The CASPROD database integrates mineralogical and geochemical data covering the entire Arctic Ocean. To facilitate a systematic description and spatial analysis of these datasets, the Arctic Ocean is subdivided into two primary physiographic provinces, the Central Arctic Ocean and the Arctic Marginal Seas, following the classification scheme proposed by Jakobsson (2002) and Martens et al. (2021). Bathymetric constraints from the General Bathymetric Chart of the Oceans (GEBCO) were employed to delineate the shelf boundaries of these seas (Jakobsson, 2002). Within this framework, the Central Arctic Ocean encompasses the deep abyssal basins, specifically the Canada, Makarov, Amundsen, and Nansen Basins. The Arctic Marginal Seas are characterized by extensive continental shelves surrounding these basins, including the Beaufort, Chukchi, East Siberian, Laptev, Kara, and Barents Seas, as well as the shelf regions adjacent to the Canadian Arctic Archipelago (Fig. 1, Table 1).

The mineralogical and geochemical composition of Arctic marine sediments is primarily controlled by the lithology of the surrounding continental source regions. As the principal sediment sources, these geologically diverse provinces bordering the Arctic Ocean impart distinct and diagnostic signatures to the adjacent marine environments (Stein, 2008; Harrison et al., 2008; Pease and Coakley, 2018; Myers and Darby, 2022).

The North American sector, encompassing the Canadian Arctic Archipelago and northern Alaska, is characterized by Precambrian rocks of the Canadian Shield, Paleozoic sedimentary successions of the Arctic Platform, and Mesozoic-Cenozoic orogenic belts associated with the North American Cordillera (Harrison et al., 2008). Geochemically, these ancient continental terrains are characterized by strongly negative ϵNd values and highly radiogenic ⁸⁷Sr / ⁸⁶Sr ratios. They serve as the primary source of Precambrian zircons and contribute substantial amounts of detrital carbonate minerals (e.g., dolomite and calcite) to the Arctic Ocean.

In contrast, the Eurasian sector exhibits a markedly different geological framework. The large drainage basins of major Siberian rivers, including the Ob, Yenisei, Lena and Kolyma rivers, erode the Mesozoic-Cenozoic sedimentary sequences of the Verkhoyansk-Chukotka region and extensive volcanic provinces, such as the Okhotsk-Chukotka Volcanic Belt (OCVB), the Kolyma-Omolon Belt and the Siberian Traps (Toro et al., 2016). As a result, sediments originating from these regions commonly yield more radiogenic ϵNd values, less radiogenic ⁸⁷Sr / ⁸⁶Sr ratios, and abundant Mesozoic–Cenozoic zircon populations. Mineralogically, the Eurasian sector represents a major source of characteristic clay minerals, including illite derived from sedimentary rocks and smectite produced by the weathering of basalts, as well as distinctive detrital minerals (Stein, 2008). Furthermore, the Siberian and Baltic Cratons contribute materials derived from high-grade metamorphic rocks to the Laptev, Kara and Barents Seas. This pronounced lithological heterogeneity establishes unique sediment-source fingerprints for each marginal seas.

Surface sediment samples included in the CASPROD database were primarily collected from the uppermost layers (typically 0–2 to 0–5 cm). For sediment cores, the core-top intervals were used to represent the modern sedimentary conditions. All datasets are provided in standardized data table (Table 2) with a uniform metadata structure containing the following categories and fields. Station metadata: station identifier (“STATION”), geographical coordinates (“LAT” and “LON”), geographic region (“REGION”), sample position within the core or sediment column (“SAMPLE_PS”). Bathymetric information: water depth derived from shipboard measurement (“WATER_DEPTH”) and from the GEBCO bathymetry (“GEBCO_DEPTH”). Water depth values were preferentially obtained from the original publications or cruise reports. For stations where shipboard depth information was unavailable, water depth was estimated using the GEBCO bathymetric grid. Provenance metadata: sample pretreatment methods (“PRETREATMENT”), analytical instrumentation (“INSTRUMENT”), and data source (“REFERENCE”).

Although the depositional ages of these surface and core-top samples may vary slightly, they are all concentrated within the Holocene. Over this relatively short geological timescale, the primary circum-Arctic sediment source configurations and geological basements have remained remarkably stable. Consequently, any minor variation in depositional age among these surface and core-top sediments is practically negligible when utilizing CASPROD as a baseline to track provenance shifts over longer geological timescales, such as Quaternary glacial-interglacial cycles.

The CASPROD database integrates four primary provenance proxy categories, providing key multi-proxy data to constrain sediment sources and transport pathways within the Arctic Ocean: Sr–Nd isotopes, detrital zircon U–Pb ages, clay mineral assemblages, and detrital mineral compositions (Table 2). The clay mineralogy dataset reports the relative abundances of four major groups: kaolinite, illite, smectite, and chlorite. The detrital mineral dataset comprises a diverse suite of mineral types, generally categorized into heavy- and light-mineral fractions (Table 2). All compiled datasets are publicly accessible through the repository described in the Data Availability section.

The CASPROD dataset synthesizes data from 117 previously published studies and publicly available databases, all of which are fully and appropriately cited. To ensure complete traceability, each data record is explicitly linked to its original source reference within the primary data tables. A comprehensive bibliography of all source publications is provided in the Supplement. Furthermore, to enable rigorous user-assessed quality, the database documents detailed metadata for each record. For detrital zircon and detrital mineral data obtained through microscopic identification, the corresponding pre-treatment methods and instruments are provided. For clay mineral and detrital data derived from XRD analysis, the dataset includes pre-treatment procedures, XRD instruments (including the type of X-ray tube, e.g., Cu or Co), whether the slit system is automatic or fixed, and the software used for data calculation.

To facilitate spatial analysis and visualization, the CASPROD database provides interpolated raster products (GeoTIFF format) for Sr-Nd isotopes, clay mineral assemblages, and detrital mineral distributions. All spatial data are projected using the WGS 1984 Arctic Polar Orthographic projection to minimize distortion at high latitudes. Interpolation was performed using Ocean Data View software (ODV; Schlitzer, 2022), specifically employing the Data-Interpolating Variational Analysis (DIVA) method (Brasseur et al., 1996; Troupin et al., 2012). DIVA is a variational interpolation technique that estimates continuous fields by optimizing a cost function to balance data fidelity, field smoothness, and consistency with physical boundaries constraints such as complex coastlines. In contrast to traditional kriging methods, DIVA explicitly accounts for anisotropic spatial correlations and incorporates boundary conditions, making it exceptionally suitable for oceanographic datasets characterized by complex basin geometries and irregular data coverage. As implemented in ODV, DIVA allows user-defined correlation lengths in both longitudinal and latitudinal directions, generates gridded estimation error fields, and supports isopycnic gridding when density-based coordinates are applied. Consequently, this approach improves the representation of spatial gradients in regions with heterogeneous data coverage, particularly between the Arctic shelf seas and central deep basins.

Although the DIVA method successfully integrates topographic features and complex coastlines to optimize spatial continuity, the accuracy of the interpolated values remains highly dependent on data density. In well-sampled regions, such as the Chukchi Sea, Laptev Sea, and parts of the Barents-Kara Seas, the grid cells represent robust local signals. Conversely, in regions with sparse sampling coverage, most notably parts of the deep Canada Basin and portions of the Canadian Arctic Archipelago, the interpolated values carry relatively higher uncertainty. Consequently, the interpolated results in these data-sparse or unsampled areas should be treated with caution and utilized primarily as a qualitative reference.

This section presents an overview of the CASPROD database inventory and examines the spatial distribution of the four primary provenance proxy groups across the Arctic Ocean. By systematically mapping the data coverage across the Arctic Ocean, we highlight regional variations in data density and identify areas where information remains limited. The geographical patterns of key geochemical and mineralogical parameters are illustrated using a series of maps and diagrams.

The multi-proxy synthesis in CASPROD effectively links shelf deposits with the geological characteristics of their respective hinterlands, providing critical fingerprints to trace pan-Arctic sediment transport. The Sr-Nd isotopic compositions reflect the crustal residence age of source rocks (White, 2013). High ⁸⁷Sr / ⁸⁶Sr ratios and low ϵNd values in the Canadian Arctic Archipelago, Beaufort, Barents, and Laptev Seas define a characteristic “cratonic” signature, indicative of sediment contributions from ancient Archean basement rocks, including North America, Baltica, and Siberia (Bazhenova et al., 2017). Conversely, the Eurasian sector exhibits “younger” isotopic signatures. The unradiogenic ⁸⁷Sr / ⁸⁶Sr and relatively high ϵNd values in the Kara, East Siberian and Chukchi Seas successfully trace sediment inputs derived from the Permian-Triassic Siberian Large Igneous Province (SLIP; Tütken et al., 2002) and the Jurassic-Cretaceous OCVB (Li et al., 2023).

Zircon U-Pb age spectra provide further constrain on sediment provenance by identifying the formation ages of felsic igneous rocks in the source regions (Moecher and Samson, 2006). The North American sector, particularly areas influenced by the Mackenzie River system, is uniquely distinguished by a prominent 1000–1500 Ma age population, diagnostic of the Grenville Orogeny (Rino et al., 2004). In contrast, zircon age distributions across the Eurasian sector are dominated by Phanerozoic populations, that trace distinct regional geological events. These include prominent Paleozoic peaks (220–360 and 420–560 Ma) in the Barents and Kara Seas associated with the erosion of the Ural Mountains and Central Asian Orogenic Belt (Puchkov and Ivanov, 2020; Parfenov et al., 2009; Campbell et al., 1992), and Mesozoic peaks (90–160 Ma) in the East Siberian and Chukchi Seas that trace magmatic activity associated with the OCVB (Akinin et al., 2020).

Clay and detrital mineral assemblages provide complementary insights into circum-Arctic lithologies and physical weathering domains. Smectite is enriched in the Kara and Laptev Seas, reflecting the erosion of basalts from the SLIP (e.g., Rossak et al., 1999; Wahsner et al., 1999; Stein et al., 2004), while its elevated abundances in the Canadian Arctic Archipelago indicate local erosion of exposed mafic gabbros (Kingsbury et al., 2018). Kaolinite signals the reworking of ancient weathering crusts, with major sources tied to Triassic-Jurassic sedimentary rocks on Franz Josef Land (Elverhøi et al., 1989; Vogt and Knies, 2009) and the weathered North American Craton supplying the Canadian Arctic Archipelago (Stevenard et al., 2022). High illite contents in the East Siberian, Chukchi, and Beaufort Seas delineate dominant physical weathering of mixed sedimentary, metamorphic, and felsic igneous lithologies (e.g., Viscosi-Shirley et al., 2003a; Stein, 2008).

Light and heavy mineral fractions further refine the provenance interpretations. Plagioclase is particularly enriched in the Kara, Laptev, and East Siberian Seas, reflecting inputs from the SLIP and OCVB. Potassium feldspar indicates granitic or gneissic sources, linking sediments in the Laptev Sea and western Kara Seas to erosion of the Siberian Craton and Uralian basement, respectively (Vogt, 1997). Quartz dominates in regions with limited igneous influence, such as areas affected by the Pechora River and Mackenzie River system and the Taimyr-Severnaya Zemlya fold belt. Dolomite enrichment in the Canadian Arctic Archipelago provides a clear fingerprint of local Cambrian-Devonian carbonate bedrocks erosion (Clark et al., 1980; Vogt, 1997; Phillips and Grantz, 2001; Stein et al., 2010; Bazhenova et al., 2017).

Within the heavy mineral fraction, pyroxene serves as a diagnostic mafic tracer (Garzanti and Andò, 2007), linking sediments near Franz Josef Land to local mafic igneous outcrops and sediments in the Kara and Laptev Seas to erosion of the Siberian Traps (Behrends, 1999; Wang et al., 2022). High-grade metamorphic indicator minerals, such as garnet and amphibole (Behrends, 1999; Garzanti and Andò, 2007), are abundant in the Barents, Kara, and Laptev Seas, reflecting erosion of crystalline basement from the Baltic and Siberian Cratons, especially the Taimyr region.

Stein (2008) documented the spatial distribution patterns of four clay minerals (illite, smectite, chlorite, and kaolinite) and four detrital minerals (amphibole, clinopyroxene, epidote, and garnet) across the Arctic Ocean (for database and down-load see Stein, 2026a, b). Furthermore, lower-resolution distribution maps of the detrital minerals of quartz, plagioclase, K-feldspar and dolomite of surface sediments from the central Arctic Ocean and the Laptev and Kara seas are also presented in Stein (2008), based on data from Vogt (1997). Compared with the database compiled by Stein (2008), the overall spatial distribution patterns as well as mean concentrations of both clay and detrital minerals in our new dataset remain largely consistent, confirming the robustness of the earlier synthesis of mineralogy. However, the present compilation incorporates recently published data from the Canada Basin (e.g., Deschamps et al., 2018), the Canadian Arctic Archipelago (e.g., Myers and Darby, 2022), the Chukchi Sea (e.g., Li et al., 2021), the East Siberian Sea (e.g., Wang et al., 2024), and the Barents Sea (e.g., Vogt and Knies, 2009), thereby improving spatial coverage across key Arctic marginal seas. The inclusion of these new datasets enables a higher-resolution and more refined characterization of mineral distribution patterns in regions that were less well constrained in Stein (2008), particularly over the East Siberian Sea shelf and within the Canada Basin. In addition, we expand the analysis to include a broader suite of provenance-sensitive detrital minerals, such as quartz, plagioclase, K-feldspar, dolomite, pyroxene, and staurolite, further strengthening source-to-sink interpretations.

Provenance accuracy is method-dependent and is linked to the analytical unit of a given proxy. For single-grain techniques like Fe-oxide chemical fingerprinting, accuracy is typically evaluated through cross-validation against a reference to quantify incorrect source assignments (Darby et al., 2015). However, this traditional concept of accuracy is not directly applicable to the integrated-signal proxies compiled in CASPROD (e.g., Sr-Nd isotopes, clay and detrital minerals, and zircon U-Pb ages), which yield bulk or distributional signatures of sediment mixtures rather than discrete grain labels. Instead, we evaluate these proxies based on discriminability, defined here as the signal-to-noise ratio (SNR) between inter-source differences and intra-source variability (Fig. 7). A high SNR (generally >1) indicates that the compositional contrast between two sources substantially outweighs their internal natural heterogeneity, allowing for reliable provenance discrimination, with larger values yielding greater discriminability.

To evaluate pairwise source-region distances and intra-source heterogeneity, we selected statistical metrics tailored to the specific data structure of each proxy. For Sr-Nd isotopes, we applied the Bhattacharyya distance (Bhattacharyya, 1946) to measure source separation in the bivariate ϵNd–⁸⁷Sr / ⁸⁶Sr space, as it accommodates substantial differences in data dispersion among source areas without assuming identical covariance structures. For detrital zircon U-Pb age spectra, we used the Kolmogorov–Smirnov (KS) distance. As a distribution-free metric quantifying the maximum vertical difference between two cumulative age-distribution functions, the KS distance is highly sensitive to overall spectral shape and serves as a standard for nonparametric age-spectrum comparisons (Vermeesch, 2012). For clay mineral and detrital mineral assemblages, we applied the Aitchison distance to measure compositional differences between source regions, because percentage of minerals are subject to the constant-sum constraint and conventional distance metrics would introduce spurious correlations (Garzanti et al., 2012). Across all proxies, intra-source heterogeneity is represented by the median distance derived from 1000 bootstrap resampling iterations within each source group.

The calculated SNR values reveal that low discriminability typically stems from substantial signal overlap between specific regions (Fig. 7). For instance, detrital zircon U-Pb age spectra cannot effectively differentiate the Kara, Chukchi, and East Siberian seas, nor can they separate the Barents Sea and Laptev Sea (Fig. 7b). This overlap occurs because these marginal seas collectively drain extensive Phanerozoic magmatic provinces, such as the Siberian Traps and the Okhotsk–Chukotka Volcanic Belt, yielding highly convergent age distributions. Similarly, clay mineral assemblages show limited discriminability for the Barents Sea (Fig. 7c), reflecting pronounced intra-regional spatial variability driven by complex hydrodynamics and the mixing of diverse local endmembers. Detrital mineral assemblages also struggle to distinguish Eurasian interior sources, particularly the Kara, Laptev, and East Siberian seas (Fig. 7d). Sr-Nd isotopes, despite generally exhibiting high SNR values (>1), show limited discriminability between the East Siberian and Chukchi seas, or the Barents and Kara seas (Fig. 7a), underscoring the challenge of differentiating provenance with similar geological affinities.

These overlaps highlight the inherent limitations of relying on any single proxy to decipher a complex, multi-source sedimentary system like the Arctic Ocean. Accurate circum-Arctic provenance reconstruction requires a multi-proxy approach (e.g., Stein, 2008; Myers and Darby, 2022). Thus, in several studies multi-proxy approaches have been used to identify circum-Arctic sediment provenances more accurately, e.g. the combination of heavy minerals, clay minerals and major and minor elements (Schoster et al., 2000; Viscosi-Shirley et al., 2003a, b) or the combination of heavy minerals, clay minerals, and chemical fingerprinting of major and minor elements in Fe-oxide minerals (Myers and Darby, 2022).

CASPROD also establishes a multi-proxy, cross-validation framework to overcome the ambiguity introduced by these overlapping signatures. By synthesizing geochemical, geochronological, and mineralogical datasets, this integrated framework provides multiple constraints on endmembers and substantially improves the accuracy of provenance identification. Furthermore, to achieve higher accuracy when interpreting highly complex mixed signals, particularly regarding ice-rafted debris, we encourage database users to couple CASPROD with other approaches, such as the Fe-oxide grain chemical fingerprinting method (Darby, 2003; Darby et al., 2015; Myers and Darby, 2022).