Quantitative climate reconstructions are essential for validating climate model performance when simulating past climate changes and thus indicating their potential to predict future climate conditions (Schmidt et al., 2014). Fossil assemblages derived from marine, lacustrine, or terrestrial archives are among the most widely used proxies providing direct quantitative information on climate (Birks et al., 2010). These approaches build upon the assumption that compositional changes within biotic communities are predominantly driven by changing climatic conditions (Birks et al., 2010). However, all proxy-based reconstructions inherently contain uncertainties, due to ecological complexities, a lack of understanding of taphonomies, and methodological limitations (Juggins, 2013). Although multi-proxy reconstructions are frequently employed for climate model evaluation, discrepancies between proxy data and model simulations – exemplified by the “Holocene conundrum” – are often attributed to proxy biases rather than modeling errors (Liu et al., 2014). This underscores the pressing need to develop and refine new and robust climate proxies.

Among terrestrial proxies, pollen analysis remains the most commonly used method for reconstructing past climate conditions on land (Herzschuh et al., 2023a), largely due to its widespread preservation across various terrestrial archives such as lake sediments, peat bogs, and soil deposits (Birks and Seppä, 2004). The popularity of pollen-based reconstructions stems from the strong and extensively studied quantitative relationships linking climate variables (e.g., temperature and precipitation) with vegetation composition and pollen assemblages (e.g. Seppä et al., 2004; Herzschuh et al., 2010). Despite these strengths, pollen proxies are subject to several significant limitations (Chevalier et al., 2020). A critical issue is their limited taxonomic resolution, typically constrained to genus or family levels, which can obscure precise ecological interpretations, particularly when closely related taxa differ in climate preferences (Tian et al., 2017). Moreover, pollen productivity is inherently biased toward wind-pollinated taxa, thereby skewing reconstructions toward certain climate conditions. For example, the contribution from productive boreal forest taxa to tundra pollen spectra leads to a warm bias (Klemm et al., 2013). Additionally, pollen signals commonly integrate information over larger spatial scales – ranging from local to regional – depending heavily on topography, basin size, landscape openness, vegetation composition, and transport mechanisms (Prentice, 1985; Sugita, 2007). Plant macrofossils offer valuable complementary data, as they provide higher spatial precision, typically representing local-scale vegetation dynamics and facilitate species-level identification, thereby overcoming the coarse taxonomic resolution inherent in pollen analysis (Birks, 2001). However, plant macrofossil records often have limited availability and preservation potential, resulting in fragmented or discontinuous fossil sequences that constrain the reliability and continuity of reconstructions. These limitations emphasize the necessity of developing additional vegetation proxies capable of providing continuous, local, and quantitatively robust climate reconstructions.

Sedimentary ancient DNA (sedaDNA) has recently emerged as a novel biological proxy that effectively traces past changes in ecological communities, offering significant potential for paleoclimatic reconstructions (Parducci et al., 2017; Capo et al., 2021). Like plant macrofossils, sedaDNA primarily originates from the vegetative parts of local and nearby plants (Niemeyer et al., 2017; Alsos et al., 2018). Yet, similar to pollen analysis, recorded DNA reads are abundant and can yield quantitative data about diverse plant taxa simultaneously. Plant metabarcoding using chloroplast DNA markers, such as the commonly employed trnL-gh fragment, enables reconstruction of past vegetation assemblages at high taxonomic resolution, mostly reaching species or genus level (Taberlet et al., 2007). Numerous sedaDNA datasets have already been published from various regions worldwide, especially from the extratropical Northern Hemisphere, demonstrating the method's broad-scale applicability (Liu et al., 2021; Alsos et al., 2022; Courtin et al, 2025). Moreover, plant metabarcoding studies of modern lake surface-sediments using this marker have provided promising reference datasets (Alsos et al., 2016; Niemeyer et al., 2017; Stoof-Leichsenring et al., 2020; Li et al., 2021), successfully establishing links between contemporary vegetation assemblages and subfossil sedimentary DNA profiles, thereby facilitating accurate reconstructions of past vegetation types from sedaDNA records (Liu et al., 2021). Also, first approaches to leverage information about median climatic conditions of selected taxa as climate indicators yielded promising results (Mayfield et al., 2024). Nevertheless, despite these encouraging developments, the reliability and robustness of plant sedaDNA as a climate proxy have not yet been systematically explored or validated.

Essentially, two main approaches exist for calibrating biological assemblage proxies. The first approach directly links modern sub-fossil assemblages with contemporary climate parameters. This is done either by directly matching modern assemblages to fossil spectra using the Modern Analogue Technique (MAT; Overpeck et al., 1985), or by applying modern taxa-climate relationships to fossil spectra using statistical methods such as Weighted-Averaging Partial Least Squares regression (WA-PLS; ter Braak and Juggins, 1993). The second approach leverages ecologically meaningful relationships by relating the occurrence of individual taxa to modern climate conditions using, for example, probability density functions (PDFs; Kühl et al., 2002), which represent the probability of occurrence across climate gradients. The individual PDFs of recorded taxa in fossil assemblages are then combined to form joint PDFs, enabling quantitative reconstruction of past climates. This PDF-based method, initially developed for plant macrofossils due to their high taxonomic resolution, has recently been advanced through the Climate REconstruction SofTware (CREST) framework allowing, among others, for weighting by abundances (Chevalier et al., 2014). CREST also provides detailed diagnostic tools to assess the climate niches of individual taxa and quantifies their contributions to climate reconstructions at the sample level. A significant advantage of calibrating proxies using modern plant occurrences is that it does not depend on the availability of an extensive modern surface-sample set for training. If, however, modern sub-fossil assemblages are available, they can be employed to independently validate this method, thereby enhancing confidence in the resulting proxy-based climate reconstructions. Hence, despite the availability of suitable statistical frameworks, their assessment and application to sedaDNA data have not yet been tested.

Although large global databases of taxonomic occurrences, such as the Global Biodiversity Information Facility (GBIF; GBIF.org, 2025), provide extensive data on species distributions, establishing accurate regional PDFs requires comprehensive modern occurrence data specifically from the target study regions (Meyer et al., 2016). This presents a substantial challenge, especially in remote and data-scarce regions such as the Arctic and Siberia. To address gaps in observational data, species distribution models (SDMs) have increasingly been employed to predict species occurrences based on their environmental niche preferences and thus allow for continuous information about taxa occurrences along the variable gradient (Elith and Leathwick, 2009; Thuiller et al., 2009). Recent methodological advances have enabled the efficient implementation of SDMs for hundreds of species simultaneously and such allowing for large-scale biodiversity approaches (Hu et al., 2025). However, despite their extensive use in contemporary ecological and biogeographic studies, SDMs have not yet been systematically integrated into quantitative paleoclimate reconstructions.

Here, we present a framework for leveraging plant metabarcoding of sedaDNA as a proxy for quantitative reconstruction of summer temperatures. The framework uses PDFs derived from modern relationships between 180 plant taxa occurrences (representing metabarcoding sequence types) and summer temperature, established using newly developed SDMs. We validated this approach using plant metabarcoding data from 203 lake surface-sediment samples across Siberia, and subsequently applied it to reconstruct summer temperature changes at a test site in eastern Siberia within the region. We also evaluated this method against methods employing modern surface sediments for calibration. We validate the calibration method and reconstruction using pollen-based calibration and reconstruction.

We used plant assemblages derived from sedimentary ancient DNA (sedaDNA) metabarcoding for quantitative climate reconstructions, applying four different calibration techniques (Fig. 1; Table 1). First, modern plant occurrence data from the Global Biodiversity Information Facility (GBIF; GBIF.org, 2025) were integrated with contemporary climate data using species distribution modeling (SDM). Model output was used to fit probability density functions (PDFs) used for reconstruction in an indicator-species approach (plantDNA_PDF/SDM). The second approach is similar to the first one, only that GBIF-derived modern taxa occurrences were directly linked to climate data. Third, we directly linked modern sub-fossil sedaDNA assemblages with contemporary climate parameters via Weighted-Averaging Partial Least Squares regression (WA-PLS) and applied these relationships to assemblages from the past (plantDNA_WAPLS). Finally, modern sub-fossil sedaDNA assemblages were matched to fossil samples via distance measures (plantDNA_MAT). The sedaDNA-based climate reconstructions were further validated through comparisons with pollen-based reconstructions.

We applied four statistical approaches – plantDNA_PDF/SDM, plantDNA_PDF/GBIF, plantDNA_WAPLS and plantDNA_MAT – to reconstruct summer temperatures using plant DNA metabarcoding spectra obtained from 203 lake surface-sediment samples collected across Siberia (79–178° E, 55–74° N). These reconstructed temperatures were compared with observed temperatures to evaluate reconstruction biases.

The plantDNA_PDF/SDM method, calibrated using modern plant occurrence data supplemented by species distribution modeling, yields a median absolute reconstruction bias of 1.1 °C across all surface samples (Fig. 2), representing 9.5 % of the total temperature range (5.0–16.6 °C) covered by the dataset (Table 1). Reconstruction accuracy is high at 47.3% of the sites, exhibiting biases below 1 °C. Observed and reconstructed temperatures are strongly correlated (r= 0.86, p<0.01). The method generally overestimates temperatures below 8 °C and underestimates temperatures above 14 °C (Fig. 2). The sensitivity analysis identifies several taxa – Larix, Persicaria amphibia and Menyanthes trifoliata and, to a lesser extent, several Ericaceae taxa – as contributing strongly to warmer reconstructed temperatures confirmed by their sensitivity ranges (Figs. A4 and A5). Betula, Anthemideae_01, and Asteraceae_01 have warm effects on the more northern samples, but cold effects on the more southern samples (Fig. A4). Taxa contributing to colder reconstructions include Saliceae, Dryas, and Hulteniella integrifolia. Additionally, the plantDNA_PDF/SDM method accurately reconstructed modern temperatures around Lake Billyakh, as indicated by a small reconstruction bias of only 0.4 °C (Table 1).

Compared with plantDNA_PDF/SDM, the plantDNA_PDF/GBIF method, which relies solely on modern plant occurrences for calibration, shows a substantially higher median bias of 2.1 °C (Fig. 2), corresponding to 18.1 % of the total temperature range (Table 1). This larger bias primarily results from pronounced reconstruction inaccuracies at the lower end of the temperature gradient. In contrast, methods calibrated using modern sediment samples – plantDNA_WAPLS and plantDNA_MAT – yield relatively low median biases (0.7 and 0.5 °C, respectively; Fig. 2), as determined by the sensitivity analysis. However, these methods exhibit significantly lower accuracy when reconstructing the modern temperature around Lake Billyakh, as indicated by notably higher biases of 3.1 °C for plantDNA_WAPLS and 3.8 °C for plantDNA_MAT (Table 1).

We applied the four approaches to reconstruct summer temperatures using 73 plant metabarcoding spectra from the Lake Billyakh sediment core covering the last 32 000 years (Figs. 3, 4 and 5). With respect to plantDNA_PDF/SDM, reconstruction uncertainties, inferred from combined PDFs, range from 1.3 to 1.9 °C for the 50 % uncertainty range (Fig. 5) and from 3.7 to 5.4 °C for the 95 % uncertainty range (not shown). During marine isotope stage (MIS) 3 and 2, summer temperatures were approximately 3.4 ± 1.5 °C cooler than present, with a distinct short-term warming peak occurring around 28 ka. At the Last Glacial Maximum (LGM), summer temperatures averaged 10.3 ± 1.6 °C. A sensitivity analysis (Fig. 3) reveals the taxa that have cold and warm effects on the temperature reconstruction in the single samples. Taxa such as Bistorta vivipara, Dryas, Eritrichium, Lagotis glauca, Papaver, and Saxifraga oppositifolia are mainly found in MIS 3 and LGM samples and have cold effects on the reconstruction. The diagnostic plots of the “combinedPDFs” show that most taxa strongly overlap in the cold range for the samples in the glacial (Fig. 4). However, some highly abundant taxa also occur today under warm conditions (e.g. Crepidinae_01 and Asteraceae_02). Several taxa that occur under intermediate warm conditions, have warm effects during the cold period of the glacial but cold effects during the warm period of the Holocene including Saliceae, Vaccinium_01, and Ranunculus_01 (Fig. 3). The temperature increased during the Late Glacial. The temperature rise accelerated during the Bølling-Allerød interstadial and reached modern-day temperatures around 10.5 ka. Peak warmth occurred in the early mid-Holocene at 5.8 ka, when temperatures were 1.6 ± 1.3 °C warmer than present. Following the Holocene Thermal Maximum (HTM), temperatures show a steady decline in the reconstruction (Fig. 4). The total temperature difference between the LGM and HTM at Lake Billyakh is 5.5 °C. The diagnostic plots of the “combinedPDFs” for an early Holocene sample (Fig. 4) strongly overlap in the warm range. Compared with the diagnostic plot of the glacial sample, less outlier taxa are observed. Alnus, Persicaria_01, and aquatic taxa like Nymphaeaceae and Potamogeton perfoliatus became abundant during the Holocene and have warm effects (Fig. 3).

Reconstructed temperature change of the other three methods shows glacial to Holocene trends (Fig. 4) similar to plantDNA_PDF/GBIF. Reconstructions of plantDNA_WAPLS and plantDNA_MAT have an overall larger reconstruction range and show more pronounced late Holocene cooling while the range and cooling intensity is lower for plantDNA_PDF/GBIF compared with plantDNA_PDF/SDM. plantDNA_WAPLS is the only method that does not show relatively warm MIS 3 samples.

We performed pollen-based reconstruction for the same sediment core as with sedaDNA-based reconstruction. Median biases and root-mean squared error of prediction (RMSEP) when comparing to similar calibration methods using sedaDNA are generally higher (Table 1). However, the variations in summer temperatures derived from sedaDNA- and pollen-based reconstruction are in good agreement, although reconstructed temperatures from sedaDNA are overall slightly warmer than from pollen (Fig. 4). The reconstructed temperature range is similar between sedaDNA and pollen_WAPLS and larger with pollen_MAT. The temperature increase during the Bølling-Allerød interstadial is more pronounced in the reconstruction with pollen_MAT compared to the sedaDNA-based reconstruction, while the temperature increase in the reconstruction with pollen_WAPLS is overprinted by strong variations. The HTM in the pollen-based reconstructions occurs slightly earlier but also during the early mid-Holocene like the sedaDNA-based reconstruction.

We demonstrated the effectiveness of sedimentary ancient DNA (sedaDNA) as a novel biological proxy for quantitative reconstruction of summer temperatures from lake sediments. This approach was successfully applied to plant DNA metabarcoding data from surface sediments and a sediment core from eastern Siberia. Calibration was performed either by linking modern plant occurrences from the GBIF database to summer temperatures using probability density functions, or by relating modern DNA assemblages directly to climate via weighted averaging partial least squares (WA-PLS) regression or modern analogue matching techniques. The sedaDNA-based reconstructions consistently provided accurate and robust climate estimates, exhibiting lower biases and smaller prediction errors compared to traditional proxies such as pollen and chironomids. The improved performance of sedaDNA-based methods primarily stems from their higher taxonomic resolution, enabling clearer ecological and climatic interpretations. Furthermore, sedaDNA reflects local vegetation conditions more directly, owing to its immediate catchment origin, whereas pollen integrates signals over broader, regional scales. However, like all assemblage-based methods, sedaDNA reconstructions exhibit systematic biases such as the “edge effect”, typically overestimating cooler climates and underestimating warmer climates. Addressing these biases, as well as carefully managing taxa with sparse occurrences, is essential for further improving the reliability of sedaDNA-based climate reconstructions. Finally, we note the risk of circularity when inferring drivers of vegetation change from the same assemblages used to reconstruct climate; therefore, our reconstructions should not be used to attribute drivers of vegetation change.