Patterns and drivers of Holocene vegetational change near the prairie–forest ecotone in Minnesota: revisiting McAndrews’ transect


  • David M. Nelson,

    1. Institute for Genomic Biology, University of Illinois, 1206 West Gregory Drive, Urbana, IL 61801, USA;
    Search for more papers by this author
  • Feng Sheng Hu

    1. Department of Plant Biology,
    2. Program in Ecology and Evolutionary Biology, and
    3. Department of Geology, University of Illinois, 265 Morrill Hall, Urbana, IL 61801, USA
    Search for more papers by this author

Author for correspondence:
David M. Nelson
Tel: +1 217 333 4376
Fax: +1 217 333 0508


  • • Holocene vegetational dynamics along the prairie–forest border of Minnesota were first documented in McAndrews’ classic work. Despite numerous subsequent paleo-studies, a number of questions remain unanswered about the vegetation history of the region. Here, pollen, stable-isotope, mineral, and charcoal data are described from three lakes near McAndrews’ sites. These data were compared with other paleoenvironmental records to reconstruct vegetation, aridity, and fire.
  • • The climate was relatively wet with increasing summer temperatures before ~8000 yr before present (BP). The rates of changes were asymmetric for the onset and termination of middle-Holocene aridity, with an abrupt increase at ~8000 yr BP and a gradual, but variable, decline from ~7800 to 4000 yr BP.
  • • Early-Holocene coniferous forests changed to mixed-grass prairie without an intervening period of tallgrass prairie or deciduous forest, whereas the retreat of prairie was characterized by transitions from mixed-grass to tallgrass prairie to deciduous forest and finally to coniferous forest. Within the middle Holocene, the composition and structures of grass-dominated vegetation varied both temporally and spatially.
  • • Fire primarily responded to changes in climate and fuel loads. Vegetation was more strongly influenced by climatic changes than by fire-regime shifts.


In the mid-1960s, John McAndrews pioneered the ‘transect’ concept in paleo-biogeography by analyzing pollen assemblages in sediments from four ponds across the prairie–forest border in northern Minnesota (Fig. 1). His results demonstrated that during the middle Holocene the prairie–forest border was ~100 km east of its location during the early and late Holocene (McAndrews, 1966, 1967). These results implied that the middle Holocene was warm and dry, although the timing of prairie expansion and contraction was equivocal because it was constrained by a total of only five 14C dates on bulk sediments. McAndrews’ work was followed by numerous paleoecological and paleoclimatic studies in Minnesota (e.g. Watts & Winter, 1966; Wright & Watts, 1969; Jacobson & Grimm, 1986; Brugam et al., 1988; Bartlein & Whitlock, 1993; Clark et al., 2001; Camill et al., 2003; Nelson et al., 2004; Wright et al., 2004) and surrounding regions (e.g. Wright, 1968; Baker et al., 1992; Dorale et al., 1992; Laird et al., 1996a; Nelson et al., 2006). Many of these studies involved pollen analysis, and a summary map of ‘prairie forbs’ confirmed that the pronounced movement of the prairie–forest border occurred during the middle Holocene (Webb et al., 1983). Recently, geochemical, sedimentological, and diatom analyses have also been used to examine Holocene climatic changes in the midwestern USA and the Great Plains (e.g. Valero-Garcés et al., 1997).

Figure 1.

Pre-European-settlement vegetational map of Minnesota and adjacent states (adapted from Küchler, 1964) with locations of some of the sites mentioned in the text. Sites used in Fig. 4 and Supplementary Material Fig. S1 are represented by squares (ML, Moon Lake; WOL, West Olaf Lake; DL, Deep Lake; SL, Steel Lake). Sites in Minnesota from McAndrews’ study (1966, 1967) are represented by circles, as are sites in Iowa (LO, Lake Okoboji; CL, Clear Lake), which are also shown in Fig. 5.

Despite the tremendous influence of McAndrews’ transect on numerous paleoenvironmental studies over the past four decades, the patterns and drivers of the pronounced shift of the prairie–forest border have not been evaluated in light of new paleorecords from northern Minnesota. One reason is that pollen records with secure chronologies are scarce from sites near McAndrews’ transect (Wright et al., 2004). In addition, pollen-independent climatic and fire reconstructions remain limited; such reconstructions are necessary to avoid the circularity of using pollen data as an indicator of environmental changes and also as a measure of the response of vegetation to environmental changes (Webb et al., 2003). Deciphering how climate and fire influence Holocene vegetational fluctuations can help to address longstanding issues concerning the controls of middle-Holocene prairie expansion. Although this expansion has often been attributed to warm and dry conditions (e.g. McAndrews, 1966, 1967; Bartlein et al., 1984; Shuman et al., 2002) other evidence suggests that such conditions may be less important. For example, paleorecords from Minnesota (e.g. Grimm, 1984) and Illinois (Nelson et al., 2006), as well as historical data from Illinois (e.g. Gleason, 1913), highlight fire as a key control on the relative abundance of woody versus herbaceous vegetation along the prairie–forest border. In addition to the relative roles of climate and fire, other important questions remain unanswered. For example, how do aridity and temperature interact to affect vegetational dynamics? What was the spatial and temporal variability of climate, vegetation, and fire within the middle Holocene?

To help elucidate Holocene changes in vegetation, climate, and fire regimes along the prairie–forest border, here we synthesize records with multiple indicators from the well-dated Holocene sediments of three lakes in northwestern Minnesota located near McAndrews’ original west–east transect. We include our previously published data (Hu et al., 1997; Hu et al., 1999; Nelson et al., 2004; Wright et al., 2004) as well as results of new analyses from the three study sites. We also compare our data with paleoecological and paleoclimatic records from adjacent areas and highlight recent studies that provide insights into Holocene environmental changes in the region.

Materials and Methods

Study sites

The three primary sites discussed in this paper span the major vegetational formations of Minnesota before European settlement, as delineated by the Marschner map based on land-survey records (Heinselman, 1974). Steel Lake (46°58′N, 94°41′W, 23 ha, 415 m above sea level (asl)) is in coniferous forest (dominated by Pinus spp.), with local stands of Populus tremuloides Michx. and Betula papyrifera Marsh. resulting from fires (Wright et al., 2004). Deep Lake (47°41′N, 95°23′W, 4 ha, 411 m asl) is just inside coniferous forest, but near a narrow band of oak woodland separating coniferous forest from tallgrass prairie. West Olaf Lake (46°37′N, 96°11′W, 58 ha, 393 m asl) is in a mix of tallgrass prairie and oak woodland. We compare data from these sites with pollen assemblages from Moon Lake (46°51′N, 98°09′W, 35 ha, 444 m asl), located in eastern North Dakota in mixed-grass prairie (Fig. 1).

The lakes (18–21 m deep) contain varved sediments (e.g. Tian et al., 2005), although varves are discontinuous during the middle Holocene. Mean January temperature varies little (−16 to −15°C) among our sites. Mean July temperature increases along a north–south gradient, from 19°C at Deep Lake to 21°C at West Olaf Lake. Mean annual precipitation and precipitation minus evaporation display strong east–west gradients across the region, ranging from 66 and −10 cm, respectively, near Steel Lake, to 48 and −30 cm, respectively, near Moon Lake (Fig. 1; Winter & Woo, 1990; Grimm, 2001). Corresponding to the east–west moisture gradient, magnesium:calcium (Mg:Ca) ratios in surface lake-water collected in July 2002 increase from 0.3 at Steel Lake to 0.7 at Deep Lake and 1.7 at West Olaf Lake. This pattern suggests that lakes in regions with drier conditions generally have greater Mg:Ca ratios in their waters.

Core sampling and analysis

Sediment cores were obtained from the deepest portion of each lake with a modified Livingstone piston sampler (Wright, 1991). Multiple cores from each lake were correlated on the basis of distinct lamination patterns. Sediment was sieved to recover macrofossils for accelerator mass spectrometry (AMS) 14C analysis, as described in Nelson et al. (2004). For Steel Lake, we used the chronology of Wright et al. (2004), which is based on 14C ages calibrated using CALIB 4.2 ( with the atmospheric 14C data set (Stuiver et al., 1998). For West Olaf and Deep lakes, 14C dates were converted to calibrated ages using CALIB 5.0.2 ( with the atmospheric 14C data set. Use of 14C ages calibrated with CALIB 5.0.2 instead of 4.2 has minimal influence on the Steel Lake chronology; the maximum difference is only 10 yr for any of the calibrated ages. Unless otherwise noted, all ages are reported as calibrated 14C yr before AD 1950 throughout this paper.

Methods for the analyses of pollen assemblages, sediment mineral composition, carbonate δ18O, and charcoal influx are described in Nelson et al. (2004). The mineral composition of bulk sediment was measured at each site, and we used the aragonite:calcite ratio to infer changes in aridity at West Olaf and Deep lakes. Low aragonite:calcite values suggest wet climatic conditions, and high values suggest dry conditions, because aragonite (a polymorph of calcite) forms preferentially over calcite as magnesium ions become concentrated in lake water when evaporation exceeds precipitation. δ18O was not used to infer climatic patterns throughout the Holocene at these two sites because aragonite, which is abundant in the middle-Holocene sediments, has an isotopic fractionation factor different from that of the coexisting calcite and thus would complicate climatic inferences from δ18O. By contrast, calcite is the only form of carbonate in the sediments of Steel Lake; thus we used carbonate δ18O to infer aridity at this site. Although δ18O may potentially indicate factors (e.g. precipitation seasonality and moisture source) other than aridity, previous studies (e.g. Nelson et al., 2004; Tian et al., 2006) have shown that, at Steel Lake, δ18O primarily indicates aridity variations within the middle and late Holocene.

Detrended correspondence analysis (DCA) was used to compare temporal variations in pollen assemblages among sites. Only taxa with ≥ 2% abundance in at least one sample from at least one site were included in the DCA. Pollen percentages of taxa with < 2% abundance were excluded from the pollen sum, and percentages of taxa with ≥ 2% abundance were recalculated from the raw pollen counts before the analysis. DCA analysis was performed using past (version 1.44; Hammer et al., 2001). Canonical correspondence analysis (conducted in PC-ORD 3.01), a constrained ordination technique (ter Braak, 1986), was used to relate temporal patterns in pollen assemblages to climate indicators and charcoal accumulation rates (environmental variables) at each site.


Radiocarbon chronologies

The chronological controls at our sites are based on a combination of new 14C dates (Supplementary Material Table S1) and 14C dates reported in previous studies (Hu et al., 1997, 1999; Nelson et al., 2004; Wright et al., 2004). The age–depth model for West Olaf Lake was constructed by fitting a third-order polynomial to 10 AMS 14C dates (Fig. 2). An age reversal occurs between the second- and third-youngest dates at this site. Both dates were included in the model because the reason for this discrepancy is unclear. The West Olaf Lake record begins ~9100 yr before present (BP), which is the age of the oldest sediments recovered from this site.

Figure 2.

Age–depth models. Depth is from sediment surface. 2σ ranges of calibrated ages are shown. The seven oldest dates from West Olaf Lake were presented in Nelson et al. (2004), and the Steel Lake chronology was first presented in Wright et al. (2004). cal. ka BP, thousands of calibrated 14C years before present.

The age–depth model for Deep Lake was constructed from seven AMS 14C dates that were fit with a second-order polynomial (Fig. 2). We only present data spanning 10 000 to 3700 yr BP because our cores were initially taken for a study focusing on the late Glacial and early Holocene (Hu et al., 1997, 1999), and the late-Holocene sections were not properly correlated with one another on the basis of lamination patterns.

The age–depth model for Steel Lake was constructed using 26 AMS 14C dates with a locally weighted polynomial regression (Fig. 2; Wright et al., 2004). The Steel Lake record spans the entire Holocene.

Aridity fluctuation

Our climatic reconstruction focuses on spatial and temporal variations of aridity, which encompass the effects of temperature on evapotranspiration and are a key control on the distribution of prairie and forest (Changnon et al., 2002). At West Olaf Lake, the aragonite:calcite ratio is low, fluctuating around a mean of 0.34, before ~8000 yr BP (Fig. 3), which suggests a generally wet climate with drier episodes. At Deep Lake aragonite first appears ~9300 yr BP, but the aragonite:calcite ratio remains negligible (~0.02) until ~8000 yr BP. These results suggest that the regional climate became slightly drier c. ~9300 yr BP but remained wet compared with just after 8000 yr BP. This interpretation is supported by the stratigraphic patterns of detrital minerals and δ18O at Steel Lake. The abundance of quartz and feldspars, detrital minerals that were probably transported to the lake through eolian erosion of dry soils (e.g. Dean et al., 1996), increases slightly after ~9400 yr BP but remains low before ~8000 yr BP. Similarly, δ18O values are relatively low before ~8000 yr BP (Fig. 3).

Figure 3.

Oxygen-isotope and mineral records used to infer aridity changes at West Olaf, Deep, and Steel lakes. cal. ka BP, thousands of calibrated 14C years before present.

A sharp increase in aridity occurred at or just after 8000 yr BP, as suggested by the abrupt increase in each indicator of aridity to peak or near-peak values (Fig. 3). This change is also recorded at other sites in central and southeastern Minnesota (e.g. Brugam et al., 1988; Keen & Shane, 1990; Shuman et al., 2002; Camill et al., 2003), as well as eastern North Dakota (Laird et al., 1996a). The regional increase in aridity may have resulted from diminishing influence of the ice sheet on the regional atmospheric circulation (Shuman et al., 2002), possibly through enhanced atmospheric subsistence (e.g. Diffenbaugh et al., 2006) or increased eastward penetration of dry Pacific air (e.g. Booth et al., 2006). Following this aridity peak, each indicator fluctuates along generally declining trends, suggesting effective-moisture variation superimposed on a long-term increase. By ~3500 yr BP the regional climate was again relatively wet in northwestern Minnesota, as inferred from the near absence of aragonite at West Olaf Lake and generally low values of detrital minerals and δ18O at Steel Lake.

Together these results indicate that the middle Holocene was the driest period since the last deglaciation, as suggested by McAndrews’ pollen data (McAndrews, 1966, 1967) and verified by subsequent studies using pollen-independent approaches (e.g. Brugam et al., 1988; Laird et al., 1996a). Within the middle Holocene, the intensity of aridity varied temporally and spatially. For example, at West Olaf Lake episodes of aridity occurred between ~6500 and ~5800 yr BP and between ~5300 and ~4900 yr BP, in addition to the aridity peak just after 8000 yr BP that indicates the driest conditions of the Holocene (Fig. 3). Aridity generally decreased from west to east across the study region within the middle Holocene, similar to the spatial pattern at present (Grimm, 2001). For example, the aragonite:calcite ratio was overall higher at West Olaf Lake than at Deep Lake during the early and middle Holocene, and aragonite was absent from the Steel Lake sediments.

Vegetational change

Pollen assemblages indicate that, at Deep and Steel lakes, herbaceous taxa (including Ambrosia, Artemisia, and Poaceae) and Quercus began replacing Pinus banksiana/resinosa-dominated forests ~9400 yr BP (Fig 4. and Supplementary Material Fig. S1; Wright et al., 2004). At Deep Lake, herbaceous taxa increased sharply between 9400 and ~9100 yr BP and then fluctuated slightly until ~8100 yr BP. By contrast, the expansion of herbaceous taxa was gradual at Steel Lake, extending from ~9400 to 8100 yr BP (e.g. Fig. 4 and Supplementary Material Fig. S1; Wright et al., 2004). DCA results confirm these patterns. At 9000 yr BP, Deep Lake has a more negative DCA1 score than Steel Lake (Fig. 5), indicating a greater abundance, and earlier expansion, of herbaceous taxa at Deep Lake. Vegetation at these sites between ~9400 and 8000 yr BP trended toward mixed-grass prairie communities of the Dakotas before European settlement, but with a generally greater abundance of woody species than in the Dakotas (Fig. 5). We cannot pinpoint the timing of prairie development at West Olaf Lake because the recovered sediment core has a basal age of ~9100 yr BP. However, it is likely that herbaceous taxa were already established near West Olaf Lake before their expansion at Deep Lake, as suggested by the greater abundance of nonarboreal pollen at West Olaf Lake at 9000 yr BP (Fig. 4a). Vegetation near West Olaf Lake 9000 yr BP was similar to that at 10 000 yr BP at Moon Lake (Fig. 1) where mixed-grass prairie, characterized by abundant Artemisia and a moderate abundance of Poaceae (Hoyt, 2000; Grimm, 2001), was already established (Fig. 5; Laird et al., 1996a). Together these results indicate that the expansion of prairie from eastern North Dakota to Minnesota was time-transgressive (Wright et al., 2004).

Figure 4.

Percentages of the key pollen types. (a) Nonarboreal pollen (NAP); (b) Ambrosia; (c) Poaceae; (d) Artemisia; (e) Quercus; (f) Pinus. ML, Moon Lake; WOL, West Olaf Lake; DL, Deep Lake; SL, Steel Lake. The order of site names is reversed in (e) and (f). The pollen sum includes all terrestrial pollen types. Three-sample moving averages are used for all plots. Pollen data from Moon Lake were obtained from the North American Pollen Database. More detailed pollen diagrams are available in Supplementary Material Fig. S1. cal. ka BP, thousands of calibrated 14C years before present.

Figure 5.

Detrended correspondence analysis (DCA) results of pollen assemblages from Steel, Deep, West Olaf, and Moon lakes. The data points occur every 500 yr and represent average DCA values within a 500-yr window of each data point. The data points are labeled in 1000-yr intervals (e.g. a label of 8 means 8000 yr before present (BP)). Also shown are pre-European-settlement pollen assemblages from Devils (North Dakota), Rice (North Dakota), and Cottonwood (South Dakota) lakes, where the vegetation was mixed-grass prairie, and Lake Okoboji (Iowa) and Clear Lake (Iowa), where the vegetation was tallgrass prairie. Pollen data from Devils, Rice, and Cottonwood lakes were obtained from the North American Pollen Database. Abbreviations of five abundant woody and herbaceous pollen types are shown to substantiate the interpretation that DCA1 primarily separates woody from herbaceous taxa (Alnus incana (Al), Ambrosia (Am), Artemisia (Ar), Asteraceae subf. Asteroideae undiff. (As), Betula (Be), Cyperaceae (Cy), Poaceae (Po), Pinus sum (Pi), Ulmus (Ul), and Quercus (Qu)). DCA2 distinguishes tallgrass (higher DCA2 values; West Olaf, Okoboji, and Clear) from mixed-grass (lower DCA2 values; Moon, Devils, Rice, Cottonwood) prairie (for samples with low DCA1 values).

As pine percentages decreased following the initial expansion of herbaceous taxa into northern Minnesota, Ambrosia and Artemisia increased sharply to their maximum abundance just after 8000 yr BP at all three sites (Fig. 4 and Supplementary Material Fig. S1). However, Ambrosia was much more abundant and variable at West Olaf Lake than at Deep and Steel lakes. Greater abundance of Ambrosia pollen suggests more extensive soil disturbance to the west, because the Ambrosia species are upland weeds that require bare ground for establishment (Grimm, 2001). Ambrosia also requires adequate growing-season moisture, yet our geochemical and mineral indicators suggest that it was generally abundant during periods of aridity (Fig. 3). This difference may be reconciled by high intra-annual variability, with sufficient moisture during the growing season and severe moisture deficits outside of the growing season, the latter of which created arid and disturbed soils to favor Ambrosia when growing-season moisture became available (Grimm, 2001).

Following the peak abundance of Ambrosia and Artemisia, pollen assemblages indicate that Quercus expanded rapidly in abundance on upland areas at all three sites, as inferred from the increase of its pollen to > 30%. As aridity generally declined from ~7800 to 4000 yr BP, the abundance of Ambrosia and Artemisia decreased and that of Poaceae increased at West Olaf Lake, although a peak in Ambrosia at around 6000 yr BP is evident at West Olaf Lake, as well as other sites (Fig. 4 and Supplementary Material Fig. S1). These peaks coincide with high and variable values of aragonite:calcite at West Olaf Lake and high values of δ18O and detrital minerals at Steel Lake, which suggests a reversal to elevated aridity and increased soil disturbance (Fig. 3).

At Deep and Steel lakes, a long-term decline in the percentages of Artemisia pollen, along with steady abundance of Poaceae and Quercus pollen, occurred within the middle Holocene (Fig. 4 and Supplementary Material Fig. S1). These changes suggest a transition to tallgrass prairie and oak woodland communities similar to those that occurred near West Olaf Lake over the last ~3500 yr. This inference is confirmed by DCA results (Fig. 5). For example, pollen assemblages at around 7500–7000 yr BP at Deep and Steel lakes were similar to those at sites in tallgrass prairie before European settlement (Fig. 5). The DCA results also suggest that, near West Olaf Lake, mixed-grass prairie transitioned to tallgrass prairie ~7000 yr BP and returned to mixed-grass prairie ~5000 yr BP, probably because of a period of relatively wet conditions centered on ~7000 yr BP and relatively dry conditions between ~5300 and ~4900 yr BP. Thus, grass-dominated communities prevailed over the regional landscape during the Holocene, but the composition of these communities varied temporally and spatially.

Between ~4000 and ~3500 yr BP, peaks in the abundance of Ambrosia pollen occur at all sites (Fig. 4a and Supplementary Material Fig. S1), probably in response to elevated aridity across the region. This interpretation is supported by higher δ18O values at Steel Lake (Fig. 3) and by evidence of dune activation in Nebraska around 3800 yr BP (Miao et al., 2007). By ~3400 yr BP pine forests (first Pinus strobus and then Pinus banksiana/resinosa after ~1900 yr BP) were again prevalent in the Steel Lake region (Fig. 4f and Supplementary Material Fig. S1) following a short period during which deciduous taxa (i.e. Betula and Ostrya/Carpinus) were abundant (Wright et al., 2004). At West Olaf Lake, herbaceous taxa gradually decreased in abundance as woodland, characterized by Quercus, Betula, and Ostrya/Carpinus, expanded between ~4000 and 2000 yr BP, but herbaceous taxa expanded slightly during the last 1000 yr (Figs 4, 5, and Supplementary Material Fig. S1).

Fire-regime variation

Charcoal accumulation rates (CHARs) in prairie–forest border communities offer qualitative information on fire occurrence as a function of fuel availability, fuel moisture, vegetational type, and ignition (Brown et al., 2005; Nelson et al., 2006). At Deep and Steel lakes, CHARs increased from an average of 1.7 particles cm−2 yr−1 before 8000 yr BP to ~2.8 particles cm−2 yr−1 at Deep Lake and ~3.1 particles cm−2 yr−1 at Steel Lake between 8000 and 4000 yr BP (Fig. 6), probably because fuels were drier and flammable types of vegetation more abundant. CHARs remained generally low during the late Holocene at Steel Lake, although they were slightly higher after 1500 yr BP, potentially because of the establishment of flammable jack pine (Pinus banksiana) forests (Wright et al., 2004).

Figure 6.

Charcoal accumulation rates (CHARs) from West Olaf, Deep, and Steel Lakes. cal. ka BP, thousands of calibrated 14C years before present.

At West Olaf Lake the Holocene average of CHARs (~25.3 particles cm−2 yr−1) is much higher than at Deep and Steel lakes (Fig. 6), which suggests higher fire frequencies and/or greater areas burned at West Olaf Lake. Within the middle Holocene, CHARs generally increased with decreased aridity at West Olaf Lake. For example, CHARs were generally higher at West Olaf Lake between 5700 and 4000 yr BP (29.7 particles cm−2 yr−1, on average) when the climate was wetter than between 8000 and 5700 yr BP (17.4 particles cm−2 yr−1, on average). Higher frequency variations in CHARs also support this interpretation. For example, CHARs were relatively low during arid periods around 5800 and 3800 yr BP, probably because fuel loads were low as indicated by peaks in Ambrosia pollen abundance and low values of Poaceae (Fig. 4). These patterns suggest that fire became more important when moisture did not limit biomass production and fuel loads. Particularly high CHARs from ~3500 to 2000 yr BP, a period of relatively wet conditions, likely also reflect greater amounts of biomass on the landscape. Woody species increased in abundance during this time, which suggests that fire did not prevent their expansion. However, once fuel loads reached certain thresholds, fire occurrence appears to have been primarily limited by moist conditions that reduced biomass flammability. In particular, increased moisture after 2000 yr BP (e.g. Tian et al., 2006) resulted in a decrease in fire occurrence at West Olaf Lake, as inferred from decreased CHARs.

Overall, CHARs from our sites suggest that Holocene prairie expansion and contraction were not driven by fire-regime shifts in northern Minnesota. For example, increasing CHARs did not accompany the expansion of prairie at Steel and Deep lakes, and decreasing CHARs did not accompany the retreat of prairie. This inference is supported by results of canonical correspondence analysis (Fig. 7). At Steel and Deep lakes the environmental arrows representing climate proxies point along axis 1 (which explains 32.8 and 31.6% of variation in pollen assemblages at these lakes, respectively) and in the direction of most herbaceous taxa that dominate middle-Holocene pollen assemblages. The environmental arrows representing fire regimes (CHAR) are shorter than those representing climate proxies at Steel Lake and are associated with axis 2 at Deep Lake. These patterns further suggest stronger linkages of plant communities to aridity than to fire. At West Olaf Lake the arrows representing aragonite:calcite and CHAR are of similar length, with Poaceae near CHAR and Artemisia near aragonite:calcite. Because Poaceae is less drought-tolerant than Artemisia (Grimm, 2001), these patterns support the interpretation that fires were more important under wetter climatic conditions and higher fuel availability.

Figure 7.

Results of canonical correspondence analysis of pollen, climate, and fire indicators from West Olaf, Deep, and Steel lakes. The percentages of the nine most abundant pollen types were used in the analysis: Pinus (Pi), Ulmus (Ul), Betula (Be), Ostrya/Carpinus (Os), Quercus (Qu), Poaceae (Po), Ambrosia (Am), Artemisia (Ar), and Chenopodiaceae (Ch). Ar:cc, aragonite:calcite; 18O, δ18O; Qtz, quartz + feldspars; CHAR, charcoal accumulation rate. The percentages of variance in pollen data explained by the axes are noted in parentheses. x, pollen samples > 8000 yr before present (BP); circles, pollen samples between 8000 and 4000 yr BP; squares, pollen samples < 4000 yr BP. 1σ error bars are shown. The scale of axis 1 at Deep Lake is reversed.


Our multi-proxy records from northwestern Minnesota help to clarify the patterns and drivers of Holocene vegetational change originally observed by McAndrews. These results are bolstered by a number of other studies in the region over the ~40 yr since McAndrews’ initial work. One pioneering study within McAndrews’ transect that used multiple indicators is that of Elk Lake. Laminated sediments from this site were analyzed for geochemical, mineral, and biological indicators to infer Holocene environmental changes (Bartlein & Whitlock, 1993; Dean et al., 1996). Pollen assemblages indicate that early-Holocene Pinus forests were succeeded by Quercus and herbaceous taxa ~8500 varve yr BP. A brief period of dominance by deciduous taxa from ~4000 to ~3000 varve yr BP was followed by plant communities dominated by Pinus (Bartlein & Whitlock, 1993). Mineral analysis reveals that the abundance of quartz increases gradually from 10 000 to ~7500 varve yr BP, reaches peak values from ~7500 to ~5500 varve yr BP, and remains generally low after ~5500 varve yr BP (Dean et al., 1996). Consistent with McAndrews’ interpretations, these results from Elk Lake suggest that aridity was generally high during the middle Holocene when herbaceous taxa were most abundant. However, the chronology of Elk Lake is up to ~1000 yr too young (Wright et al., 2004), complicating detailed comparisons with other records.

Shifts in the prairie–forest border appear to be well constrained at Deming Lake (Clark et al., 2001), a site between Deep and Steel lakes and ~2 km southeast of Elk Lake. The vegetational history at this site is quite similar to that documented at Deep and Steel lakes. For example, at each site Ambrosia, Artemisia, and Poaceae were abundant during the early part of the middle Holocene, whereas Poaceae and Quercus were dominant during the later portion of the middle Holocene. Clark et al. (2001) used δ13C analysis of charcoal particles to reconstruct changes in the relative abundance of C3 and C4 plants at Deming Lake. This δ13C record, along with that of Nelson et al. (2004) from Steel Lake, revealed that C4 plants were more abundant under drier conditions and C3 woody species more abundant under wetter conditions. By contrast, at West Olaf Lake, where middle-Holocene drought severity was greater and soil disturbance more extensive, C4-plant abundance was negatively correlated with aragonite:calcite, suggesting that such conditions favored C3 weedy species over C4 grasses (Nelson et al., 2004).

Our results uncovered aspects of the vegetational and climatic history in northern Minnesota that have not been extensively discussed in the literature. For example, prairie taxa began expanding into the Deep Lake and Steel Lake areas ~9400 yr BP, well before the pronounced change to drier conditions ~8000 yr BP on the basis of geochemical and mineral data. Prairie development may have initiated with a slight increase in aridity that began at ~9400 yr BP. Alternatively, the expansion of herbaceous taxa may have resulted from the direct effects of increased summer temperatures in the region from 10 000 to ~8000 yr BP (Nordt et al., 2007), when the regional climate remained relatively wet as inferred from our data (Fig. 3). Increased temperatures may have caused the northward retreat of P. banksiana or P. resinosa, which dominated the landscape around Deep and Steel lakes before ~9400 yr BP (Fig. 4; Jacobson, 1979; Wright et al., 2004). This interpretation is consistent with the fact that today the southern limit of P. banksiana in North America extends no farther than central Minnesota, Wisconsin, and Michigan, because this species cannot regenerate under high summer temperatures (Schoenike, 1976; Wright, 1992).

Our results also offer evidence for complex temporal and spatial variations of Holocene climate and vegetation near McAndrews’ transect. Early-Holocene coniferous forests at Deep and Steel lakes changed toward mixed-grass prairie without an intervening period of tallgrass prairie or deciduous forest. By contrast, the retreat of prairie at these sites was characterized by transitions from mixed-grass to tallgrass prairie to deciduous forest, and finally to coniferous forest. These patterns can be attributed to the severity of aridity and its rates of change. The sharp increase to maximum aridity at the beginning of the middle Holocene caused a rapid expansion of herbaceous taxa, whereas the gradual decline caused prairie to retreat westward over several thousand years. Spatial variation in climate and vegetation is evidenced by nonuniform middle-Holocene plant communities across the area. For example, while more mesic tallgrass prairie and woodland communities generally expanded at Deep and Steel lakes from ~7000 to ~5000 yr BP, the vegetation at West Olaf Lake reverted toward more xeric mixed-grass prairie.

McAndrews’ work and subsequent studies near his transect also instigated climatic and vegetational reconstructions in other nearby regions. Pollen data from Roberts Creek (Baker et al., 1992) and stalagmite isotopic data from Coldwater Cave (Dorale et al., 1992) in northeastern Iowa suggest that prairie expansion did not occur until after 6000 yr BP. This delayed expansion was attributed to a steep moisture gradient between southeastern Minnesota and northeastern Iowa. Recent work at sites in Illinois confirmed a more complex scenario there (Webb et al., 1983): an early pulse of aridity and prairie development around 8500 yr BP was interrupted by mesic conditions and prairie contraction until the marked expansion of prairie c. 6000 yr BP (Nelson et al., 2006). These vegetational changes differed from those near the prairie–forest border in northern Minnesota and probably resulted from distinct climatic conditions and fire regimes.

Charcoal influx data from Deming Lake support the interpretation that fire regimes primarily responded to changes in climate and fuel loads near the prairie–forest border in northern Minnesota. These results are consistent with modern ecological studies in Minnesota indicating that climate exerted a strong control on the distribution of prairie and forest (e.g. Faber-Langendoen & Tester, 1993). By contrast, charcoal and pollen data from Illinois suggest that fire played an important role in the development and maintenance of prairie communities (Nelson et al., 2006), which concurs with paleorecords from southeastern Minnesota (e.g. Grimm, 1984) and historical records from Illinois (e.g. Gleason, 1913). These differences in the relative effects of climate and fire on vegetational distributions may be related to the degree of moisture limitation. For example, biomass production in Illinois was probably high enough to prevent fuel limitation for fire occurrence even during the driest periods of the middle Holocene; by contrast, aridity limited biomass production and fuel loads at West Olaf Lake in northwestern Minnesota. Biomass production at Deep and Steel lakes was also probably high enough to prevent fuel limitation and yet fire did not appear to strongly influence prairie expansion and contraction at these sites, potentially because conditions were overall drier in Minnesota than Illinois. Another possible explanation for the greater importance of fire in Illinois is more frequent anthropogenic ignitions, but testing this idea remains a challenge because of difficulty in distinguishing anthropogenic from natural burning.

Beyond the immediate scope of McAndrews’ work, the needs to understand anthropogenic climatic change have motivated a number of recent high-resolution studies of late-Holocene aridity variations. At Moon Lake, decadal-scale analyses of diatom assemblages revealed short-term moisture variations (Laird et al., 1996b) that had been unnoticed by most previous studies with lower sampling resolution. Subsequent studies from Minnesota and the Great Plains further documented the spatial extent of late-Holocene aridity variations (e.g. Yu & Ito, 1999; Tian et al., 2006) and their ecological impacts (Brown et al., 2005). A significant discovery from several of these studies is that drought events of greater severity and duration than the 1930s Dust Bowl occurred throughout the late Holocene. These droughts may be related to changes in atmospheric circulation patterns and solar forcing (Laird et al., 1996b; Yu & Ito, 1999; Tian et al., 2006), but the mechanisms remain poorly understood. Although these late-Holocene droughts were relatively mild compared with the prolonged middle-Holocene aridity, climatic boundary conditions (e.g. insolation and ice sheet area) of the late Holocene were more similar to those of today, and thus understanding the patterns of these droughts is particularly important for distinguishing natural and anthropogenic climatic changes. McAndrews’ pioneering work provided a broad framework of Holocene environmental changes for many recent and ongoing high-resolution studies. Therefore McAndrews’ legacy continues to influence our understanding of climatic and vegetational changes in the North American midcontinent.


We thank H. E. Wright, J. Tian, A. Henderson, M. Chipman, G. Clarke, D. Devotta, K. Wolfe, C. Barrett, and three anonymous reviewers for helpful comments on the manuscript. I. Stefanova counted the Steel Lake and most of the Deep Lake pollen samples. J. Tian conducted the isotopic and XRD analyses for Steel Lake. We also thank contributors to the North American Pollen Database. This work was supported by a Packard Fellowship in Science and Engineering (FSH) and NSF ATM-0318404 (FSH). DMN was supported by a fellowship from the Institute for Genomic Biology.