Atmospheric river activity during the late Holocene exceeds modern range of variability in California

The study investigates how atmospheric river (AR) activity and associated extreme precipitation in California have varied over the last 3,200 years, addressing the limitation of short instrumental records for flood hazard assessment. ARs are narrow corridors of concentrated water vapor that deliver substantial precipitation to western North America and are responsible for many of California’s largest floods. Recent winters (e.g., 2022–2023 with 31 AR storms) demonstrate the dual role of ARs in ending droughts and triggering damaging floods and landslides. Climate models project increases in extreme precipitation and runoff in California due to strengthening ARs, but uncertainties remain because instrumental data span only recent decades. Lake sediments can archive extreme precipitation beyond instrumental periods. This work develops and applies a high-resolution sedimentary proxy (Si/Al and clay layers) at Leonard Lake to reconstruct AR-related integrated vapor transport (IVT) and place modern extremes in a late Holocene context.

- Prior studies link ARs to major floods in California and show projected increases in extreme precipitation and runoff, implying elevated flood risk but with uncertainty due to limited records. - California lacustrine archives typically resolve centennial–millennial pluvial variability (e.g., Sierra Nevada and southern California basins) but often lack the temporal resolution to identify individual or decadal extreme events. - Event deposits in lake sediments, characterized by lithological, grain-size, and geochemical ‘fingerprints’, can capture extreme precipitation; elevated siliciclastic elements (Si, Ti, Al) are associated with runoff from intense storms and with clay deposition. - Calibration-in-time approaches have been used to quantitatively link sediment proxies to instrumental climate variables, enabling reconstruction of hydroclimatic metrics such as IVT. - Regional paleoclimate reconstructions indicate periods such as the Medieval Climate Anomaly (MCA) and Little Ice Age (LIA) with characteristic aridity or wetness, providing external benchmarks for validating new records.

- Field and coring: In April 2014, two surface cores and three overlapping long cores (total composite length 446 cm) were retrieved from the depocenter of Leonard Lake (Mendocino County, CA; 39.2725°N, 123.3633°W; 613 m a.s.l.) using Bolivia and Livingston corers; cores were split, logged, imaged, and sampled. Bathymetry was mapped using a Norcross H22PX depth sounder with Trimble GeoXH geolocation. - Stratigraphy and sedimentology: Visual descriptions at 0.5-cm resolution classified intervals as clay layers, fine laminations, homogenous sediments, or sand layers. Grain size was measured contiguously at 0.5-cm after oxidation (NaOCl), carbonate removal (10% HCl), and analyzed on a Beckman Coulter LS13 320. - Geochemistry and physical properties: Water content (50 °C), LOI at 550 °C (organic matter) and 960 °C (carbonates) at 1-cm intervals; non-carbonate inorganics by mass balance. Elemental concentrations were measured at 1-cm resolution on split cores with a Thermo Scientific Niton XL3t GOLDD p-XRF using certified standards (NIST 165a, 2709a; JMS-1). - Geochronology and age modeling: 17 AMS 14C dates on discrete organics (multiple labs) and short-lived radionuclides (210Pb, 137Cs, 226Ra) from the upper 60 cm (gamma spectrometry). 137Cs peak at 28 cm indicates 1963 CE. Background 210Pb reached ~50 cm. A Bayesian Plum (v0.3.0) age–depth model was constructed; to propagate uncertainty within geoChronR, a Bacon age ensemble (n=1000) was generated using modeled 210Pb ages and 137Cs. Plum and Bacon models agreed within ~3.3% on average. Mean deposition time: 7.4 yr cm−1 overall; 1.98 yr cm−1 during the 1948–2015 calibration period. Average age uncertainty ±74.5 yr over 3200 years; ±3.5 yr during calibration. - Proxy selection and calibration-in-time: Si/Al and percent clay were used as proxies for extreme-precipitation-driven siliciclastic influx. PCA on compositional data confirmed grouping of siliciclastics (Si, Ti, Al, K) and distinct behavior of Fe and organic matter (with Br). Proxy–instrument relationships were assessed using Pearson correlations across age ensembles with effective sample size and false discovery rate (FDR) control. - Instrumental AR metrics and regression: Annual sums (5-year binned) of IVT (kg m−1 s−1) and IWV (kg m−2) for ARs at 40°N from the Gershunov database were calculated. For each iteration, a selected age ensemble binned Si/Al (or clay) to 5-year intervals aligned with instrumental bins. Linear regression with scaling to mitigate regression dilution linked proxies to IVT (and IWV); ensemble regression and age ensembles propagated both calibration and chronological uncertainties to reconstruct IVT back 3200 years. Local streamflow (Russian and Navarro Rivers) was also tested but not retained due to non-significant correlations.

- The Leonard Lake core contains recurring light greenish-gray clay layers interspersed with fine laminations and homogeneous zones; clay layers correspond to extreme precipitation events delivering fine siliciclastic material to the depocenter. Fine laminations are associated with elevated Fe/Al and suboxic–anoxic benthic conditions. - Si/Al is strongly and significantly correlated with integrated vapor transport (IVT): median r = 0.63 (all ensemble members significant after FDR control; p_median ≈ 0.02). Si/Al also correlates with integrated water vapor (IWV): median r = 0.48, with 96% of ensembles significant. - Clay percentage correlates positively with Si/Al; visually distinct clay layers occur above ~9% clay. Clay values during −1050 to −550 CE exceed the instrumental-period maxima, while instrumental-period Si/Al spans nearly the full record range. - Reconstructed median IVT shows values comparable to the modern period (~40,000 kg m−1 s−1) last occurred before the MCA onset (~950 CE), followed by a multi-century decline to a nadir near −27,500 kg m−1 s−1 in the 1500s CE, and then rising during the Little Ice Age (1300–1850 CE). - The largest reconstructed median IVT peaks occur around 70 CE and 860 CE, reaching ~44,000 kg m−1 s−1, indicating pluvial episodes exceeding the instrumental-era range. - The late 20th century exhibits the highest median IVT since the start of the MCA, with an overall increase through the LIA. - Regional coherence: The reconstruction aligns with blue oak tree-ring reconstructions of heavy precipitation (increases since the 1700s), with low IVT during MCA aridity, and with Sacramento River flow reconstructions showing extremely dry intervals in the early 1300s and individual dry years in the early 1500s. - The data suggest that California has experienced pluvial episodes larger than any recorded in the instrumental era, particularly two to three millennia ago, underscoring a broader natural range of AR-driven hydroclimate variability.

The research question—whether extreme AR-driven precipitation in California has exceeded modern levels over millennial timescales—is addressed by calibrating sedimentary Si/Al and clay proxies to instrumental IVT and reconstructing paleo-IVT with propagated age and calibration uncertainties. The strong proxy–IVT relationship and the clear stratigraphic signal of clay layers provide confidence that Leonard Lake reliably records extreme precipitation events. The reconstruction demonstrates that multiple late Holocene pluvial episodes were more intense than any observed in the instrumental record, implying that current planning based solely on modern data may underestimate potential flood risk. Trends through key paleoclimate intervals (low IVT during the MCA; rising IVT through the LIA) align with independent regional records, indicating that extreme precipitation modulates pluvial–drought cycles over centuries to millennia. The comparison with broader hemispheric gradients suggests AR activity may be decoupled from simple latitudinal temperature gradient controls, reflecting the distinct dynamics of ARs and potential sub-regional influences (e.g., tropical Pacific SSTs, insolation-driven thermodynamics). The findings contextualize recent AR activity within a long-term framework and emphasize the necessity of incorporating paleo-informed extremes into hydrologic risk assessments for California.

This study provides the first quantitative late Holocene reconstruction of AR-related integrated vapor transport for the western United States using a high-resolution, well-dated lake sediment record. By calibrating Si/Al and clay layers from Leonard Lake to instrumental IVT, the authors show that California experienced pluvial episodes two to three millennia ago that exceeded the magnitude of instrumental-era events, while also documenting increased AR-related activity during the Little Ice Age and high late-20th-century values. These results expand the known range of hydroclimatic variability and have direct implications for flood hazard estimation, water resource planning, and emergency management under a changing climate. Future research should target additional high-resolution sites across the region to map spatial variability in paleo-AR activity, assess potential non-stationarity in proxy–climate relationships, integrate multi-proxy approaches to improve robustness, and couple reconstructions with climate model simulations to elucidate dynamical drivers.

- Potential non-stationarity: The relationship between Si/Al (and clay) and extreme precipitation/IVT could have changed through time due to shifting catchment processes or climate states; this cannot be fully accounted for. - Site-specific factors: Variations in lake level, basin geometry during early lake development, and vegetation cover could alter sediment transport and preservation, influencing proxy expression independent of AR intensity. - Chronological uncertainty: Although age modeling and ensembles were used, average age uncertainty over the full record (~±74.5 years) limits precise event-level attribution; the reconstruction is best interpreted at multi-decadal to centennial scales. - Instrumental comparability: Calibration relies on the recent period and binning (5-year), which may smooth high-frequency variability; streamflow correlations were positive but not significant and thus not used, limiting direct hydrologic calibration. - Event specificity: The record cannot resolve individual storm events; it reconstructs aggregated AR-related intensity metrics (IVT) over time.