An extensive search was conducted based on published articles, reports, theses, and datasets to gather all available literature on deep-time nitrogen isotopes. Initially, a keyword-based search combining geological period and nitrogen isotope was performed on Google Scholar, yielding over 3000 relevant literature sources after removing duplicates. A significant portion of the articles, however, only discussed previously published δ15N data, rather than presenting newly measured data, which were manually excluded from the data compilation. Additionally, geochemical databases such as PANGAEA (https://www.pangaea.de/, last access: 1 April 2025), EarthChem (https://www.earthchem.org/, last access: 1 April 2025), SGP (https://sgp-search.io/, last access: 1 April 2025), and NOAA (https://www.ncei.noaa.gov/, last access: 1 April 2025) were queried to ensure comprehensive coverage of dataset sources (Diepenbroek et al., 2002; Gard et al., 2019; Farrell et al., 2021). Where overlaps existed between datasets and publications, journal articles were prioritized as the primary data sources. Further filtering excluded studies on non-marine sediments, entries lacking essential metadata (e.g., geological age, latitude and longitude), and a limited number of Cenozoic records with inaccessible data. Ultimately, the curated dataset includes 417 valid sources published between 1983 and 2024, representing a comprehensive compilation of nitrogen isotope records for deep-time marine sediments.

Data from each publication were stored in various formats, including tables within the main text, supplementary files, or shared databases. Data extracted from tables and supplementary files were initially processed by computer algorithms, followed by manual verification and supplementation. For databases, data files were downloaded manually. In cases where publications did not provide direct data, data points were extracted from figures using GetData Graph Digitizer (ver. 2.24), and these entries were labeled as “plot” in the Notes section. Each publication was then organized into an individual data file with clear labeling of sources and unique site identifiers. These files were subsequently merged into a master dataset based on standardized column headers. In the final master dataset, additional metadata were curated, including geological age, latitude and longitude, lithology, depositional facies, and metamorphic grade. High-resolution ages and paleocoordinates were calculated and converted, where applicable.

Given that biogeochemical and paleoenvironmental studies based on nitrogen isotopes require the assessment of the depositional environment and post-depositional alteration, geochemical data apart from δ15N are crucial (Tribovillard et al., 2006; Robinson et al., 2012). Therefore, we collected other contemporaneously published geochemical data of the same samples as δ15N from the literature relevant to the formations in our database. All available data from each research site were included as comprehensively as possible, rather than excluding entries solely due to the absence of δ15N values. This approach allows for the potential interpolation of the time-series data. However, geochemistry fields with fewer than 100 data points in the final compilation were excluded due to their limited analytical utility, such as Mo and Fe isotopes.

To ensure the reliability and applicability of the data, each entry underwent a rigorous screening and evaluation process. Initially, we assessed the data source and its spatiotemporal context. All studies included in the database were required to report verified δ15N values with clear data provenance and well-defined spatiotemporal information. Data entries lacking traceable sources were excluded. Similarly, entries without precise geographic or temporal information were not considered. Data from geological settings representing highly localized environments or with high metamorphic grades, such as samples affected by hydrothermal activity (Martin and Stüeken, 2024) or highly metamorphosed minerals (e.g., mica; Jia and Kerrich, 2000; Busigny et al., 2003), were also excluded given that their δ15N compositions likely record alteration processes rather than seawater signatures. This filtering criterion was applied based on descriptions in the original literature rather than a fixed metamorphic grade threshold. The δ15N values for bulk rock and decarbonated rock were classified as primary entries (δ15N_bulk), while values for specific phases, such as fossil shells, kerogen, clay-bound nitrogen, and porphyrins, were categorized solely as secondary entries (δ15N_sp). Only primary entries were analyzed in the data visualizations presented later in this study.

Only δ15N obtained through standardized, widely accepted techniques were included in the database. These primarily consist of elemental analyzer-isotope ratio mass spectrometry methods applied to bulk rock, decarbonated fractions, or kerogen (Song et al., 2023), as well as denitrifier-based mass spectrometry methods for microfossils (Ren et al., 2012; Smart et al., 2018). Studies employing non-standard or unvalidated methods, such as stepwise combustion (Ishida et al., 2017), were excluded. Data from highly metamorphosed settings (e.g., hydrothermal alteration), terrestrial lakes and rivers, modern organisms and their metabolic products, and liquid phases were flagged and omitted from the database (e.g., Bebout et al., 1999; Chase et al., 2019; Xia et al., 2022). For data from the same site but at different depths or lithologies, or for measurements of different components in the same layer (e.g., bulk sediment and decarbonated sediment), or replicate analyses of the same homogenized sample, each entry was recorded separately to accurately capture variability.

Metadata on paleocoordinates, depositional setting, lithology, and metamorphic grade are included, wherever available. Entries were not excluded due to missing such metadata, as these can potentially be supplemented in future research. When such metadata were not directly reported in the literature, we attempted to estimate them using supplementary data or external sources, such as paleogeographic reconstructions. For entries for which metadata could not be determined, blank values were assigned.

Each data entry was assigned a unique Sample ID to distinguish it from other data entries. Geographic location information includes the modern latitude and longitude (Latitude and Longitude) referencing WGS84 (World Geodetic System 1984), obtained directly from original literature or external sources whenever possible. For studies that do not provide exact coordinates, approximate locations are estimated based on geographic descriptions or accompanying maps, using tools such as Google Maps. Additionally, we record the broader sampling region (e.g., country or oceanic region) and specific sampling site details (such as province, county, or uplift names). The location fields also include the name of the drilling site or outcrop section (SiteName), which identifies the precise drilling location or outcrop at which samples were collected, providing valuable geographic context. Certain SiteNames are uniquely associated with major drilling projects (e.g., ODP, IODP), which is important for subsequent data supplementation and analysis. Some samples also have a SampleName, as designated by the original authors – typically a code or non-unique label reflecting the naming format in the primary literature. Although multiple samples in the database may share the same SampleName, each entry has a distinct Sample ID to ensure the uniqueness of each record.

We also provide paleolatitude and paleolongitude (PaleoLatitude and PaleoLongitude), calculated based on the geological age of each sample and using paleogeographic reconstruction tools such as PointTracker v7.0, built on the plate rotation model of Scotese and Wright (2018). Paleocoordinate data are crucial for understanding the historical shifts in sample locations and their relationship to depositional environments (Percival et al., 2022; Li et al., 2025). To maintain consistency, all geographic coordinates are standardized to two decimal places.

Each entry includes not only absolute age data but also a series of geologic age-related fields to provide precise temporal context. These fields enable targeted data retrieval at a range of geological timescales, facilitating comparisons with newly added data. The GeologicalAge field captures broad temporal frameworks, recorded as Epoch for the Phanerozoic (e.g., Early Triassic) and Era for the Precambrian (e.g., Neoproterozoic). For more refined stratigraphic resolution, the Stage field (e.g, Induan) is used, with the System as a substitute for Precambrian samples (e.g., Ediacaran). The Age field records the absolute age of each sample, following the International Chronostratigraphic Chart, GTS v202309. The Formation field notes the geological unit (formation or member) from which the sample was collected, aiding in understanding its depositional context and relation to surrounding strata (Murphy and Salvador, 1999). However, Formation data are generally limited to outcrop sections, as ocean drilling samples lack specific formation designations. The Unit field identifies particular stratigraphic units or geologic event layers, such as the Cretaceous pre-OAE2 or OAE2 (Jenkyns, 2010), which aids in correlating samples within recognized geological events. The RelativeDepth field records the sample's relative depth in the section or drill core, which is essential for high-resolution age analyses and sedimentation rate calculations.

When precise ages and geological age information for each sample were provided in the original source, these values are prioritized. However, for data from the Common Era (i.e., negative ages), they are uniformly assigned a value of 0 Ma, meaning that all such data are treated as reference values for modern top sediments. Otherwise, age assignments follow two methods based on data availability. (1) For records with at least two samples or stratigraphic horizons of known age (e.g., radiometrically dated layers or well-defined stage boundaries), we constructed an age-depth model. This model linearly interpolates ages between these tie-points along the RelativeDepth axis, assuming a constant sedimentation rate within each interval between stratigraphic age tie-points. While this assumption is effective for establishing the relative temporal sequence of samples, which is critical for capturing first-order stratigraphic trends, it necessarily introduces uncertainties in absolute age determination due to potential variability in sedimentation rates or local stratigraphic features. (2) For records lacking sufficient data for an age-depth model, a single age was assigned to all samples. When only one age constraint (e.g., a radiometric date from a nearby stratum) is available, that specific age is applied. In the absence of any direct age control, the median age of the corresponding geologic stage is used as a default. It should be noted that assigning a uniform age to a suite of samples, particularly using the median stage age, carries significant uncertainty, theoretically on the order of the duration of the entire geologic interval (which can approach 100 Myr for long stages of the Precambrian). Profiles constrained by a single radiometric date, which is the predominant method for dating sequences older than 600 Ma, are generally more reliable than those relying solely on a median stage age.

The dataset includes analyses of isotopic compositions, elemental concentrations, and specific components. To maintain consistency, all units were standardized during data collection, as original publications sometimes report these data in varying units. (1) Isotopic data include δ15N, δ13C, δ18O, and δ34S, all expressed in ‰ relative to international standards. Nitrogen isotopes are reported relative to atmospheric nitrogen (Air N₂), carbon and oxygen isotopes relative to the Vienna Pee Dee Belemnite (VPDB) standard, and sulfur isotopes relative to the Vienna Canyon Diablo Troilite (VCDT) standard (Hoefs, 2021). (2) Elemental concentrations include TN (total nitrogen), TOC (total organic carbon), TS (total sulfur), CaCO₃, TC (total carbon), TIC (total inorganic carbon), P, Al, K, Si, Ca, Ti, Na, Mg, Fe, as well as iron species data and LOI (loss-on-ignition), and they are reported in weight percent (wt %). Concentrations of other trace elements are standardized to parts per million (ppm). (3) Some data originally reported as oxide concentrations were converted to elemental concentrations based on stoichiometric ratios, such as P₂O₅. (4) Additional derived values include ratios of iron species, dry bulk density, and rock eval indices (Peters, 1986; Poulton and Canfield, 2005). These indices comprise alkenone content (C37, in nmol g⁻¹), oxygen index (OI, mg CO₂ g TOC⁻¹), hydrogen index (HI, mg HC g TOC⁻¹), maximum pyrolysis temperature (Tmax⁡, °C), free hydrocarbons (S1, mg HC g⁻¹), hydrocarbons generated from rock pyrolysis (S2, mg HC g⁻¹, and CO₂ released from organic matter pyrolysis (S3, mg CO₂ g⁻¹). Some inaccessible data points were visually extracted from figures using scatterplot recognition techniques, which are marked as “plot” in the Notes field. Data with values exceeding detection limits or those erroneous (e.g., negative values for element concentration) were excluded from the dataset.

The petrological characteristic fields encompass information on lithology, depositional facies, and metamorphic grade, which provide essential contextual support for subsequent isotopic geochemistry analyses. (1) Lithology: The Lithology field records the original descriptions provided by authors, using terms such as “black shale”, “mudstone”, “limestone” and “breccia”. The LithType field classifies these lithologies into broader categories, primarily as carbonate and siliciclastic (Tucker and Wright, 2009), with minor entries for phosphorite and iron formations. (2) Metamorphic Grade: The metamorphic grade field reflects the extent of metamorphism the samples have undergone, based on original terminology whenever possible. Common terms include specific metamorphic facies (e.g., amphibolite, greenschist) as well as general descriptors like “unmetamorphosed” and “low grade”. For Cenozoic samples, which are generally assumed to have undergone minimal metamorphic alteration (Winter, 2014), any entries lacking detailed descriptions are uniformly designated as “unmetamorphosed”. (3) Depositional Setting: This field records the depositional environment of each sample, with terms like “neritic”, “peritidal”, “slope” and “abyssal” preserved from the original literature. For many ocean drilling samples, depositional settings are inferred from WaterDepth: depths of 500–2000 m are classified as “bathyal” and depths exceeding 2000 m are designated as “abyssal”.

Data in the database primarily originate from published literature and are traceable via DOI. Some data come from public databases such as PANGAEA, SGP, and NOAA. Each record includes multiple fields for source information, such as first author, publication year, article title, reference, DOI, and data source. Metadata fields have been standardized and cleaned via code to ensure consistency and machine readability, removing special characters while retaining complete citation formats. This structure allows users to trace data provenance, with DOI or Reference fields facilitating direct searches on Crossref for verification.

Latitude and longitude values were checked to confirm they fall within the valid ranges of -90 to 90 and -180 to 180, respectively. Sample coordinates were cross-referenced with country names and public national boundaries to ensure geographic accuracy. Modern sample coordinates were projected onto a global map with administrative boundaries (Figs. 2–3) to verify logical placements. If coordinates appeared on land or in other unexpected locations, each entry was manually reviewed and corrected as needed.

Paleolatitudes and paleolongitudes were calculated using the G-Plates model (Scotese and Wright, 2018) and PointTracker v7.0 software, ensuring alignment with each sample's geological age and geographical context. Site locations were plotted on paleogeographic maps (Fig. 4) for further evaluation; any inconsistencies in paleocoordinates were flagged, reviewed, and adjusted accordingly.

Frequency histograms and time-series scatter plots were generated to identify potential outliers in the dataset. All flagged extreme values underwent secondary validation against their original sources to confirm the accuracy. This process led to the correction of erroneous entries introduced during unit conversions and the removal of invalid data points that fell outside instrumental detection limits. Extreme δ15N values falling outside a conservative range (<-10 ‰ or >+40 ‰) were excluded from the final compilation (e.g., Thomazo et al., 2011; Hammarlund et al., 2019). This decision was based not on the validity of the individual measurements, but on the need to prioritize data representativeness for global-scale analysis. The excluded values, even if explained within their original publicational context, are statistical outliers that have not been corroborated and could unduly influence broad interpretations.

We conducted a comprehensive check for duplicate entries, especially for samples with similar GPS coordinates. All suspected duplicates were carefully compared, and necessary corrections were made to eliminate redundancy.

To minimize errors, geological age data were entered using a standardized template to prevent typos, inconsistencies, or incorrect values. Automated analyses and cross-verification ensured that numerical ages corresponded accurately with designated eras and geological stages. A mismatch between a numerical age and its geological stage often indicates an outdated age in the original reference (e.g., Wang et al., 2013). To address this, we recalibrated the outdated estimations by building new age-depth models based on the current geologic stage boundaries from the International Chronostratigraphic Chart (GTS v202309).

Citation information within the reference field was obtained through automated methods from the CrossRef platform, ensuring uniformity in citation formatting (Hendricks et al., 2020). We used scripts to extract comprehensive bibliographic details for each publication, including author names, title, publication year, journal name, volume, page numbers, and DOI. This automation significantly reduced potential spelling errors and inconsistencies that may arise in manual entry. Extracted citation data were cross-checked against original entries in the database, and any discrepancies or errors were corrected manually by the data management team to maintain source accuracy and completeness.

Given that the data are concentrated in the Phanerozoic, for which ages are more precisely constrained, we performed a detailed stratigraphic breakdown of age distribution by stage within the Phanerozoic (Fig. 5). The distribution is uneven, with the highest data densities in recent periods, particularly the Holocene (0–12 ka), Late Pleistocene (12–129 ka), and Chibanian (129–770 ka). The high data density in the Quaternary primarily reflects the abundance of high-resolution records from various ocean drilling projects, whose individual cores contributed large and densely sampled datasets. In contrast, older geological periods exhibit data clusters around key events, such as biotic radiations, mass extinctions, and oceanic anoxic events (Bush and Payne, 2021). Notable gaps or low-density intervals occur in the mid-Cambrian to Early Ordovician, Silurian to Early Devonian, mid-Carboniferous to Early Permian, mid-Triassic to Early Jurassic, and Late Jurassic to Early Cretaceous.

Overall, δ15N values exhibit a unimodal distribution centered around +5 ‰, with a mean of 5.1 ± 9.1 ‰ (1σ; Fig. 6a). When examining the modal values of the era-specific kernel density distributions, the Cenozoic exhibits the highest mode, followed by the Precambrian, with significantly lower modal densities in the Paleozoic and Mesozoic (Fig. 6b). The Precambrian data, which have a dispersed distribution (Fig. 7), indicate an unstable nitrogen cycle, a state potentially driven by the evolution of microbial metabolisms and later overprinted by metamorphism (see Ader et al., 2016; Stüeken et al., 2024 for further discussion). LOWESS smoothing results reveal δ15N peaks in the Neoarchean, Paleoproterozoic, and Ediacaran, i.e., periods closely associated with significant oxygenation events (Kipp et al., 2018; Koehler et al., 2019; Pellerin et al., 2024).

An examination of δ15N record reveals first-order variations on multi-million-year timescales since the Cryogenian. The LOWESS curve shows extended intervals of relatively elevated δ15N (> +5 ‰) during the Cambrian/Ordovician transition, the Carboniferous–Permian, and the late Cretaceous–Cenozoic (Fig. 7). These broad peaks are separated by periods of lower δ15N values during the Ediacaran–Cambrian, Ordovician–Devonian, and Triassic–Cretaceous. The prolonged intervals (except for the Cambrian/Ordovician transition) of elevated δ15N broadly coincide with known periods of sustained cool climates or major glaciations (i.e., the Sturtian–Marinoan glaciations, the Late Paleozoic Ice Age, and the Cenozoic Icehouse), whereas the low δ15N intervals generally align with warmer greenhouse periods (i.e., most of the late Ediacaran–early Carboniferous and the Mesozoic) (Montañez et al., 2011; Macdonald et al., 2019). This tectonic-scale pattern mirrors observations from orbital-scale glacial-interglacial cycles (Ren et al., 2017) and transient hyperthermal events like the Paleocene/Eocene Thermal Maximum (Junium et al., 2018), suggesting that climate exerts a first-order influence on the marine nitrogen cycle. The underlying mechanisms may involve variations in ice sheet extent and sea level, which affect the distribution of oxygen-minimum zones (OMZs) and the proportion of water-column denitrification versus sedimentary denitrification (Algeo et al., 2014; Wang et al., 2022). However, the correlation is not straightforward; for instance, the increase in δ15N began in the Late Cretaceous, coinciding with the onset of global cooling but preceding the major expansion of Antarctic ice sheets in the Cenozoic (Judd et al., 2024). Therefore, the exact mechanisms coupling climate and nitrogen cycle evolution remain an unsolved question for future research, ideally integrating Earth system models with the spatial δ15N data presented here.

Spatial trends in data density within the DSMS-NI database reveal substantial variability in both modern (Fig. 2) and paleogeographic distributions (Fig. 4). Ocean drilling sites are primarily concentrated along continental margins and deep-sea basins, with significant gaps in central oceanic regions (National Research Council, 2011). For older strata (pre-Cretaceous), sampling sites are clustered in North America, Europe, China, and South Africa (Fig. 2). In terms of latitude, δ15N sampling in older strata is sparse in the modern equatorial region and the mid- to high-latitude areas of the Southern Hemisphere, aside from some Southern Hemisphere samples collected from Cenozoic ocean drilling sites (Fig. 3). When modern coordinates are converted to paleolatitudes and mapped onto paleogeographic reconstructions, the Cenozoic Era provides the most extensive latitudinal coverage, with the Quaternary period contributing the highest number of sites, followed by the Cretaceous (Figs. 4 and 9). In terms of marine spatial distribution, δ15N data since the Cretaceous reflects global patterns to a certain degree (Fig. 4). However, pre-Jurassic data remain spatially concentrated, with Paleozoic sites limited to just two or three main areas. High-latitude sampling is generally scarce, with Paleozoic sites predominantly in the Southern Hemisphere and Mesozoic sites mainly in the Northern Hemisphere (Fig. 4).

To visualize spatial trends, average δ15N values from each Phanerozoic period were mapped onto paleogeographic reconstructions for the respective period (Fig. 4). Significant spatial differences exist in δ15N distribution for different geological periods. In modern ocean sediments, elevated δ15N values (notably > +5 ‰) are concentrated in regions influenced by upwelling, such as the Arabian Sea, southeastern Indian Ocean, eastern equatorial Pacific, southwestern South America, and the western coast of Mexico (Fig. 4a; Tesdal et al., 2013; Du et al., 2025b). In contrast, lower δ15N values (significantly < +5 ‰) are typically found in restricted basins or broad continental shelves, such as the Black Sea, the Mediterranean Sea, the Baltic Sea, and the South China Sea. The global mean δ15N (approximately +5 ‰, as observed in open ocean like the Southern Ocean) lies between these extremes. The modern spatial distribution of δ15N can provide a valuable framework for interpreting past marine conditions, as δ15N serves as an indicator of nutrient supply, upwelling intensity, and the extent of oceanic oxygen minimum zones (Altabet et al., 1999; Godfrey et al., 2025). However, analyzing spatial patterns in deep-time δ15N records is inherently limited by the scarcity of data, particularly from open-ocean settings, making it difficult to estimate global mean values and relative spatial gradients. For the Paleogene and Neogene, δ15N values were generally higher in the open ocean than in continental margin and restricted basins (Fig. 4b–c). In the Paleozoic and Mesozoic, δ15N values are generally negative, lacking prominent hotspots except in the Carboniferous and Permian. This pattern may reflect a systematic bias, as available data are predominantly derived from continental shelf environments (Judd et al., 2020), which tend to exhibit lower δ15N values compared to the open ocean. Despite differences in paleogeographic position and absolute δ15N values, rapid shifts in δ15N exhibit consistent directional changes (increase or decrease) during key Phanerozoic transition events, such as the Permian-Triassic boundary (Knies et al., 2013; Du et al., 2021, 2023) and the Late Cretaceous (Meyers et al., 2009; Junium et al., 2018; Du et al., 2025b). Given the current uneven distribution of sampling sites, further δ15N studies of multiple regions are crucial for enhancing our understanding of the spatial characteristics of nitrogen cycle evolution in deep time.

Each record (row) in the database includes detailed temporal and spatial metadata, along with lithology, metamorphic grade, and depositional facies information, where available. These metadata are essential for evaluating the geological context and fidelity of nitrogen isotope data. However, this version of the database has certain limitations: it may not capture all possible geological age uncertainties or precise depositional environment details for some records; significant gaps remain in the compilation of data for certain materials and time intervals (e.g. Quaternary). Consequently, users may need to independently assess and refine the metadata (e.g., chronological constraints) and supplement missing data (e.g., coral-bound δ15N records) as necessary for their specific applications. Despite our extensive efforts to accurately identify and quality-control each entry, given the vast dataset, some overlooked errors or data inconsistencies may remain. Users are encouraged to report any issues or omissions to the authors, as corrections will be incorporated into future database versions. We plan to release a new version of the dataset annually on Zenodo and update it on the Geobiology Database website. Each version will incorporate corrections to identified errors and integrate newly published data from the previous year to the fullest extent possible. This systematic update cycle is designed to ensure the dataset's accuracy, relevance, and long-term value for the research community.

In addition to δ15N data, the database provides geochemical information such as TOC, total TN, δ13C_org, and major and trace element concentrations. These supplementary data are valuable for assessing factors that may influence nitrogen isotopes, such as organic matter preservation and redox conditions. Even when not directly paired with δ15N values, we retain all relevant data to enable users to conduct correlation analyses via interpolation or other methods. Researchers are welcome to contribute additional geochemical data from the same sites or samples as they become available, allowing for updates and refinements in subsequent database releases.

When applying the database to deep-time studies, certain filtering criteria can be used. For instance, samples may be selected based on lithology, metamorphic grade, and other metadata to ensure that the data aligns with specific geological research contexts. Temporal, paleolatitude, and paleodepth information are critical for paleogeographic reconstructions and spatiotemporal distribution analyses, particularly when investigating paleoclimate change and global biogeochemical cycles. Further analysis of variations in latitude, basin characteristics, and water depth has the potential to yield significant insights. Given the rapid variability of nitrogen isotopes and their pronounced regional characteristics, filling temporal and spatial gaps and enhancing resolution are of great value – particularly for pivotal periods like the Ordovician-Silurian mass extinction, the Early Devonian terrestrial plant radiation, and the Late Jurassic-Early Cretaceous supercontinent breakup. The database is also especially suited for comparative studies of key geological periods, such as the Permian-Triassic boundary extinction and the Cretaceous OAE2. Given the inherent limitations of our simplified age-depth models, we recommend that users seeking higher chronological precision for time-series analysis incorporate additional stratigraphic constraints (e.g., paleomagnetic or cyclostratigraphic data) to develop finer-scale age models, where necessary. To support these applications, we have also provided a software tool on Zenodo, allowing users to generate heatmaps of δ15N data distributions for specific time intervals. These heatmaps visualize the average spatial distribution of δ15N for any selected geological interval, offering preliminary validation for user hypotheses and aiding in uncovering the evolution of the global nitrogen cycle.