Asian continental isoscapes of tree-ring δ¹⁸O: implications for paleoclimate and paleoenvironment

ABSTRACT

Continental-scale isoscapes of tree-ring oxygen isotopes (δ¹⁸O_TR) are crucial for understanding atmospheric circulation dynamics, interpreting climatic significance, and tracing wood provenance. However, continental-scale δ¹⁸O_TR isoscapes remain underdeveloped. We compiled 313 multi-year averaged δ¹⁸O_TR records across Asia and generated isoscapes using two machine learning approaches: XGBoost and Random Forest. Results reveal a ‘sandwich’ pattern: depleted values at high (>50°N) and low (<30°N) latitudes, enriched values at mid-latitudes (30°N–50°N). This pattern closely resembles the distribution of precipitation δ¹⁸O (δ¹⁸O_P). Correlation and commonality analyses confirm δ¹⁸O_P as the primary driver of δ¹⁸O_TR isoscape patterns across Asia. Continental-scale δ¹⁸O_TR–elevation relationships are generally insignificant, except in Indian Summer Monsoon regions showing significant negative correlations (r = −0.69, p < 0.05). These findings suggest that δ¹⁸O-based paleoaltimetry reconstructions work best in regions with dominant moisture sources such as Indian Summer Monsoon regions. This study provides Asia’s first continental-scale δ¹⁸O_TR isoscapes, establishing a foundation for atmospheric circulation and dendroprovenancing research.

By complying 313 multi-year averaged tree-ring oxygen isotope records across Asia and generated isoscapes using thin-plate spline regression interpolation, this study provides Asia's first continental-scale tree-ring dδ¹⁸O isoscapes, establishing a foundation for atmospheric circulation and dendroprovenancing research.

INTRODUCTION

The term isoscapes was introduced in 2005 to describe spatially explicit models of isotopic variation across geographic regions [1]. Among these, isoscapes of precipitation oxygen isotopes (δ¹⁸O_P) have become particularly valuable tools for understanding atmospheric circulation, hydrological processes, and both past and recent climate variability [1,2]. Significant progress has been made in constructing δ¹⁸O_P isoscapes on global [3] and regional scales [4]. Techniques such as spatial interpolation, dynamical modeling, and machine learning have produced increasingly detailed products [1,5]. However, the limited and sparse distribution of observational stations—particularly in remote and mountainous areas—continue to constrain the spatial resolution and completeness of δ¹⁸O_P isoscapes. This limitation underscores the importance of incorporating complementary proxy records, such as tree rings, to enhance isoscape reconstructions across broader spatial scales.

Tree-ring oxygen isotopes (δ¹⁸O_TR) have demonstrated their strong potential as proxies for extending δ¹⁸O_P isoscapes. δ¹⁸O_TR is influenced by both the δ¹⁸O_P and relative humidity (RH) [6,7]. Through mechanistic models and transfer functions, δ¹⁸O_TR has shown promising capacity for reconstructing δ¹⁸O_P [8]. Advances in pre-treatment techniques and mass spectrometry have broadened δ¹⁸O_TR applications across a wide range of timescales [9], from intra-annual resolution [8,10] to millennial-scale hydroclimate reconstructions [11,12]. Spatially, δ¹⁸O_TR studies now encompass regions from the low-latitude tropics to mid- and high-latitude environments [13,14]. These methodological and spatial advances therefore provide a key foundation for developing large-scale δ¹⁸O_TR isoscapes.

Regional δ¹⁸O_TR isoscapes have been used to investigate atmospheric moisture transport, climate variability, and wood provenance. For example, δ¹⁸O_TR isoscapes (a network of δ¹⁸O_TR) have identified the Indian Summer Monsoon (ISM) as the dominant moisture source for Southwest China [15]. Meanwhile, a network of 24 δ¹⁸O_TR chronologies across Asia has been used to decipher the combined influence of South Asia Summer monsoon precipitation and El Nino Southern Oscillation (ENSO) on regional δ¹⁸O_TR variability [16]. In North America, δ¹⁸O_TR isoscapes have revealed a decreasing influence of the Labrador Sea on forest moisture regimes [13]. Additionally, δ¹⁸O_TR isoscapes have proved to be helpful in wood provenance assessments, such as identifying timber origins in North America [17], Europe [18], tropical Africa [19] and tropical South America [19]. Despite these promising applications, continental-scale δ¹⁸O_TR isoscapes remain underdeveloped, largely due to the limited data availability.

Asia represents an ideal setting for advancing δ¹⁸O_TR isoscape research due to its diverse atmospheric circulation regimes, complex topography, and a relatively dense distribution of δ¹⁸O_TR records [9]. This study establishes the first δ¹⁸O_TR isoscape for the Asian continent, constructed from multi-year mean values compiled in previous work [9] and generated using two machine learning approaches—XGBoost and Random Forest—as described in earlier studies [2,5]. Our objectives for the large-scale δ¹⁸O_TR isoscape investigation were: (i) to characterize the spatial distribution of δ¹⁸O_TR across Asia; (ii) to identify environmental variables that mostly influence the isotopic patterns; and (iii) to quantify the relationship between δ¹⁸O_TR and elevation. To conduct this investigation, we propose two hypotheses. First, the spatial distribution of δ¹⁸O_TR is expected to reflect that of precipitation δ¹⁸O_P, capturing the isotopic signature of source water incorporated during tree growth. Second, we anticipate that the relationships between δ¹⁸O_TR and environmental variables—such as δ¹⁸O_P, RH, and elevation—will vary regionally, shaped by Asia’s climatic diversity and topographical complexity.

RESULTS AND DISCUSSION

Observed δ¹⁸O_TR isoscapes

The observed δ¹⁸O_TR values ranged from 15.20‰ to 37.10‰ (Fig. 1a and b). Observed May–September δ¹⁸O_P from GNIP and TNIP stations varied between −18.05‰ and 2.93‰ (Fig. 1c). Reconstructed May–September δ¹⁸O_P, derived from δ¹⁸O_TR and CRU RH data, ranged from −18.92‰ to 1.17‰ (Fig. 1d). Similarly, δ¹⁸O_P reconstructed using CHELSA RH varied from −8.84‰ to −0.60‰, with a mean of −9.07‰ and a median of −8.53‰ (Fig. S1). The mean difference between δ¹⁸O_TR and δ¹⁸O_P reflects several fractionation processes occurring after precipitation infiltrates the soil and during water transport and assimilation in trees [6]. While root uptake does not fractionate oxygen isotopes, leaf water becomes enriched through transpiration, and this signal is incorporated into cellulose during synthesis [6]. These processes collectively result in δ¹⁸O_TR values being higher than δ¹⁸O_P.

Figure 1.

Observed and modelled spatial patterns of δ¹⁸O_TR and δ¹⁸O_P across Asia. (a) Observed δ¹⁸O_TR (313 sites). (b) δ¹⁸O_TR isoscape based on XGBoost. (c) May–September δ¹⁸O_P (252 sites) from the GNIP and TNIP [20]. (d) Reconstructed May–September δ¹⁸O_P derived from δ¹⁸O_TR using CRU TS 4.07 (1993–2022) RH data [21]. Black dotted lines in (a), (c) and (d) delineate the boundary of the ISM domain.

The Asian δ¹⁸O_TR isoscapes exhibit a characteristic ‘sandwich’ pattern (Fig. 1a and b), characterized by more depleted values at high latitudes (>50°N) and low latitudes (<30°N), and relatively enriched values at mid-latitudes (30°N–50°N). This spatial configuration closely mirrors the observed May–September δ¹⁸O_P distributions from the GNIP and TNIP (Fig. 1c), thereby supporting our first hypothesis. The δ¹⁸O_P values reconstructed from observed δ¹⁸O_TR isoscapes and CRU RH data also replicate this ‘sandwich’ pattern (Fig. 1d). A similar structure was obtained using RH from the CHELSA dataset for δ¹⁸O_P, reinforcing the robustness of these results (Fig. S1). These spatial patterns suggest a distinct climate transition zone, where we identify a monsoon boundary at approximately 31°N–33°N on the central Tibetan Plateau, which is consistent with previously reported boundaries based on extensive δ¹⁸O_P observations (34°N–35°N) [20]. Notably, the southeastern Tibetan Plateau shows the lowest δ¹⁸O_TR values due to isotopically depleted rainfall from its exceptionally wet upstream regions (e.g., Cherrapunji) [9]. These findings demonstrate the potential of δ¹⁸O_TR isoscapes in capturing large-scale isotopic gradients and atmospheric circulation patterns including dynamics of the Indian and East Asian monsoon systems.

Simulated δ¹⁸O_TR isoscapes

We modeled δ¹⁸O_TR isoscapes using RH data derived from the CRU TS 4.07 dataset, along with simulated May–September δ¹⁸O_p values from multiple global isotope-enabled climate models (isoGSM, ECHAM5 and LMDZ4). As anticipated, the modeled δ¹⁸O_TR isoscapes also successfully reproduced the characteristic ‘sandwich’ pattern observed in empirical δ¹⁸O_TR measurements (Fig. 2c, e and g vs. Fig. 1b). These modeled δ¹⁸O_TR distributions also resembled the spatial patterns generated by δ¹⁸O_P simulations (Fig. 2b, d, f), which exhibit depleted values in high-latitude regions and low-latitude monsoon zones. These highlight the coherence between δ¹⁸O_P and δ¹⁸O_TR. Similar results were observed when using CHELSA RH and simulated May–September δ¹⁸O_P (Fig. S2).

Figure 2.

Spatial distribution of RH and simulated δ¹⁸O in precipitation and tree rings across Asia. (a) Mean May–September RH derived from the CRU dataset. (b), (d) and (f) Simulated May–September δ¹⁸O_P for the same period from three global isotope-enabled models: IsoGSM (b), ECHAM5 (d) and LMDZ4 (f). (c), (e) and (g) Corresponding simulations of δ¹⁸O_TR derived from each model, respectively.

Spatial drivers of δ¹⁸O_TR: δ¹⁸O_P vs. RH

Multiple analytical approaches, including correlation analysis and commonality analysis, further demonstrate the predominant influence of δ¹⁸O_P in determining δ¹⁸O_TR isoscapes. The scatter plot reveals a negative relationship between RH and δ¹⁸O_TR, with higher RH values corresponding to depleted δ¹⁸O_TR values (Fig. 3a). In contrast, we observe significant positive relationships between δ¹⁸O_P and δ¹⁸O_TR across all isotope-enabled models, indicating that enriched δ¹⁸O_P values correspond to enriched δ¹⁸O_TR values (Fig. 3b–d). Quantitatively, correlations between observed δ¹⁸O_TR and simulated δ¹⁸O_P (r = 0.55–0.64, p < 0.01, n = 313) are consistently stronger than those with RH (CRU: r = −0.48, p < 0.05; CHELSA: r = −0.47, p < 0.05; n = 313, Fig. 3a and Fig. S3). A commonality analysis (model: δ¹⁸O_TR = CRU RH + δ¹⁸O_P) reveals that the combined effects of δ¹⁸O_P and RH explain 54.80%–62.73% of observed δ¹⁸O_TR variance (Table 1). Notably, the individual effect of δ¹⁸O_P (33.04%–39.81%) substantially exceeds that of RH (15.53%–25.86%), further confirming the dominant role of δ¹⁸O_P in determining δ¹⁸O_TR patterns across Asia. Similar results were obtained when using CHELSA RH data in the commonality analysis model (Table S1).

Figure 3.

Relationships between observed δ¹⁸O_TR and RH/δ¹⁸O_P across 313 sampling sites in Asia. (a) Scatter plot illustrating the correlation between observed δ¹⁸O_TR values and mean May–September RH based on the CRU dataset. (b–d) Relationships between δ¹⁸O_TR and simulated δ¹⁸O_P from three global isotope-enabled models: IsoGSM (b), ECHAM5 (c) and LMDZ4 (d). Pearson correlation coefficients are shown at the top.

Table 1.

Commonality analysis results showing the unique and shared contributions of CRU-RH and simulated δ¹⁸O_P to the spatial variability of δ¹⁸O_TR across Asia.

Model names of δ18Op	Individual effect of RH on δ18OTR (%)	Individual effect of δ18Op on δ18OTR (%)	Joint effect on δ18OTR (%)	Total effect on δ18OTR (%)
IsoGSM	15.53	33.04	7.39	55.96
ECHAM5	25.86	39.81	−2.94	62.73
LMDZ4	24.57	31.88	−1.65	54.80

The dominant role of δ¹⁸O_P on δ¹⁸O_TR has also been observed in other regions such as eastern Canada [13], Eurasia [22], the Amazon [23] and the Europe Alps [24], although we acknowledge that the specific climatic regimes differ from those in Asia. The δ¹⁸O_P signal is first incorporated into plant water through soil moisture, and is subsequently transferred into photosynthetic products before ultimately being recorded in cellulose during synthesis [6].

δ¹⁸O_TR–elevation relationships

Our second hypothesis, concerning the relationship between δ¹⁸O_TR and elevation, is also supported. The relationship between δ¹⁸O_TR and elevation demonstrates remarkable heterogeneity across the Asian continent, with distinctive patterns emerging at regional scales. No significant correlation between δ¹⁸O_TR and elevation is observed at continental scale (Fig. 4a, r = 0.04, p > 0.05). This lack of a continental-scale trend likely reflects the continent’s highly complex topography, the diversity of climate regimes, and the influence of multiple moisture sources—including local evapotranspiration, as well as long-distance transport from the Indian, Pacific and Atlantic Oceans [20,25].

Figure 4.

Relationship between δ¹⁸O_TR and elevation across Asia and regional monsoon domains. (a) Relationship between δ¹⁸O_TR and elevation for all 313 sites. (b and c) Same input data as (a) but for regional patterns within the ISM and EASM zones, respectively. For each panel, the left sub-figure displays the spatial distribution of sampling sites, and the right sub-figure shows the corresponding scatter plots of δ¹⁸O_TR vs. elevation. Pearson correlation coefficients are shown in the top right corner of the scatter plot. The definitions of each subregion are presented in Fig. S4.

In contrast, a significant negative δ¹⁸O_TR–elevation relationship (Fig. 4b, r = –0.69, p < 0.05) is found in the ISM region, with a lapse rate of −0.15‰/100 m. This observation aligns with previous studies dealing with river water isotope studies on the southern Tibetan Plateau [26,27], precipitation isotope analyses in monsoon-dominated Tibetan Plateau regions [20], and leaf wax n-alkane δD measurements in Eastern China [28]. The observed lapse rate corresponds closely to the −0.13‰/100 m precipitation lapse rate previously reported for the monsoon Tibetan Plateau [20]. This phenomenon can be attributed to adiabatic cooling of rising water vapor, during which heavy isotopes preferentially precipitate due to their stronger molecular bonds [29].

Outside the ISM region, other Asian regions exhibit markedly different δ¹⁸O_TR–elevation patterns (Fig. S5). The EASM region shows a weak non-significant negative correlation (Fig. 4c, r = −0.10, p > 0.05, n = 102, Fig. 4c), while the arid westerlies region demonstrates a significant positive relationship (Fig. S5a, r = 0.30, p < 0.05, n = 52). High-latitude areas also display a weak non-significant negative relationship (Fig. S5b, r = −0.13, p > 0.05, n = 29). The unexpected positive δ¹⁸O_TR–elevation relationship in arid westerlies region likely arises from the interaction of multiple moisture sources, enhanced contributions from recycled surface water, and sub-cloud evaporation processes [20,26,27]. These factors could overprint the original isotopic signal of source moisture and may obscure the expected negative δ¹⁸O-elevation gradient. Such positive isotope–elevation linkages are further supported by satellite-based observations of water vapor δD [30], river water δ¹⁸O analyses [26,27], and leaf wax n-alkane δD measurements [28] in Asia.

These region-specific δ¹⁸O_TR–elevation linkages have significant implications for paleoelevation reconstructions. Specifically, isotope-based reconstructions of paleoelevation are most reliable in regions dominated by a single, consistent moisture source with substantial long-distance transport, such as the ISM system. Conversely, in areas influenced by multiple or variable moisture sources, isotope–elevation relationships may be biased and should therefore be interpreted with considerable caution.

Regional differences for drivers of spatial δ¹⁸O_TR pattern

The drivers of observed δ¹⁸O_TR across Asia are region-specific, confirming our second hypothesis (Fig. 5). In the ISM region, δ¹⁸O_P emerges as the primary driver of δ¹⁸O_TR (r = 0.64–0.69, p < 0.05), followed by RH (CRU: r = 0.36, p < 0.05; CHELSA: r = −0.18, p > 0.05, Fig. 5). Commonality analysis further supports this pattern, revealing that δ¹⁸O_P consistently explains a larger proportion of δ¹⁸O_TR variance than RH (Fig. S6). Such dominant influence of δ¹⁸O_P on δ¹⁸O_TR has been supported by intra-annual isotope observations and process-based isotopic modelling on the southeastern Tibetan Plateau [10]. In the EASM region, RH demonstrates a stronger statistical relationship with δ¹⁸O_TR (CRU: r = −0.49, p < 0.05; CHELSA: r = −0.54, P > 0.05) compared to δ¹⁸O_P (ECHAM5: r = 0.2, p < 0.05; LMDZ4: r = 0.09, p > 0.05; isoGSM: r = 0.39, p < 0.05, Fig. 5), which is also supported by the results of commonality analysis (Fig. S7). Such finding is corroborated by δ¹⁸O_TR observational data and validated through process-based isotopic modeling approaches [31]. These may explain why δ¹⁸O_TR is often employed to reconstruct RH in the EASM region [32,33]. RH affects δ¹⁸O_TR mainly in two ways: (i) when RH is low, plants lose more water through their leaves, which makes the leaf water richer in ¹⁸O. This enriched water is used to make tree rings; and (ii) low RH also causes more evaporation from the soil, making the water taken up by roots richer in ¹⁸O. Both effects lead to higher δ¹⁸O_TR [6,32].

Figure 5.

Pearson correlation coefficients between observed δ¹⁸O_TR and key drivers (RH from CRU and CHELSA, and simulated May–September δ¹⁸O_P from three isotope-enabled models) across four Asian subregions. Non-significant correlations (p > 0.05) are marked with ×; color intensity indicates correlation strength.

In the arid westerlies region, it seems challenging to disentangle the relative contributions of RH and δ¹⁸O_P to δ¹⁸O_TR due to their similar correlation coefficients (CRU-RH: r = −0.69, p < 0.05; CHELSA-RH: r = −0.48, p > 0.05; δ¹⁸O_P: r = 0.45–0.55, p < 0.05, Fig. 5). Commonality analysis using CHELSA RH and δ¹⁸O_P from ECHAM5 and LMDZ4 revealed a stronger influence of δ¹⁸O_P on δ¹⁸O_TR, whereas analyses using CRU RH and δ¹⁸O_P from IsoGSM, ECHAM5, and LMDZ4 indicated a greater contribution from RH (Fig. S8). Additionally, commonality analysis combining CHELSA RH and δ¹⁸O_P from IsoGSM also showed RH as the dominant driver (Fig. S8). Such results may be attributed to the high contribution of recycled moisture to local precipitation, which enhances the coupling between RH and δ¹⁸O_P. Moreover, in the high latitude region, RH exhibits a slightly stronger overall influence on δ¹⁸O_TR (CRU: r = −0.40, p < 0.05; CHELSA: r = −0.43, p > 0.05) compared to δ¹⁸O_P (ECHAM5: r = 0.33, p > 0.05; LMDZ4: r = 0.35, p > 0.05; IsoGSM: r = 0.40, p < 0.05, Fig. 5). Commonality analysis only using CRU RH and δ¹⁸O_P from IsoGSM revealed a slightly greater contribution from δ¹⁸O_P to δ¹⁸O_TR. In contrast, analyses using other combinations consistently indicated a higher contribution from RH than δ¹⁸O_P (Fig. S9). Such linkages may result from the influence of permafrost-affected source water, which can overwrite or obscure the climatic signals imprinted in δ¹⁸O_TR [34]. This may complicate the interpretation of δ¹⁸O_TR in high-latitude permafrost regions, where no single dominant climatic driver is evident (r < 0.6) [35,36].

Applications

The development of the first Asian δ¹⁸O_TR isoscapes presented in this study opens up a wide range of research opportunities across paleoclimate and (paleo)environmental sciences. In paleoclimate research, these maps offer a valuable framework for reconstructing hydroclimatic variability, tracking monsoon dynamics. In ecological studies, δ¹⁸O_TR isoscapes can support investigations into wood provenance and migration. For hydrology, combining δ¹⁸O_TR with climate data enables improved tracking of water cycle changes, identification of moisture sources. Beyond these fields, potential applications include forensic timber tracing and archaeological artifact sourcing.

SUMMARY AND OUTLOOK

In summary, to the best of our knowledge, this study represents the first attempt to generate δ¹⁸O_TR isoscapes at a continental scale for Asia, offering a foundational framework for exploring large-scale hydroclimatic processes and enhancing dendroprovenancing applications. The resulting isoscapes reveal a pronounced ‘sandwich’ pattern, characterized by depleted δ¹⁸O_TR values at high (>50°N) and low (<30°N) latitudes, and enriched values across the mid-latitudes (30°N–50°N). This spatial distribution closely aligns with May–September δ¹⁸O_P observed in GNIP and TNIP datasets. Our reconstructed δ¹⁸O_TR field, informed by δ¹⁸O_P outputs from global isotope-enabled models and RH, effectively reproduces this latitudinal gradient. We identify δ¹⁸O_P as the primary driver of δ¹⁸O_TR variability across the continent based on correlation and commonality analysis. Notably, the relative importance of these drivers varies regionally: δ¹⁸O_TR is primarily driven by δ¹⁸O_P in the ISM, by RH in the EASM and high-latitude regions, while in the arid westerlies the relative contributions of RH and δ¹⁸O_P are difficult to disentangle. While the δ¹⁸O_TR–elevation relationship is negligible at the continental scale, substantial heterogeneity emerges at regional levels. A significant negative correlation is observed in the ISM domain, contrasting with weak or even positive correlations in the EASM, arid westerlies, and high latitude regions. These findings suggest that isotope-based paleoaltimetry may only be viable in regions dominated by consistent moisture sources and long-range atmospheric transport, such as the ISM. Collectively, our results underscore the utility of δ¹⁸O_TR isoscapes as proxies for reconstructing δ¹⁸O_P distributions, diagnosing shifts in atmospheric circulation (e.g. ISM variability), and informing dendroprovenancing analyses.

While this study offers valuable insights into the variability of δ¹⁸O_TR isoscapes, some limitations of this study should be acknowledged. First, we assumed that May–September δ¹⁸O_P could serve as a direct proxy for source water in modeling δ¹⁸O_TR, without explicitly accounting for soil-related processes such as soil depth, composition evaporation, root uptake depth, and residence time of soil water. While this assumption may hold in humid regions, it is likely less valid in arid environments, where trees may rely on moisture accumulated prior to the growing season or derived from snowmelt [37]. Second, when reconstructing δ¹⁸O_P using δ¹⁸O_TR and RH (e.g., Fig. 1d), and generating modeled δ¹⁸O_TR isoscapes from RH and δ¹⁸O_P (e.g., Fig. 2), we treated several key parameters as constants (e.g., f₀ = 0.42, ε* = 9‰, ε_k = 29‰). However, these values are known to vary among species and even within individuals, which may introduce uncertainty into the modeled δ¹⁸O_TR. Third, while our δ¹⁸O_TR isoscapes capture large-scale a “sandwich” pattern (Fig. 1b), caution is warranted when interpreting fine-scale features, particularly in heterogeneous landscapes. Fourth, our analysis is based on multi-year averaged δ¹⁸O_TR values, which capture the climatological patterns of δ¹⁸O_TR isoscapes but do not reflect interannual variability.

Consequently, building on the present findings, we suggest the following research directions need to be taken into account in future studies to strengthen the future applicability of isoscapes in the fields of ecology, hydrology, climate, geology, forensics: (i) global δ¹⁸O_TR isoscapes development—expand current efforts to a global scale by integrating more extensive datasets, improving spatial resolution and model fidelity; (ii) proxy-model integration and comparisons—combine δ¹⁸O_TR with δ¹⁸O_P isoscapes, model simulations [38–41], and other hydrological and ecological proxies (e.g. river water, tap water, leaf water) [4,42] to enhance multi-proxy reconstructions of environmental processes; (iii) temporal δ¹⁸O_TR isoscapes generation for the past 50–60 years—generate annually resolved δ¹⁸O_TR isoscapes to capture interannual variability over the past 50–60 years [2,5]; (iv) application of the δ¹⁸O_TR isoscapes for dendroprovenancing—utilize δ¹⁸O_TR isoscapes to enhance dendroprovenancing applications, building on the pioneering work in the USA [17] and Italy [18].

MATERIALS AND METHODS

Collection and preprocessing of δ¹⁸O_TR and δ¹⁸O_P data

We compiled 313 δ¹⁸O_TR records across Asia through systematic literature review [9], expanding the dataset through to April 2025. To standardize species differences, broadleaf δ¹⁸O_TR values were adjusted by +1.9‰ based on established fractionation differences between broadleaf and coniferous species [43,44]. δ¹⁸O_P data were obtained from 252 monitoring stations (233 from GNIP, 19 from TNIP [20]) collected in April 2023. May–September mean δ¹⁸O_P values were calculated for each site to align with the active vegetation period when trees record isotopic signals. We also reconstructed May–September δ¹⁸O_P using δ¹⁸O_TR and RH data through established transfer functions [7,8]. Due to the variability in both temporal resolution and record length across datasets, we chose not to report a single representative value in the abstract. Detailed information on the lengths of δ¹⁸O_TR, GNIP, and TNIP records, as well as the equations used, is provided in the Supplementary Material (Figs S10–S14).

δ¹⁸O_TR isoscape generation

Building on studies that apply machine learning to isoscape generation [2,5], we also employed XGBoost and Random Forest to produce 0.5°×0.5° δ¹⁸O_TR isoscapes using our compiled δ¹⁸O_TR data, along with a suite of geographic, climatic, and categorical variables. These include longitude, latitude, elevation, climate data and climate classifications. Isoscapes were extended to non-forested regions for comparison with δ¹⁸O_P distributions. Details are provided in the supporting materials (Figs S15–S18, Tables S2–S4).

Modelling δ¹⁸O_TR isoscapes

For comparison, we modeled δ¹⁸O_TR spatial patterns using δ¹⁸O_P simulations from three validated isotope-enabled general circulation models (isoGSM [41], ECHAM5 [40] and LMDZ4 [39]) and RH data from CRU TS 4.07 [21] and CHELSA [45] datasets. RH was calculated using following equations [46]:

(1)

(2)

where T_mean and VAP represent monthly mean temperature and vapor pressure, respectively.

In calculating δ¹⁸O_TR from the simulated δ¹⁸O_P and RH in Figure 2, we interpolated the spatial resolution of the simulated δ¹⁸O_P to match the resolution of RH. The equation used to model δ¹⁸O_TR is provided in the Supplementary data, with parameter settings identical to those used for reconstructing δ¹⁸O_P in Fig. 1d.

δ¹⁸O_TR–elevation relationships

Elevation–δ¹⁸O_TR relationships were examined using elevation data from original publications or extracted from the ETOPO (Earth Topography) Global Relief model. Analysis was conducted at the continental scale and within four subregions: Indian Summer Monsoon, East Asian Monsoon, arid westerlies and high latitudes. Details about each of the defined subregions are provided in Supplementary data. Moreover, we examined the relationships between δ¹⁸O_TR and geographic position (longitude and latitude). Our results show that spatial position—particularly latitude—exerts a measurable influence on δ¹⁸O_TR across Asia (Fig. S19), consistent with the “sandwich” pattern observed in the isoscape. As the illustrative figure provides limited additional information and largely overlaps with the main figures (e.g., Fig. 1b), it is presented in the supplementary materials rather than in the main text to maintain brevity.

Driving factors on δ¹⁸O_TR isoscapes in Asia and subregions

To elucidate the influence of δ¹⁸O_P and RH on δ¹⁸O_TR isoscapes, we extracted mean May–September δ¹⁸O_P and RH values for 313 tree-ring δ¹⁸O_TR sampling sites. Scatter plots were constructed to explore bivariate relationships between δ¹⁸O_TR and both δ¹⁸O_P and RH. Additionally, to quantitatively assess the individual and shared contributions of these environmental variables to spatial variation in δ¹⁸O_TR, we employed commonality analysis [47]. Using a similar scatter plot, we investigated the relationships between δ¹⁸O_TR and both δ¹⁸O_P and RH within each subregion.

All analyses were performed using R version 4.4.3 [48].

DATA AVAILABILITY

The tree-ring oxygen isoscapes generated in this study (Fig. 1b) are available on Zenodo: https://zenodo.org/records/17503284. Model outputs from LMDZ4, ECHAM5 and isoGSM were obtained from the following repositories: (i) SWING2 Project Database: https://data.giss.nasa.gov/swing2/; (ii) Zenodo Repository: https://zenodo.org/records/1249604; and (iii) University of Tokyo Database: https://isotope.iis.u-tokyo.ac.jp/∼kei/tmp/isogsm2/flxmon.iso-n.nc. The CRU and CHELSA climate datasets are available at: (i) CRU: https://crudata.uea.ac.uk/cru/data/hrg/cru_ts_4.07/; (ii) CHELSA: https://chelsa-climate.org/downloads/. Digital Elevation Model (DEM) data (ETOPO_2022_v1_60s_N90W180) are available from NOAA: https://www.ncei.noaa.gov/products/etopo-global-relief-model. The GNIP dataset is accessible at https://www.iaea.org/services/networks/gnip. The TNIP dataset is provided by National Tibetan Plateau Data Center http://data.tpdc.ac.cn/en.

FUNDING

This work was supported by the National Key Research and Development Program of China (2024YFF0807902), the National Natural Science Foundation of China (42425706 and 42488201), the Strategy Priority Research Program (Category B) of Chinese Academy of Sciences (XDB0710000), the International Partnership program of the Chinese Academy of Sciences (170GJHZ2023074GC) and the Second Tibetan Plateau Scientific Expedition and Research (STEP) program (2024QZKK0301).

AUTHOR CONTRIBUTIONS

Chenxi Xu: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, resources, supervision, writing–original draft, writing– review & editing. Ru Huang: data curation, formal analysis, investigation, methodology, software, validation, visualization, writing—original draft, writing—review & editing. Wenling An: data curation. Qingyu Zhao: data curation. Yaru Zhao: data curation. Yucheng Liu: data curation. Jussi Grießinger: writing– review & editing. Wolfgang Jens-Henrik Meier: writing– review & editing. Zhengtang Guo: conceptualization, project administration, writing–review & editing.

Conflict of interest statement. None declared.