorological station, the monthly mean temperature ranged from 20 to 28 °C, with an average of 20 ± 6.8 °C. The historical monthly precipitation averaged 147 ± 82 mm, ranging from 50 to 282 mm. Based on the monthly precipitation and temperature variations (Supplementary Fig. 2) and the authors’ experience, the growing season in the research area spans approximately from March to October. Increment cores were collected from trees at breast height using hollow drills with a 5.15 mm diameter. To build a long tree-ring chronology and minimize non-climate signals, around 20 trees with no obvious injury or disease were selected for sampling. Two tree cores were taken from each tree, drilled through the pith center, and then placed into prepared plastic straws and sealed for storage. Sampling information such as station, tree number, and diameter at breast height was recorded during the process. In total, approximately 40 core samples were retrieved from Nanling in this study.

In February 2023, we obtained tree-ring cores from Pinus elliottii Engelm. in the city center of Shenzhen (114.07°E, 22.56°N; at an elevation of 21 m above sea level). A total of 18 core samples were obtained from 10 individual trees. In April 2023, we gathered samples from Cunninghamia lanceolata (Lamb.) Hook. in Hengshan National Nature Reserve (112.73°E, 27.31°N; at an elevation of 858 meters above sea level) in Hunan Province. This area is located in southern China and features a subtropical monsoon mountainous humid climate, with an average annual temperature of 11.3 °C and a relative humidity of 82%. Due to the difficulty of extracting tree cores from this species, we obtained only four unbroken samples.

Tree growth periods can vary with latitude, temperature, water availability, and species differences. While this variability may introduce uncertainties in single-year growth conditions, the long-term trends captured in tree-ring data minimize their impact on the conclusions of this study. Additionally, comparisons across multiple sites and latitude bands further mitigate the influence of site-specific variations.

Crossdating

The tree-ring samples were handled in accordance with standard dendrochronological procedures⁴⁴. We polished the samples with progressively finer sandpaper to smooth the surface until the ring boundaries were clearly distinguishable under the microscope (Supplementary Fig. 3). Based on careful visual crossdating between trees, a calendar year was preliminarily assigned to each growth ring. Ring widths were subsequently measured using a Lintab 6 ring analyzer and the supporting TSAP software (Frank Rinntech, Heidelberg, Germany) with a resolution of 0.001 mm. The COFECHA program was used to check the quality of the crossdating, excluding the possibility of false rings or missing rings⁴⁴. Finally, the ARSTAN program was applied to standardize ring-widths, eliminating the growth trend related to age and growth fluctuations of individual trees.

Crossdating was successful for the Nanling and Shenzhen stations, but failed for Hengshan due to the limited number of samples. However, given the high humidity at Hengshan, the possibility of missing growth rings is minimal, and the boundaries of annual rings were quite clear. Therefore, we included the samples from Hengshan in this study, with carefully recognizing the annual rings based on measured ring widths.

Sample preparation

We selected tree cores for preparation based on ring width and year span. Annual growth rings were separated using a scalpel from the selected cores, and samples from the same year were combined to obtain sufficient material for ¹⁴C analysis. We obtained annual samples spanning 100 calendar years from 1921 to 2020 at Nanling, 38 years at Shenzhen (1984–2022), and 34 years at Hengshan (1988–2022). Every ten tree samples coupled one parallel samples. These samples were cut into very small pieces to ensure uniformity, and then treated to extract α-cellulose following the method described by Michczyńska et al.⁴⁵ α-cellulose is a very stable polymer that remains unchanged and immobile over long time periods, thus representing the isotopic composition of the original plant material.

The major α-cellulose extraction procedures include the following steps:

Base-Acid-Base-Acid (BABAB) Treatment

This step removes the alcoholic, phenolic, and carboxylic groups from the main wood components.

Bleaching

The samples are bleached with NaClO₂ to obtain holo-cellulose.

Extraction of α-Cellulose

A sequence of treatments with a concentrated sodium base is used to extract α-cellulose.

For detailed procedures, please refer to the MSZ method described by Michczyńska et al.⁴⁵.

The α-cellulose samples were graphitized using the AGE3 (Automatic Graphitization Equipment)⁴⁶. AGE3 is a third-generation fully automatic graphitization system developed by Ionplus, a Swiss company, designed to prepare organic carbon samples ranging from 0.2 to 1.0 mg in size. Subsequently, the graphite samples were analysed for ¹⁴C using the Accelerator Mass Spectrometer (AMS) at the Guangzhou Institute of Geochemistry, Chinese Academy of Sciences (GIGCAS).

Calculation of Δ¹⁴C

The ¹⁴C levels were represented as Δ¹⁴C, which was fractionation- and age-corrected deviation from the standard pre-industrial ¹⁴C content in atmospheric CO₂. Δ¹⁴C values were calculated based on the following formula provided by Stuiver et al.⁴⁷:

$${\Delta }^{14}{{{\rm{C}}}}=\left(\frac{{{{{\rm{A}}}}}_{{{{\rm{SN}}}}}{{{{\rm{e}}}}}^{\lambda (1950-x)}}{{{{{\rm{A}}}}}_{{{{\rm{ON}}}}}}-1\right)\times 1000\textperthousand$$

(1)

where ${{{{\rm{A}}}}}_{{{{\rm{SN}}}}}$ and ${{{{\rm{A}}}}}_{{{{\rm{ON}}}}}$ is the $\frac{{14}_{C}}{{12}_{C}}$ ratios of sample and oxalic acid-II reference standard. $\lambda$ is the decay constant of ¹⁴C ($\lambda =\frac{1}{8267}\,{{{{\rm{year}}}}}^{-1}$). $x$ is the formation year of sample.

Quality Assurance/Quality Control

To ensure the reliability and consistency of tree-ring Δ¹⁴C measurements, rigorous quality assurance and quality control procedures were implemented. Tree-ring dating was performed using crossdating techniques, which are widely recognized in the dendrochronological research community, to precisely identify the growth year of each ring. After dating, microscopic cross-sectional cutting was conducted to isolate individual rings, minimizing overlap with adjacent years.

To further reduce variability, all samples were prepared by trained personnel following standardized protocols across sites. Firstly, all tree-ring samples followed the specific protocols described in the previous sections. Secondly, reference materials including a ¹⁴C-free wood blank (AVR, >55 kyrs BP)⁴⁸ and two types of secondary standard materials (IAEA-C5, FIRI-F)^49,50 were used for background corrections and quality control throughout the entire process, mirroring the procedures for the tree-ring samples (α-cellulose extraction, graphitization, and AMS analysis). In principle, for every ten tree-ring samples, coupled one AVR, one IAEA-C5, and one FIRI-F. Thirdly, parallel samples were processed to check reproducibility, with 13 parallel pairs in total for Nanling. The standard Δ¹⁴C deviation of the replicates ranged from 0.1 to 4.5‰.

Fourthly, for the graphitization and AMS testing procedures, phthalic anhydride was used as the blank, while oxalic acid-II and IAEA-C7 were used as secondary standards⁴⁹. Each AMS wheel was monitored by 13 replicates of phthalic anhydride, oxalic acid-II, and IAEA-C7, respectively, for approximately 90 α-cellulose samples. Accuracy and precision for these procedures were better than 3.0‰. Fifthly, to ensure the quality of AMS analysis, ten to twelve runs were carried out for each AMS wheel. The precision of individual tree-ring AMS analyses ranged from 1.7 to 2.8‰.

A pooled standard deviation calculation determined that replicated measurements are within the 4.5‰ level or better. The instrumental error of AMS is better than 3.0‰.

AMS System at GIGCAS

The AMS measurements in this study were conducted using the CAMS-500 AMS at the GIGCAS. This system, manufactured by National Electrostatic Corporation was installed in April 2014, marking a major milestone in the development of high-precision radiocarbon dating in China⁵¹. The system’s Model 1.5SDH pelletron accelerator operates at a terminal voltage of 500 kV, and features a 134-sample multi-cathode source for analyzing graphite samples. The system uses a Canberra Passivated Implanted Planar Silicon solid-state detector to accurately detect ¹⁴C ions, ensuring high-precision measurements.

The GIGCAS AMS system has undergone extensive development over nearly a decade, with continuous improvements made to its performance. Hardware optimizations such as realigning the AMS equipment in 2018, which increased the accelerator transmission efficiency from 37% to 41–42%, and stabilizing the stripping gas pressure with temperature-controlled cooling water, have measurably enhanced the system’s precision. These improvements have also contributed to a stable stripping gas pressure ( ± 0.2 µTorr), essential for consistent performance. In addition, the AGE3 automated graphitization equipment, installed in 2016, facilitates the efficient preparation of samples, making GIGCAS the first lab in China to operate this system.

The system’s reliability is further ensured by rigorous tuning techniques and data processing protocols, which are based on methodologies used at KECK AMS Laboratory at the University of California, Irvine (UCI, https://sites.ps.uci.edu/kccams/). To maintain optimal performance, the system undergoes careful warm-up procedures before each set of measurements, ensuring temperature balance in the magnets and accelerator. Beam intensity is also carefully controlled, with the ¹²C- beam intensity maintained at 50–70 µA to minimize divergence and ensure accurate results.

Quality control is a cornerstone of GIGCAS AMS operations. Each batch of samples is accompanied by NIST SRM 4990 C (oxalic acid-II) primary standards, phthalic anhydride background monitors, and IAEA-C7 secondary standards, to ensure both accuracy and precision.

With these advancements, GIGCAS now processes over 5000 samples annually, demonstrating the system’s high throughput and robust performance. The system’s precision and reliability were further validated by its strong performance in the 2022 Glasgow International Radiocarbon Inter-comparison activity.

Trajectory Frequency calculation

The Hybrid Single-Particle Lagrangian Integrated Trajectory Model (HYSPLIT version 5.3.0) was used to analysis the geographical provenance of the air masses that potentially transported the atmospheric ¹⁴C to the observation sites⁵². We calculated the backward trajectories starting from the beginning of the bomb test and extending one to two months beyond the end of the test period. A new trajectory was generated every 6 hours, with each trajectory simulated for ten days. The meteorological input data were sourced from the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) Reanalysis Data Archive provided by the National Oceanic and Atmospheric Administration (NOAA) Air Resources Laboratory⁵³, accessed via ftp://arlftp.arlhq.noaa.gov/archives/reanalysis using FileZilla version 3.67.0 (https://filezilla-project.org/).

Ecological half-life

The ecological half-life (T_1/2,_e) was widely used to reveal the decline speed of ¹⁴CO₂ levels in air with the combined effect of atmosphere transport, CO_2ff emission and exchange between reservoirs in the environment. T_1/2,_e is suitable used for explain the falling speed when the ¹⁴C levels decreased exponentially, which indicating that the exchange between carbon reservoirs dominated the decline trend. The calculation method is based on Xu et al.¹³:

$${{{{\rm{C}}}}}_{t}={{{{\rm{C}}}}}_{0}\cdot {{{{\rm{e}}}}}^{(-\left({\lambda }_{{{{\rm{r}}}}}+{\lambda }_{{{{\rm{e}}}}}\right))}$$

(2)

where the sum of ${\lambda }_{{{{\rm{r}}}}}$ and ${\lambda }_{{{{\rm{e}}}}}$ is the effective decay constant of ¹⁴C, in which ${\lambda }_{{{{\rm{r}}}}}$ is the physical decay constant and ${\lambda }_{{{{\rm{e}}}}}$ is the ecological decay constant. Thus, ${\lambda }_{{{{\rm{e}}}}}$ is related to the ${T}_{1/2,{{{\rm{e}}}}}$.

$${T}_{1/2,{{{\rm{e}}}}}=\frac{{{\mathrm{ln}}}(2)}{{\lambda }_{{{{\rm{e}}}}}}$$

(3)

Fossil fuel CO₂ emission estimate

According to the cycle of CO₂ and ¹⁴C in the atmosphere, CO_2ff could be calculated as below⁵⁴:

$${{{{\rm{CO}}}}}_{2{{{\rm{ff}}}}}=\frac{{{{{\rm{CO}}}}2}_{{{{\rm{bg}}}}}({\Delta }_{{{{\rm{obs}}}}}-{\Delta }_{{{{\rm{bg}}}}})}{{\Delta }_{{{{\rm{ff}}}}}-{\Delta }_{{{{\rm{obs}}}}}}$$

(4)

The CO_2ff levels in Chinese cities or regional station based on different backgrounds were calculated using the equation above. ${\Delta }_{{{{\rm{obs}}}}}$ and ${\Delta }_{{{{\rm{bg}}}}}$ are the ${\Delta }^{14}C$ values at the observation and background stations, respectively. ${\Delta }_{{{{\rm{ff}}}}}$ is defined as −1000‰. ${{{{\rm{CO}}}}}_{2{{{\rm{bg}}}}}$ is the CO₂ concentrations at background station. The CO₂ data from Mauna Loa (MLO) which belong to Global Greenhouse Gas Reference Network were used in this study as the background CO₂ levels, which available from 1970 to 2020 (https://gml.noaa.gov/ccgg/trends/)⁵⁵.

Statistical analysis

To evaluate whether regional differences in atmospheric Δ¹⁴C were statistically significant, we conducted pairwise comparisons between observation sites using a one-tailed paired sample t-test, under the assumption that the first site in each pair exhibited higher Δ¹⁴C values. Prior to the t-tests, the Shapiro–Wilk test was applied to the distribution of paired differences to assess normality. If the normality assumption was not met (p < 0.05) or if the sample size was too small (n < 3), we substituted the t-test with a non-parametric Wilcoxon signed-rank test.

For each test, we calculated the mean difference, 95% one-sided confidence interval (CI), and Cohen’s d as a measure of effect size when the t-test was used. To account for multiple comparisons, we applied a Bonferroni correction to the raw p-values, adjusting the significance threshold accordingly. All statistical analyses were performed in Python version 3.6 using the scipy, pingouin, and statsmodels libraries. A corrected p < 0.05 was considered statistically significant.