The Little Ice Age signature in a 700-year high-resolution chironomid record of summer temperatures in the Central Eastern Alps

Despite the fact that the Little Ice Age (LIA) is well documented for the European Alps, substantial uncertainties concerning the regional spatio-temporal patterns of temperature changes associated with the LIA still exist, especially for their eastern sector. Here we present a high-resolution (4–10 years) 700-year long mean July air temperature reconstruction based on subfossil chironomid assemblages from a remote lake in the Austrian Eastern Alps to gain further insights into the LIA climatic deterioration in the region. The record provides evidence for a prolonged period of predominantly cooler conditions during AD 1530–1920, broadly equivalent to the climatically defined LIA in Europe. The main LIA phase appears to have consisted of two cold time intervals divided by slightly warmer episodes in the second half of the 1600s. The most severe cooling occurred during the eighteenth century. The LIA temperature minimum about 1.5 °C below the long-term mean recorded in the mid-1780 s coincides with the strongest volcanic signal found in the Greenland ice cores over the past 700 years and may be, at least in part, a manifestation of cooling that followed the long-lasting AD 1783–1784 Laki eruption. A continuous warming trend is evident since ca AD 1890 (1.1 °C in 120 years). The chironomid-inferred temperatures show a clear correlation with the instrumental data and reveal a close agreement with paleotemperature evidence from regional high-elevation tree-ring chronologies. A considerable amount of the variability in the temperature record may be linked to changes in the North Atlantic Oscillation. Electronic supplementary material The online version of this article (10.1007/s00382-018-4555-y) contains supplementary material, which is available to authorized users.


SI 1. Sediment dating and age-depth modeling
The chronology for the MUT sediment sequence was constrained by sixteen 210 Pb activity determinations (down to 6.7 cm depth) and three accelerator mass spectrometry (AMS) radiocarbon dates derived from terrestrial plant macrofossils for the deeper section of the core (Ilyashuk et al. 2015). Lead-210 activity was analyzed by extracting the grand-daughter 210 Po and counting it in an alpha spectrometer (Flynn, 1968) in the University of Gdańsk, Poland. The 210 Pb-based ages of recent sediments were calculated by applying the constant rate of supply (CRS) model to the unsupported 210 Pb inventory (Appleby and Oldfield 1978). AMS radiocarbon dating was carried out at the Poznan Radiocarbon Laboratory (Poland) and the Beta Analytic Radiocarbon Dating Laboratory (Miami, FL, USA). Given that large chronological uncertainties can arise under translation of 14 C ages into calendar years if 14 C dates are calibrated individually (Blaauw et al. 2011), all dates were calibrated simultaneously, taking into account their stratigraphic order. Stratigraphically constrained 14 C calibration was performed on-line using the OxCal Bayesian software (Bronk Ramsey 2009, 2010) by applying the Northern Hemisphere terrestrial IntCal13 calibration curve (Reimer et al. 2013). The age-depth relationships (Fig. S1) were established by means of Poisson mediated deposition model (P Sequence) that supposes the essentially random nature of the deposition process (Bronk Ramsey 2008). A reasonable value for the k parameter, which is defined by the number of the deposition events per unit depth, was chosen to be of 5, taking into account fine-grained sediment deposition in the lake and fine sampling resolution (0.22-cm slicing) of the sequence. Ages were interpolated to each sample depth based a secondorder polynomial regression between median values of posterior probabilities at each chronological tie point, assuming that polynomial fitting will result in a smoother age-depth curve, avoid large shifts in modeled sediment accumulation, compared to linear interpolation, and thereby provide a better approximation of the true sediment accumulation history in the lake. As with all age-depth models, each interpolated value represents one of a range of possible values for its depth and our interpretations must be considered within this context. According to the age-depth model, an average temporal resolution is ca. 4.8 years per sample in the sediment sequence. Every sample was analyzed at the top 11.5 cm of the core (resulting in the temporal resolution of the core interval of ~4-5 years), and every other one at the lower part (resulting in the temporal resolution of the interval of ~10 years).

Fig. S1
The Bayesian age-depth model (P Sequence, k = 5) for the MUT sediment drawn through the 210 Pb-CRS ages and posterior distributions of calibrated 14 C dates. The uncertainty envelope (in blue) represents the 95.4% confidence intervals of the age model. The distributions of individually calibrated dates are shown as grey histograms. The complete age-depth curve (thick line) employed in the present study is constructed by a second-order polynomial fit between median points of the modeled ages

SI 2. Chironomid analysis
Subfossil chironomid analysis was performed on 100 sediment samples of 0.22 cm thickness, which were processed following the standard procedures for subfossil midges given in Walker (2013). Subsamples were deflocculated in 5% KOH for 30 minutes and then sieved with a 100-μm mesh. Chironomid larval head capsules were hand-picked out from the remnant in a Bogorov counting tray under a stereomicroscope at 20-40× magnification and dehydrated in 100% ethanol. Afterwards head capsules were permanently mounted ventral side up on microscope slides in Euparal® (Carl Roth Gmbh, Karlsruhe, Germany) mounting medium for identification. Chironomids were identified under a compound microscope at 200-400× magnification.
Chironomid taxonomy followed Brooks et al. (2007) and Andersen et al. (2013). A minimum of 100 chironomid head capsules were counted and identified in each sample. The chironomid stratigraphy for MUT showing the relative abundance (%) of all taxa present in the lake during the past 700 years and established zonation is given in the Fig. S2.

SI 3. Segmented regression analysis
Providing a statistical basis for spotting trends in the chironomid-inferred reconstructed temperature record, breakpoint years, with confidence intervals (CIs), and separated periods of significant trend change were determined by applying a segmented regression approach with the SegReg software (Oosterbaan 2011). In the 700-year long record, the analysis identified a significant breakpoint at AD 1783 (95% CI [1754, 1811]) ( Fig. S3a). Taking into account the CI of the point and chronological uncertainties of the record, the year AD 1800 is further used as a trend changing point to estimate a multi-centennial (AD 1300-1800) cooling trend in the record (Fig. S3b). A separate segmented regression for the subsequent time interval (AD 1800(AD -2010 revealed a breakpoint at AD 1894 (95% CI [1850,1938]), where the horizontal stretch (no trend) is followed by a sloping line with a significant regression coefficient (Fig. S3c).