Millennial Variations and a Mid-Holocene Step Change in Northern Mid-latitude Moisture Gradients

Holocene records reveal a constantly varying hydroclimate characterized by responses to precession punctuated by decadal-to-centennial ‘megadroughts’ and rapid state shifts. How such changes relate across space and time can reveal the underlying dynamics and how external forcing, intrinsic variability, and various feedbacks interact to alter societally critical water supplies. Here, Holocene water-level changes were examined in two groups of North American lakes to systematically characterize millennial-scale hydroclimate variability and potential large-scale state changes. The records constrain changes at the ends of the large hydroclimate gradient between the semi-arid Rocky Mountains and the humid Atlantic coast. Geophysical surveys and 40 radiocarbon-dated sediment cores provide direct measures of past shoreline positions from the 12 lakes. None exhibited stable Holocene water levels. Together they show a steep east-west gradient from 9-5.5 ka and again from 4.5-2.1 ka. The gradient was unusually weak from 11-9, 5.5-4.5, and after 2.1 ka when Rocky Mountain lakes reached their maxima. Consistent with interconnected atmospheric circulation and land surface energy budget changes, the gradient steepness correlates with mid-continental summer temperature changes (r = 0.73) with a cool, wet mid-continent associated with a weak hydroclimatic gradient, such as during the anomalous mid-Holocene uctuation from 5.5-4.5 ka. The millennial-scale variability interacted with long-term trends to rapidly increase effective moisture in the west at 5.5 ka, potentially as part of state shifts extending to the Sahel. The interconnected changes underscore the possibility that poorly diagnosed centennial-to-millennial variability could accelerate some Holocene trends to produce abrupt shifts without requiring strong threshold effects.


Introduction
Interglacial temperatures may be broadly stable (Marsicek et  Geological evidence of past lake-level changes represent an archive of Holocene hydroclimate changes that can be systematically evaluated over large areas to investigate such dynamics (Harrison et al. 2015). Across the large precipitation gradient in mid-latitude North America, where annual precipitation falls from >1500 to <700 mm/yr between the Atlantic coast and the Rocky Mountains, many small lakes and ponds responded to Holocene changes (Stone and Fritz 2013; Liefert and Shuman 2020). As they did, lake shorelines changed position and sediment stratigraphies preserved evidence of these shifts (Winkler et al. 1986;Digerfeldt et al. 1992). Where lakes rose, the area of ne-grained muds deposited beneath deep water expanded. Where lakes fell during dry periods, the area of muds contracted as shoreline sands expanded basinward. The changes appear clearly in geophysical (e.g., ground-penetrating radar) surveys of the lake sediments and can be analyzed and radiocarbon dated when sampled by transects of sediment cores collected in shallow water within tens of meters of the modern shorelines (Digerfeldt 1986;Pribyl and Shuman 2014).
Over the past decade, over a dozen lakes at either end of the mid-latitude moisture gradient across central and eastern North America have been examined using the same systematic approach (Pribyl and Shuman 2014) involving 1) geophysical surveys of dozens of lakes in each region to identify target sites with well-de ned sequences of paleoshorelines, 2) collection of transects of sediment cores from different water depths and distances from shore to sample paleoshoreline deposits in each target lake, 3) analyses and radiocarbon dating of the core sedimentology, and 4) systematic reconstruction of lake elevations based on application of a decision-tree algorithm to the core data (Shuman et al. 2009 . The decision-tree approach creates ensembles of reconstructions for each lake by iteratively classifying sediments within each core as representing shallow or deep environments and then using the classi ed data from multiple cores to systematically infer the changing elevations of the boundary between the two facies (i.e., the lower elevational limit of shoreline sediments) through time as a measure of lake-level history (Marsicek et al. 2013;Pribyl and Shuman 2014).
In the northeastern United States, closely co-located lakes were studied to replicate the reconstruction results (Newby et al. 2014; Shuman and Burrell 2017), which were found to include replicable patterns of both multimillennial trends and multi-century variations. The variations correlate with pollen-inferred annual precipitation reconstructions, which accurately drove forward models of transect core stratigraphic features (Shuman et al. 2019). Here, the results are compared to those from lakes in the semi-arid Rocky Mountain region at the opposite end of the moisture gradient, which also reveal a range of Holocene variations including evidence of north-south changes in effective precipitation over centuries to millennia  Serravezza 2017).
These lake-level reconstructions are synthesized here to constrain the dynamics of the east-west precipitation gradient controlled by delivery of moisture from the Atlantic Ocean. Potential in uences include orbital forcing and ocean variability (Webb III et al. 1998 . The lakes only represent two end-member regions, and interpretations are inherently limited as a result, but the comparisons provide evidence of Holocene changes in the continental moisture gradient grounded in direct physical evidence of past water-level changes. A larger network of nearly 200 lakes with some geomorphic or stratigraphic evidence of lake-level change shows substantial long-term hydroclimate changes across North America, but most of the records involved lack detail (Liefert and Shuman 2020). Just as maps and atlases of dendroclimatic drought reconstructions reveal the power of using a single, coherent approach over broad geographic areas to study the synoptic patterns of hydroclimate change at interannual to decadal scales (Cook et al. 1999(Cook et al. , 2007(Cook et al. , 2015, the comparison of reconstructions derived from 40 radiocarbon-dated sediment cores from 12 lakes clustered in four U.S. states offer a systematic perspective on the centennial-to-millennial dynamics of a major mid-latitude moisture gradient spanning several thousand kilometers.

Methods
The twelve lake-level reconstructions discussed here cover at least the past 4000 years and ten span the whole Holocene (Table 1) All of the reconstructions were developed using surveys of paleoshoreline deposits made using groundpenetrating radar (GPR) and derive directly from grain size, loss-on-ignition, or x-ray uoresce (XRF) data that document facies changes in at least two cores collected at different water depths; all records were interpreted using the same decision-tree approach (

Results
Declines in the reconstructed water levels indicate changes in the elevational extent of sandy littoral sediments, which interrupted deep-water mud accumulation in near-shore cores from the 12 lakes; reconstructed increases represent periods of shoreward expansion of muds (Fig. 1). The records indicate that all of the lakes experienced multiple millennia when they were lower than today, although the timing varies by region. Additional rapid changes or multi-century variability further distinguish the different regions with added variability among individual lakes resulting from local hydrogeological in uences or reconstruction limitations ( Fig. 1).
Regional averages a rm patterns common across multiple lakes (Fig. 2). Eastern lakes all indicate signi cantly lower than modern water levels prior to ca. 8 ka ( Fig. 1-2, blue lines), but western lakes indicate maximum lowering both in the late-Pleistocene (>11 ka) and during portions of the mid-Holocene at ca. 8-2 ka ( Fig. 1-2, orange and black lines).

Major millennial-to-centennial features
In eastern lakes, water levels rose rapidly between 10-8 ka with additional increases common afterward, especially in coastal lakes (New Long and Deep ponds). Multiple multi-century uctuations affected these records since 8 ka with low water phases widely recorded at 4.2-3.9, 2.9-2.1, and 1.3-1.2 ka (late-Holocene gray bars, Fig. 1-2), which are well constrained by multiple 14  In most western lakes, water levels rose earlier than eastern lakes ( Fig. 1-2). Most record late-Pleistocene minima, which speci cally date to the Younger Dryas chronozone from ca. 12.6-11.3 ka at Lower Paintrock Lake, Lake of the Woods, and Upper Big Creek Lakes (Fig. 1). Lakes in northwest Wyoming then reached maxima at near-modern levels from ca. 10.5-9.2 ka (gray bar, Fig. 1-2), and Colorado lakes, and the eastern most northern Wyoming lake (Duncan Lake), followed with maxima from ca. 9.2-8.2 ka (orange lines, Fig. 1-2).
These lakes then fell with low levels recorded from ca. 8-5.5 ka in northwestern Wyoming (black lines, Fig. 1-2) and from ca. 7-5.5 ka in Colorado (orange lines, Fig. 1-2). Two Colorado lakes (Hidden, Emerald) have no records of early Holocene variations and only the contain stratigraphic evidence of late-Holocene changes (Shuman et al. 2009).
An increase in water levels at ca. 5.5 ka terminated or interrupted the mid-Holocene low water phase at nearly all western lakes ( Fig. 1-2). The rise is not signi cantly different in time between Lake of the Woods in northern Wyoming and Emerald Lake in central Colorado . After the rise, many of the southern lakes fell again and reached second minima before ca. 2.1 ka. Pronounced low water episodes developed by ca. 4 ka at Paintrock, Little Windy, Big Creek, and Emerald lakes ( Fig. 1-2). Nearly all lakes have been near their modern high levels for the past two millennia.

Principal Components Analysis
The rst principal component of the lake-level dataset captures 55% of the variance and represents a common long increase in moisture across all sites (top, Fig. 3). The second component, representing 15% of the variance, summarizes the north-south differences between northwest Wyoming (blue symbols, middle map in Fig. 3) and Colorado (orange symbols, middle map in Fig. 3). The second component scores (middle, Fig. 3) capture a set of latitudinal contrasts created when 1) northern lakes declined before southern lakes before ca. 8 ka and 2) the decline of southern lakes from ca. 4.2-2.1 ka when northern lakes began to rise (Fig. 2).
The third component represents major changes in the east-west gradient and 13% of the variance in the dataset (bottom, Fig. 3). The scores emphasize that the moisture contrast between these areas was muted in the early Holocene before ca. 9 ka when western lakes were high and eastern lakes low, ampli ed as western lakes fell and eastern lakes rose from ca. 9-5.5 ka, relaxed during a millennial anomaly from ca. 5.5-4.5 ka when most western lakes rose and the two eastern inland lakes (Davis and Blanding) fell, and ampli ed again as Colorado lakes fell and eastern lakes continued to rise from ca. 4.5-2 ka ( Fig. 1-2).
PCA speci c to each region (Fig. 4) reveals similar patterns with the rst component of the western and eastern subsets well predicted by combinations of precession trends and the east-west millennial variations.
PC1 of both the eastern and western sub-sets of records track the long-term Holocene increase. The difference between the two (DE-W in Fig. 4C) is not signi cantly different from PC3 of the full dataset (Fig. 3, bottom).
Adding the DE-W curve (or PC3 from Fig. 3) to a scaled precession curve (thin "inso" curves in Fig. 4A-B) replicates the major features of the western PC1 scores (r = 0.95) with a prominent step shift at ca. 5.5 ka; subtracting the DE-W curve replicates the eastern PC1 scores (r = 0.95) with a step change at ca. 4.5 ka (Fig.   4).
The trends recorded by PC3 of the whole dataset (

Discussion
The records synthesized here con rm that lakes across mid-latitude North America have uctuated on a wide range of time scales despite their locations in humid or snow-dominated regions. Their uctuations indicate multi-millennial to centennial variations in the moisture gradient across semi-arid to humid mid-latitude North America during the Holocene when external climate forcing was modest and monotonic. At face value, they imply three major geographic patterns of hydrologic change during the Holocene (Fig. 3).
First, effective moisture increased over much of the region since the last glacial period. Late-Pleistocene climates may have been arid across much of the region as most lakes in both the eastern and western clusters were low from 15-11 ka (Fig. 1-2). In addition to effects related to the ice sheets, the change may relate to long- Second, the east-west gradient has not been stable, but varied in strength on multi-millennial time scales (Fig.   2-3). A rise in the level of western lakes marks the beginning of the Holocene and represents a reduced gradient before 8 ka compared to today (Fig. 2). Other mid-continental moisture records also provide evidence of high effective moisture in the early Holocene (prior to the mid-Holocene arid period there), such as represented by dust deposition at Elk Lake, Minnesota (Dean 1997) (Fig. 5), diatom records from sites such as Moon Lake, South Dakota (Laird et al. 1996), and numerous fossil pollen records (Bartlein et al. 1984 The effects would have favored the weak early-Holocene gradient, but after 8 ka, eastern lakes rose, western lakes fell, and the east-west gradient appears to have become steeper than today (Fig. 2).
Subsequent weakening of the gradient re-developed from 5.5-4.5 before it returned to a near modern difference ( Fig. 2-4). The weak gradient (wet west-dry east) coincided on millennial time scales with cool midcontinent summers whereas warm continental summers correlated with a strong gradient (dry west-wet east) at other times in the Holocene (Fig. 5). The relationship could relate to reinforcing changes between precipitation, soil moisture, and the resulting effects of latent heating on air temperatures ( In Colorado, maximum aridity occurred later from 8-2 ka (Fig. 2-3). The north-south difference in the timing of the Rocky Mountain aridity could represent long-term shifts in winter precipitation patterns similar to those associated with interannual to multidecadal variations over the Paci c Ocean (Wise 2010; Pederson et al. 2011). However, the difference between PC2 and PC3 (Fig. 3) indicates that the western lakes were in uenced by multiple dynamics. Both regions experienced at least a modest reduction in the aridity at ca. 5.5-4.5 ka, although the wet millennial period appears to have been drier than today in the south (e.g., Upper Big Creek and Emerald lakes).
The increase in western water levels at ca. 5.5 ka may relate to large-scale hemispheric phenomena (Fig. 6-7 The Rocky Mountains have some of the highest correlations with Sahelian precipitation outside of Africa, especially in the northern hemisphere summer months (Fig. 7A). Reduced precipitation in the Sahel tends to coincide with increased African surface pressures and increased 500 mb geopotential heights (Fig. 7B) just as expected in response to a reduction in the African monsoon in response to precessional forcing (Claussen et al. 2017;Kelly et al. 2018). Easterly (dust laden) winds from the Sahel are then enhanced, while mid-latitude westerly zonal velocities decline between central North America and western Europe (Fig. 7C). Speci c humidity increases over the Americas (Fig. 7D) as water vapor is advected northward into (Fig. 7F) and not eastward away from (Fig. 7C) the North American mid-continent where uplift produces precipitation (Fig. 7E).
The changes in surface heating in Africa, thus, propagate to North America through dynamics linked to the North Atlantic Subtropical High, which have consequences for the east-west moisture gradient and have analogs in mid-Holocene simulations (Kelly et al. 2018).
If the change at 5.5 ka represents a step-shift in Holocene climates that connected Africa, the North Atlantic, and North America (Fig. 6)(deMenocal et al. 2000a; Marsicek et al. 2018), it also marks the beginning of the anomalous millennium when the east-west moisture gradient was reduced (Fig. 3-4). The period from 5.5-4.5 ka appears as a prominent millennial anomaly in many regions of mid-latitude North America (Fig. 5) (Shuman, submitted) and more broadly in the northern hemisphere ( Great Plains, loess activity declined at 5.5-5.2 ka, remained low for >300 years, and then increased after 4.9-4.5 ka (Fig. 6) (Dean 1997). A prominent A b soil horizon amid the Holocene Bignell Loess in the western Great Plains dates to the same wet millennial period (Miao et al. 2007a) when prairie pollen records mark a millennial decline in Ambrosia (ragweed) pollen (Grimm 2001) and diatoms indicate a millennial freshening of Moon Lake, North Dakota (Laird et al. 1996). Further west, submerged tree stumps in Lake Tahoe Potentially, as indicated by combinations of precessional trends and the millennial east-west changes (Fig. 4), the apparent step change at 5.5 ka could represent the interference of long-term Holocene trends and a prominent millennial-scale variation (Fig. 8). In some areas, such as in the northern Rocky Mountains (Fig. 4B) and the Sahel as recorded by ODP658c (Fig. 8A-B), the two changes constructively interfere to accelerate longterm trends, but in other areas, such as Elk Lake, Minnesota (Fig. 5) and Lake Bosumtwi, Ghana (Fig. 8A), the two patterns may have negatively interfered to produce a millennial oscillation (Fig. 5). The outcome depends upon the relative magnitudes and signs of the trends and millennial uctuation, such that negative interference could appear as a delayed abrupt shift like recorded at ca. 4.5 ka in northeastern lakes (Fig. 4A) Thornalley et al. 2009). The interference with orbital and ice volume trends differs, however, from that observed during abrupt Pleistocene events associated with North Atlantic overturning such as the Younger Dryas (YD) because the east-west gradient in North America decreased after the YD and increased after the anomalous millennium from 5.5-4.5 ka (Fig.   8C). In Africa, Ghana and the Sahel appear to have changed in concert during the YD, but in opposition at 5.5 ka (Fig. 8A).     Orange shading marks the anomalous millennium between the later two. Time series of the area of the Laurentide ice sheet ( lled gray) (Dyke 2004) and June insolation anomalies (gray line)(Berger and Loutre 1991) represent the long-term Holocene climate forcing trends, but other records including the difference between eastern and western lake-level records (regional PC1 score differences, the inverse of "△EW" from Fig. 4, bold blue line) record millennial variations, such as from 5.5-4.5 ka. Other series include carbon isotopes from the North Atlantic (thin blue line) (Oppo et al. 2003); mean temperatures of the warmest month from mid-latitude North America (as anomalies from the Holocene mean, orange line), mid-continent North America (a 6-record mean, bold black line), and Sharkey Lake, Minnesota (thin black line) (Shuman and Marsicek 2016); and dust deposition at Elk Lake, Minnesota (tan lled) (Dean 1997).