Paleoenvironments from robust loess stratigraphy using high-resolution color and grain-size data of the last glacial Krems-Wachtberg record (NE Austria) Quaternary Science Reviews

The complex interplay of dust sedimentation, pedogenesis, and erosion/reworking in the formation of loess-paleosol sequences (LPS) challenges paleoenvironmental proxies. Here we show that color and grain size are essential parameters characterizing loess pro ﬁ les and support robust stratigraphies as a basis for reconstructions in the context of local geo-ecological and large-scale paleoclimatic evolution. Detailed paleoenvironmental records from the period since the arrival of anatomically modern humans to the last glacial maximum are scarce in the Alpine surroundings. The c. 7.5 m thick LPS Krems-Wachtberg, NE Austria, known for its well-preserved Upper Paleolithic context at a depth of 5.5 m, formed between 40 and 20 ka BP by quasi-continuous dust-sedimentation, interrupted by phases of incipient pedogenesis and local reworking. The new KW2015 composite is based on three sections studied and sampled at 2.5 cm resolution. Color and grain size data support a robust stratigraphy for reconstructions of the pedosedimentary evolution. The marked transition from oxidized to reduced paleosols of KW2015 around 34 e 35 ka corresponds to the Middle-to Upper Pleniglacial transition as part of a general cooling trend from marine isotope stage (MIS) 3 to 2, intensely modulated by millennial- scale climatic ﬂ uctuations as recorded in the Greenland ice core data. The distinct response of KW2015 to these trends highlights that reconstructing LPS evolution based on a robust stratigraphy is a prerequisite to paleoenvironmental proxy interpretation.

Comparably few studies provide detailed stratigraphic information and evaluate the formation processes of LPS, which is prerequisite to interpreting their parameter variations (Antoine et al., 2013;Meszner et al., 2013;Sprafke, 2016;Meyer-Heintze et al., 2018). Pedogenic processes overprint loess during phases of absent or low sedimentation and reflect ecological conditions. The tight chronology for the last glacial reference LPS Nussloch (Moine et al., 2017) demonstrates strong links between millennial climatic events and paleosol types/intensities in Central European LPS (Rousseau et al., 2017a). However, interpretations of LPS can be complicated by reworking and erosion events Sprafke, 2016).
Depending on the scientific background of researchers, stratigraphic descriptions of LPS strongly differ in detail and unit designation; the unified nomenclature set by the former INQUA subcommission of loess stratigraphy in the 1960s (Fink, 1965(Fink, , 1969 has been abandoned. Soil horizon designations appear reasonable for loess stratigraphy as pedogenic pigmentation causes most visible variations in LPS (Bronger, 1976(Bronger, , 2008Schirmer, 2000Schirmer, , 2016Sprafke, 2016). Spectrophotometers quantify color variations of LPS and enhance the robustness and detail of stratigraphic logs derived from field descriptions (Sprafke, 2016). Here, we argue that a combination of this approach with detailed granulometry (Vandenberghe, 2013; captures the main forming processes of LPS in the interplay of dust deposition, pedogenesis and local reworking, bridging the gap to proxy interpretation. In the Alps and their forelands only few paleoclimate records are available for the timespan since the arrival of anatomically modern humans (AMH; c. 43.5 ka) and the end of the Last Glacial Maximum (LGM; c. 20 ka) (Heiri et al., 2014;Nigst et al., 2014;Mayr et al., 2019;Stojakowits et al., in press). During this period, the c. 7.5 m thick LPS Krems-Wachtberg (NE Austria) formed by more or less continuous dust accumulation, interrupted by phases of incipient pedogenesis and reworking Einw€ ogerer et al., 2014;Lomax et al., 2014). Different parts of the section were exposed during archeological excavations that lasted from 2005 until 2015 (H€ andel et al., 2014; H€ andel, 2017). In total seven archeological find horizons (AH) are reported, of which the Gravettian (Mid Upper Paleolithic) AH 4 at c. 5.5 m depth provided the vast majority of findings, including part of an occupation layer with infant burials and hearths dated to c. 31 ka cal BP (Einw€ ogerer et al., 2006;Brandl et al., 2014;Fladerer et al., 2014;Simon et al., 2014).
The LPS Krems-Wachtberg is located in a (paleo-)geographical bridge position from Central-to SE-Europe ( Fig. 1) and provides high-resolution paleoenvironmental information for the transition from the milder Marine Isotope Stage (MIS) 3 to MIS 2, corresponding roughly to the Middle-(MPG) to Upper Pleniglacial (UPG; Moine et al., 2017;Zens et al., 2018). High-resolution MS data (Hambach, 2010;Groza et al., 2019) and a horizon-wise study of the upper 6 m  did not unravel the complex evolution of the complete section. Before the Krems-Wachtberg excavation was backfilled for house construction in 2015, the entire LPS was accessible in three complementary profiles (KW-A, -B, eC). This paper presents a detailed revised stratigraphy of the LPS Krems-Wachtberg (KW2015 composite) that combines field documentation with high-resolution color and GS data. The reconstruction of the local interplay of dust deposition, pedogenesis and reworking provides semi-quantitative paleoenvironmental information, which is discussed in the context of the Central European MPG/UPG paleoclimatic evolution. This study highlights that complex terrestrial archives like LPS require a detailed stratigraphic analysis involving the reconstruction of their forming processes before parameter variations along vertical sections are interpreted as paleoclimatic and -environmental proxies.

Study site
For more than 100 years, the thick loess cover at Krems a. d. Donau (NE Austria) has attracted both loess researchers and archeologists (Penck, 1903;Fink, 1976;Fink and Kukla, 1977;Einw€ ogerer et al., 2014;Sprafke, 2016;H€ andel, 2017). The c. 7.5 m thick LPS Krems-Wachtberg is located at 250 m a.s.l. on a slightly SE-inclined plateau 50 m above present day Danube level on a spur created by the incision of the Danube and Krems River into crystalline rocks of the Bohemian Massif ( Fig. 1; Fuchs and Grill, 1984). Weathered Neogene to Pleistocene fluvial sediments are overlain by loess packages of up to 37 m thickness, depending on paleotopography Sprafke, 2016). At Krems-Schießst€ atte loess reaches back to the Early Pleistocene Olduvai magnetic chron (Fink and Kukla, 1977). The SE tip of the spur is a strategic panorama position used since the Early Upper Paleolithic, as witnessed by a number of archeological find layers in the youngest loess. The Wachtberg area has recently gained attraction as residential area, resulting in the abandonment of old vineyards and numerous prospections, core drillings, rescue excavations, and two long term archeological campaigns at Krems-Hundsteig (2000e2002) and Krems-Wachtberg (2005e2015) (Neugebauer-Maresch, 2008Einw€ ogerer et al., 2014;H€ andel, 2017). H€ andel et al. (2014) summarize the history of the Krems-Wachtberg excavation and give an overview of the main loess profiles (Fig. 2), which are all characterized by a similar stratigraphy with very little lateral variation. Several local marker horizons in the weakly differentiated loess sediments have supported a stratigraphic framework for the whole excavation, with layers slightly dipping southeast (H€ andel et al., 2014). 39 macroscopically identifiable geological horizons (GH) were labelled in 2005 from the topsoil to the gravel/rock at the bottom (GH 39). GH 5 to GH 38 are yellowish to slightly brownish and partly greyish/grey loess sediments. Some units with a higher amount of coarse material from upslope (weathered crystalline rock and fluvial deposits) and partly sharp boundaries result from reworking/erosion Heiri et al., 2014;.

Previous profiles and general stratigraphy
All samples taken from various researchers during the ten years of exposure are documented in the excavation database; most are displayed in the profile sketches of H€ andel et al. (2014). High-resolution data of MS (Hambach, 2010) and numerous luminescence ages (Lomax et al., 2014) derive from an almost 8 m thick section at North Profile 2005e6 (Fig. 2AeD). The South Profile 2005e7/10e11 on the opposite site of the excavation was very close to the double infant burial (Fig. 2B;Einw€ ogerer et al., 2006), but did not expose the lower 2 m of the LPS (H€ andel et al., 2014). From this profile,  published field descriptions and results of sedimentological and geochemical analyses from each visible GH. In the two deep trenches to the bedrock that were dug in 2009 and 2013/2014 (Fig. 2EþF), the horizons beneath GH 28 were labelled GH 50e62 (61 ¼ gravel, 62 ¼ bedrock) in sounding DE98-100 (West Profile, 2009e11) and GH 80e90 (89 ¼ gravel, 90 ¼ bedrock) in sounding AB94-95 accessible to us in autumn 2013, which was extended by 1 m to both east and north in 2014 (sounding A-C94-96).

Geochronological framework
All published ages from the 2005e2015 Krems-Wachtberg excavation are shown in Fig. 3, using the compilation from the supplement of Heiri et al. (2014), complemented by luminescence ages from Groza et al. (2019) and the radiocarbon age of the mammoth scapula covering the double infant burial (H€ andel, 2017). The high-resolution environmental magnetic study of Hambach et al. (2008) was the first paleoenvironmental and chronological interpretation for the whole LPS Krems-Wachtberg and was adopted by . The MS age model (Fig. 3, cf. Heiri et al., 2014) is based on the assumption that phases of enhanced   (Hambach, 2010;Rasmussen et al., 2014). Relative paleointensity variations traced by paleomagnetic analyses indicate a formation of AH 4 contemporary to the Mono Lake event around 34.2 ± 1.2 ka Laj et al., 2014). This estimate agrees with three thermoluminescence (TL) ages of "baked loess" directly below hearth 1, with a weighted mean of 33.9 ± 2.3 ka (Z€ oller et al., 2014). A set of 38 luminescence ages from North Profile 2005-6 spans the whole LPS and indicates more or less continuous sedimentation between 40 and 20 ka and no major hiatus; smaller gaps cannot be identified due to relatively large errors (Lomax et al., 2014). Three luminescence laboratories contributed with different approaches to certain parts of the record, whereby Groza et al. (2019) analyzed 16 additional samples from the whole sequence; however, due to their scatter, the obtained ages did not lead to a more precise chronostratigraphy (Fig. 3).
Radiocarbon ages with smaller errors and higher confidence are restricted to datable material. H€ andel (2017) compiled 44 radiocarbon ages (41 from charcoal, 3 from bone) from Krems-Wachtberg and Krems-Hundsteig to determine the main Upper Paleolithic occupation periods and phases of local reworking within this episode. All 19 radiocarbon ages from the recent Krems-Wachtberg excavation calibrated with Calib Rev 8.1.0 (Stuiver and Reimer, 1993) using the IntCal20 calibration curve (Reimer et al., 2020) are shown in Fig. 3. Charcoals from AH 5 in GH 28 yielded ages from 32.0 to 33.5 ka cal BP. Two age clusters for AH 4 in GH 26 partly overlap within the 2s-error. Most ages from the AH 4.4/4.3 (in situ) group around 30.5 to 31.5 ka cal BP and are therefore 1e3 ka younger than dates by alternative methods (Hambach, 2010;. Radiocarbon ages for AH 4.11 span the timeframe of both AH 4.4/4.3 and 5, and thus confirm its stratigraphic and archeological interpretation as a layer created by slope processes that includes material from earlier Gravettian occupations (H€ andel, 2017). This layer is sealed by GH 25, identified as double ash layer (DAL) of two subsequent steppe fires and dated to 30.1e30.7 ka cal Horizontal age axis shows NGRIP d 18 O variations including counting error for comparison . Radiocarbon ages from H€ andel (2017) calibrated with Calib 8.1.0 (Stuiver and Reimer, 1993) using the IntCal20 calibration curve (Reimer et al., 2020), with close up of ages from AH 4 compared to the NGRIP d 18 O variations with GI marked.
The example of the youngest AH 4 radiocarbon age from charcoal in the infant burial illustrates possible offsets related to different calibrations (section 2.2). TL ages of baked loess from beneath hearth1 by Z€ oller et al. (2014). Luminescence ages from Lomax et al. (2014) and Groza et al. (2019) marked by (L) and (G), respectively, in the legend. Environmental magnetic age model by Hambach (2010) shown for comparison (cf. Heiri et al., 2014).  (Weninger and J€ oris, 2008) place AH 4.4/4.3 into the stadial between GI 5.1 and 5.2, which is supported by the absence of pedological features in GH 26 (H€ andel, 2017). IntCal20 ages are few hundred years younger, reaching partly GI 5.1. Taking into account age errors, calibration insecurities, and a NGRIP 2s counting error of 1 ka during this period ( Fig. 3; Rasmussen et al., 2006) we follow the stratigraphic reasoning by H€ andel (2017).

Profiles and sampling
The stratigraphies along different outcrop walls at Krems-Wachtberg are macroscopically similar (H€ andel et al., 2014). In autumn 2013, we created three complementary profiles of the LPS few meters distant from each other (KW-A, -B, eC) as parts of the new composite profile KW2015 (Fig. 2AþB, Fig. 4). KW-A is 4 m thick and reaches from GH 5 into GH 22. Profile KW-B is 2.2 m thick, embraces AH 4 and contains the sequence of GH 21 to GH 30. KW-C  (2016) complemented by a key to the main pigmenting processes. Note that pedological horizons do not necessarily reflect in situ pigmentation by soil formation, but soil may be reworked by slope processes.
T. Sprafke, P. Schulte, S. Meyer-Heintze et al. Quaternary Science Reviews 248 (2020) 106602 in the excavation basement is 2.4 m thick and encompasses the section from GH 28/80 down to the bedrock (GH 39/90) (Figs. 2F and 4). New MS data from sampling positions close to our three profiles were published by Groza et al. (2019) and are similar to data from North profile 2005e6 . This indicates that the 7.2 m thick composite KW2015 is representative for the entire Krems-Wachtberg excavation. Before sampling, the three profiles were thoroughly cleaned and documented following FAO (2006). Detailed field descriptions are refined by colorimetric (Fig. 5) and granulometric (Fig. 6) data measured on samples taken continuously in 2.5 cm resolution (continuous column sampling [CCS] after Antoine et al., 2009). All 312 samples (positions in Fig. 4) were air dried, sieved to fine earth size (<2 mm) and homogenized for further analyses.

Laser-granulometry
Granulometric data obtained at high resolution from LPS are a  (FAO, 2006;Sprafke, 2016), except for C horizons (loess sediments) and color data. Background shows from right to left the real RGB colors and four tuning steps (RGB k1 … 3, max ) of each sample. Left diagram exemplifies the variations of R real and the four tuning steps (R k1 … 3, max ), as equally calculated for the parameters B and G. Color variables of the CIELAB color space indicate the lightness (L*), red (a*), and yellow (*b) components of individual samples (details lined out in section 3.3). (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) powerful tool to reconstruct aeolian sedimentation dynamics (Antoine et al., 2009;Vandenberghe, 2013;Újv ari et al., 2016;. Post-sedimentary in-situ alteration has to be taken into account, as well as reworking along slopes including the admixture of local material, e.g. by slope wash or solifluction (Sprafke et al., 2013;Schulte et al., 2016;Vandenberghe et al., 2018 Schulte et al., 2016). Organic matter was removed before measurement by treating the samples with 0.7 ml 30% H 2 O 2 at 70 C for several hours. This process was repeated for up to 3 days until a bleaching of the sediment occurred (Allen and Thornley, 2004). To keep particles dispersed, the samples were treated with 1.25 ml Na 4 P 2 O 7 (0.1 mol*l À1 ) for 12 h on an overhead shaker (Pye and Blott, 2004;ISO 11277, 2009). This established routine allows comparison to GS data of over 20 Eurasian LPS, measured with the same protocol and device (cf. Fig. 7). Heatmaps (Fig. 6A) visualize high-resolution GS data without the loss of vertical (samples with depth) or horizontal (116 GS classes) information . In addition to the GS data provided by the LS 13320 (GS classes according to ISO 14688, 2017), we calculated the grain-size index (GSI ¼ [% 20e63 mm]/[% <20 mm]), which is interpreted as index for wind intensity (Rousseau et al., 2002;Antoine et al., 2009). To visualize granulometric deviations from baseline loess due to variations in aeolian dynamics, pedogenesis and reworking, we define a standard loess from representative Central to SE-European sites stored in the RWTH GS database defined by the following criteria: 20e50 mm GS range, low scatter (sorting), light color Particles <1 mm are extremely underrepresented during aeolian accumulation and mainly form in situ. The complex refractive index considered by Mie theory takes into account more complex pattern of laser scattering, which are enhanced by pedogenically formed submicron minerals. The input values for the calculation vary for different mineral properties. For the Fraunhofer approximation (FH), however, the mineral properties are not taken into account. The difference of GS frequencies calculated with FH and with Mie (DGSD) quantifies clay mineral neoformed by chemical weathering . The variability of low concentrated GS ranges (e.g. the submicron) within percentage-frequency distributions can be misleading due to the compositional data effect (closed number space, constant sum of 100%) (Aitchison, 1986;Roberson and Weltje, 2014). Therefore, the vertical variability of bulk sample DGSDs is analyzed as centered log-ratio transformed GS differences. The centered log-ratio transformation (clr) is suitable to transform grain size distributions from closed to real number space (Roberson and Weltje, 2014). The clr-transformation is applied to both GSDs prior to the calculation of the differential.
The resulting DGSD clr is more robust against other GS influencing processes especially during sediment accumulation by sheet wash, saltation or enhanced background sedimentation .
A Konica Minolta CM-5 spectrophotometer (RWTH Aachen) measured the air dried fine earth samples in a glass beaker on a circular field of 3.5 cm diameter. The procedure is identical to the one described by Gocke et al. (2014). L*, a*, and b* are the distinct color components plotted against RGB background colors (Sprafke, 2016). To facilitate the interpretation of the profile, a tuning of each RGB variable around its average value, proportional to the distance of the maximum and minimum values relative to the profile average, was done in three steps (RGB k1 , RGB k2 , RGB k3 ). Maximum tuning (RGB max ) is achieved by transforming the RGB variables around their average values to the complete scale of this color space (0e255; Fig. 5).

Color-based stratigraphy
Variations of color parameters or their ratios reflect complex forming processes in the interplay of dust sedimentation, pedogenesis and reworking and their interpretation is not always straightforward. Sprafke (2016) uses color data to enhance field descriptions objectively, with stratigraphic units classified with master horizons (A, B, C and combinations) and their subordinate characteristics (FAO, 2006). Pedological subdivisions of LPS (Bronger, 1976(Bronger, , 2008Schirmer, 2000Schirmer, , 2016 are reasonable, as visible changes in the stratigraphy of loess profiles are mainly related to post-depositional alteration of the deposits. Numerical color values support the precise separation of stratigraphic units and the classification of the units in relation to each other based on changes in certain color components (Sprafke, 2016). Color-based stratigraphy supports thorough field documentation but remains entirely descriptive and does not discriminate between in situ and reworked material.
C horizons represent the unaltered parent material (loess). A horizons are surface soil horizons characterized by increase in organic matter and thus often darker in color (lower L*-value) than loess. B horizons are subsurface soil horizons that show enhanced weathering and more brownish to reddish coloring (higher a*-and b*-values). The combination of two master horizon symbols indicates a transitional horizon, whereby the first letter characterizes forming in reductive environment above the permafrost table (Antoine et al., 2009). At present, temperate zone loess profiles do not contain cryic (perennially frozen) horizons and the reductaquic principal qualifier is not adequate for KW2015, as all Cr horizons are less than 25 cm thick (IUSS Working Group WRB, 2014). In the following, we refer to these specific loess paleosols as tundra gleys (Bibus, 1974;Frechen et al., 1999;Nigst et al., 2014;Moine et al., 2017).
In addition to the colored stratigraphic log, a transparent sand/ gravel signature layer is added to the profile sketches, in order to illustrate units different from aeolian dust overprinted by pedogenesis of variable degree; these are units affected by higher wind strength or local reworking (Antoine et al., 2009;Vandenberghe et al., 2018).

Field description
KW2015 exhibits no well-developed paleosol. Silty yellowish loess horizons with variable sand contents, partly accompanied by more brownish colors are repeatedly intercalated by pale horizons. The succession below the cultural horizon AH 4 is more homogenous with diffuse horizon boundaries, whereas the units above are more distinct and more variable. Most a priori defined GH  can be well identified (Fig. 8).
The KW2015 composite is subdivided into ten major units, labelled IeX, with subunits a, b, c etc. (Fig. 4 þ 5). Unit X at the bottom, overlying the basal rock, is loess sediment comprising a thick pale horizon. Two weak brownish horizons divided by loess are present in the superimposed unit IX. Unit VIII consists of weakly differentiated loess sediments with incipient signs of reduction and humification. The reworked cultural layer AH 4.11 and superimposed sandier loess are located in between pale horizons (Unit  (Hambach, 2010) shown for comparison. Note that reworked (rw) units are characterized by decreases in GSI and increases in sand contents (black and grey arrows), which allows the differentiation of in situ soils (unit IX) from soil-bearing brownish/darker sediment (color trends indicated by red and white arrows). Next to two oxidized terrestrial soils, there are seven tundra gleys (TG) related to in situ reduction. The left column of this figure provides the semi-quantitative paleoenvironmental interpretation (adapted from Heiri et al., 2014) based on reconstructed processes. Colors in the bars at the bottom indicate contribution of dust deposition (yellow), chemical weathering (grey for predominant reduction, brown for predominant oxidation, among other weathering processes), and permafrost (blue). These processes and environments lead to semi-quantitative paleoclimatic inferences, indicated below. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.) VII). Sandy, brownish loess sediments with wavy layers form unit VI, which is sharply cut by unit V, characterized by two pronounced bleached horizons with sandier loess in between. Centimeter-sized ice wedges at the boundary of units VI and V have revealed a polygonal network during the excavation. Loess sediments with more pale horizons are superimposed (Unit IV). The lower half of Unit III is characterized by brownish loess sediment with higher sand contents, microcracks and laminae. Unit IIIa is even stronger pigmented, has more coarse components and a wavy to pocket-like lower boundary. The loess sediments above are more homogenous, showing only weak differentiation, with pale horizons and a grey, sandier horizon in the upper part of unit II. Above we note an increase in coarse mineral components (unit I).

Color variations
The main trends of color variations (Fig. 5) summarize as follows: the lightness (L*) of the overall pale yellowish substrate is highest in the lowermost meter of the LPS, directly above the bedrock, and within the upper 2.7 m. Distinct maxima indicate bleached loess sediments (IIIb, IVb, Vg, VIIId), whereas minima are related to darker (IIa, Va, VIIe-f, IXa þ d) or brownish horizons (IIIa, IIIf-h, VIb). Bleached horizons below 2.7 m often have relatively higher L*, but their main characteristic is reduced redness (a*) and yellowness (b*).
The a* and b* values oscillate in a similar way throughout most parts of the profile. In the upper 2.7 m there are higher a* and b* values compared to the loess sediments below. Maxima in these parameters well reflect the brownish sand enriched layers (IIIa, IIIfh, VIb) and two incipient paleosols in unit IX. Minima are related to bleached horizons. Below AH 4 there are more variations in b* then a*. Unit Va records more expressed a*, relative to b*.
The visualized RGB colors confirm the presence of yellowish loess sediment with subtle color variations for KW2015, but tuned RGB values strongly enhance the contrast and detail (Fig. 5). Accordingly, the LPS is divided in three main parts, with the upper 2.7 m being more yellowish, reddish and lighter, the middle part more blueish and darker, and the lowermost 1.3 m again lighter. The distinct colors support the stratigraphic log (see section 4.4).

Granulometry
The heatmap of GS distribution (Fig. 6A for the KW2015 composite, complete figure in supplement Fig. S1) indicates an overall rather uniform character of the loess sediments with a general maximum in the coarse silt fraction around 40 mm. Lower concentrations in coarse silt are found in units with high contents of medium to coarse sand (units I, III, V, VI, base of X) or higher clay content (unit IX). The GS distribution of typical loess from KW2015 (sample KW-A_46) overall agrees with results from representative Central to SE-European loess (from the LPS Garzweiler, Remagen, Ostrau, Titel, Vlașca) measured until 2019 with the same device (Fig. 7A). Fig. 7B compares the GS distribution of this European average loess 2019 (EAL_19) to samples affected by pedogenesis (sample KW-C_52) and reworking (sample KW-A_84) of KW2015. Sample KW-C_52 from a weak paleosol (unit IXe, Fig. 4) has a higher clay content, a medium silt shoulder, and a main mode shift to smaller GS. Sample KW-A_84 from reworked deposits (unit IIIh, Fig. 4) has distinct medium to coarse sand modes.
Deviations of all individual samples from KW2015 to EAL_19 are visualized in a heatmap (Fig. 6C, for GS > 4 mm; complete figure in supplement Fig. S2). Next to marked increases in medium to coarse sand or clay in distinct units, we note that the sequence below 2.7 m (units IVe to X) has slightly higher fine sand contents compared to the EAL_19. The sequence above 5 m (units I to VII), has a higher variability in GS distribution and overall slightly lower amounts of medium silt. This increase in GS variability is also shown in different GS parameters, with examples in Fig. 6A. Contrary to the GSI variation in the LPS Nussloch (Antoine et al., 2009), the GSI variations of KW2015 do not always correlate to the sand contents (Fig. 8), which is especially visible in units with higher amounts of medium to coarse sand and peaks in GS median (Fig. 6A). Clay contents range between 6 and 10% with maxima in unit IX and minima in unit VII. Noteworthy is a constant increase in clay from unit VII (6%) to the top of the sequence (9%). The DGSD clr values as measure for the amount of pedogenic clay

Stratigraphy of the KW2015 composite
The stratigraphy of KW2015 is based on detailed field logging supported by high-resolution colorimetric data and complemented by a high-resolution GS data set. 10 major units (I-X) were visible in the field, while color data facilitate the distinction and classification of up to 9 subunits per major unit (a-i; ig. 4). Most horizons of KW2015 classify as C horizons or as transitional to A (CA/AC) or B (CB/BC) horizons, depending on the contribution of dark or brown colors, respectively. Pale horizons with low L* and/or low a*/b* values classify as C(r) or Cr horizon depending on their intensity. Only two horizons of the whole section are sufficiently distinct from loess to classify without a C master horizon symbol: Unit IIIa is a brownish sandy horizon labelled GH 10 in the field assessment, classified as Bw horizon. Unit Va, the very dark horizon on top of GH 19, was interpreted as fossil manganese band by , but this was not confirmed by their geochemical data. Until composition and genesis of unit Va are clarified it is classified as A horizon based on its color.

Color-based loess stratigraphy and granulometry
Pale yellow, aggregated, carbonate bearing substrate dominated by coarse silt that formed mainly by accumulation of windblown dust constitutes the original baseline material in LPS (P ecsi, 1990;Pye, 1995;Sprafke and Obreht, 2016). Loess GS variations primarily reflect changes in sedimentation dynamics, whereas color reflects mainly post-depositional alteration (pedogenesis). Both parameters for unravelling this dynamic interplay are now available in high resolution from the LPS Krems-Wachtberg and support a robust stratigraphy (KW2015). Understanding the complex pedosedimentary evolution is prerequisite to paleoenvironmental reconstructions.
Quantified color variations support a detailed and robust stratigraphic log for weakly differentiated loess profiles (Sprafke, 2016). Master horizons (A, B, C and combinations) and subordinate characteristics (h, r, t, w etc.) from pedology (FAO, 2006) capture the high stratigraphic variability of KW2015 (Fig. 4 þ 5). This approach is chosen for a robust subdivision of complete loess profiles, whereas reconstructions of (sub)unit evolution requires morphological observations and further data (e.g. from granulometry). Color data strongly support the definition of boundaries and a consistent classification of units, but do not replace profile or drill core documentation. . KW2015 not only reproduces, but considerably refines the GH stratigraphy defined during the archeological excavations (H€ andel et al., 2014) and provides a consistent scientific terminology (Fig. 8).
Six well-developed pale horizons interpreted as tundra gleys in units IV-VII and X, three reworked brownish yellow layers in units III and VI and two weak paleosols in unit IX were clearly visible in the field, whereas less developed units could be robustly defined based on color data (Fig. 4). It is important to note that the horizon designations from pedology used here are descriptive and not genetic. Interpreting the LPS evolution requires a differentiation of paleosols from reworked soil material based on field observation and GS data. The dark and brownish AB and BC horizons below AH 4 show increases in finer GS indicating pedogenesis, whereas all BC and Bw horizons above AH 4 contain higher proportions of sand and gravel, admixed during soil reworking (Fig. 8).
The detailed KW2015 color-based stratigraphy complemented by granulometry differentiates LPS units according to the predominance of aeolian, pedogenic or slope processes during their formation, which helps to select adequate proxies representing these process groups. Such differentiation is relevant for LPS locations with an overall higher weathering degree, lower sedimentation rates and/or variable topography. Soil formation is a downward oriented alteration of already deposited sedimentary layers (Rousseau et al., 2017b) and slope processes lead to erosion and reworking of sediment and soil . Sedimentological approaches, i.e. using vertical LPS parameter variations as quantitative estimates for process intensities and factors through time (commonly referred to as proxies) are applicable, when the predominance of aeolian sedimentation is verified (C, CA and CB horizons, if not reworked). In all other cases, pedogenesis has to be included in parameter interpretation. The stratigraphy of KW2015 indicates numerous phases of pedogenesis and reworking, therefore a differentiated approach to unravel its formation is required.

Granulometry
Throughout KW2015, coarse silt, typical for loess (Pye, 1995;P ecsi and Richter, 1996), is the predominant GS (Fig. 6A). Compared to the EAL_19 measured with the same device only the samples between 0.5 and 1.4 m and between 2.3 and 3.2 m of KW2015 have a comparable frequency in the mode fractions (light colors in the coarse silt fractions of Fig. 6C). Unit II has the closest match to typical loess from Central Europe and may be defined as local reference loess. Slightly larger and more frequent modal fractions likely indicate higher local wind-speed or closer deflation centers (Nugteren and Vandenberghe, 2004;Újv ari et al., 2016). Additional submodes in the medium to coarse sand range (units I, III, V, VI, and the bottom of the LPS) result from admixtures of local coarse material by sheet wash or solifluction (Sprafke et al., 2013;Zeeden et al., 2016), which is indicated by layering of units III and VI and partly sharp unit boundaries.
Deviations in GS along LPS are usually related to changes in aeolian dynamics during the formation of the record (Vandenberghe, 2013;Újv ari et al., 2016). GS ratios represent certain process groups more adequate than variations of GS classes, due to the compositional data effect. The GSI unravels variations in wind intensity (Antoine et al., 2009), if in-situ alteration resulting in clay mineral formation and local reworking are excluded . At the LPS Nussloch, increases of the GSI are accompanied by higher fine sand contents due to a shift of the GS mode into the direction of coarser GS resulting from higher wind intensity (Antoine et al., 2009). At KW2015 only the lower part of the sequence (units VII-X) has comparable patterns in both, GSI and sand content. Above units with highest sand contents are accompanied by lower GSI (Fig. 8). Local reworking (units marked by "rw" in the KW2015 composite) along the slope mobilized coarse material, which creates distinct medium to coarse sand modes (Fig. 6A), independent of smaller shifts in the GS mode by changes in wind intensity.
Sheet wash mobilizes fine sand to coarse silt better than more cohesive finer fractions, which strongly affects the GSI, as coarse silt (20e63 mm) is the numerator of this ratio. This proxy of past aeolian dynamics is apparently inadequate for settings, where coarser components are admixed by non-aeolian processes. This finding highlights that GS proxies require a critical evaluation related to the regional climatic and local topographic context . KW2015 sand contents are the most suitable parameter to reflect local changes between morphodynamic activity (open vegetation cover, intensified wind-speed and reworking) and stability (closed vegetation cover, low wind-speed and possibly pedogenesis; Fig. 9).
The overall clay contents show relative maxima in the weak paleosols of unit IX which can be attributed to pedogenesis, given the brownish pigmentation and a weak soil structure. The slight clay increase from unit VII to the top of the sequence likely reflects a change in dust source, as our stratigraphy does not point to significant pedogenesis in this part of the sequence, except reduction leading to tundra gley formation.  report a marked increase in carbonate contents from unit VI (20e25%) to the top (35e40%), which may contribute to the clay fraction if finely dispersed during loessification (Sprafke and Obreht, 2016). Absolute clay contents as proxies for weathering intensity areonly recommendable after verification of sedimentary homogeneity and cross-checking with other parameters.
The DGSD clr signature of the LPS is a direct indicator for postdepositional chemical weathering processes and successfully applied in several Central European LPS, e.g. Garzweiler, Düsseldorf-Grafenberg, Ringen or Frankenbach, all located in the Rhine catchment (Zens et al., 2018;Schulte and Lehmkuhl, 2018;Fischer et al., 2019). These LPS have DGSD clr values around 2 to 7 (dimensionless) in well-developed interstadial soils, which is considerably more than for KW2015, where DGSD clr values range around zero in the entire dataset (Fig. 6B, left side). The incipient paleosol (AC/BC) in the lower part of unit IX is the only paleosol with a pronounced signature of the DGSD clr , indicating the formation of clay minerals by silicate weathering. The positive deviations to EAL_19 in the mSi (medium silt) fractions are probably due to pedogenic formation of stable microaggregates (Fig. 6C). These deviations occur in the entire Unit IX. The slightly increased values between Unit I and Unit V (visible in the 42 nm DGSD(clr) curve) may result from slight chemical weathering or a change in the dust source. The latter is more likely, as sediments interpreted as loess or tundra gleys are often not altered by considerable silicate weathering when a carbonate buffer is abundant (Blume et al., 2010).

Colors and magnetic susceptibility
Throughout the sequence, color parameters are more variable than DGSD clr , showing their high potential sensitivity for initial pedogenesis. Pigmenting soil forming processes like oxidation, reduction or humification (Cornell and Schwertmann, 2003;Blume et al., 2010) can already take place in the presence of carbonate that largely buffers further chemical weathering (cf. Bronger, 1976). However, colors cannot distinguish soil from soil sediment. Interestingly, the DGSD clr values are low in reworked units of KW2015, suggesting that aquatic reworking removes clay from soil sediments, whereas enough iron oxides remain as brownish pigments, attached to the remaining silt and sand grains. Color variations are increasingly applied in loess research as proxy for paleoenvironmental evolution (Luki c et al., 2014;Krauss et al., 2016;Vlaminck et al., 2016). The Redness Index (RI) or a* are used to detect horizons affected by pedogenesis resulting in the formation of iron oxides as reddish or brownish pigments (Luki c et al., 2014;Vlaminck et al., 2016). Calculations of the RI recognize that increasing hematite contents not only increase a* but also reduce the lightness of a soil and thus include L* in the denominator in the formula for the RI (Viscarra Rossel et al., 2006). Being sensitive to dark pigments, the RI is not applicable to LPS with humic horizons (Sprafke, 2016). For KW2015, we note that the maxima in reddening are related to reworked layers in unit III. Dark red components, partly as coatings around sand grains indicate iron remobilization in this better drained material, but they may also originate from weathered Neogene gravels exposed upslope on the Wachtberg hill (Fuchs and Grill, 1984). Unit VI is equally reworked, with a maximum peak in a*, relative to the surrounding material. Only in unit IX the increases in a* seem to result from in-situ oxidation (Fig. 8).
The overall increase in L* in the upper 3 m (units I-IV) can be explained with a 10e15% increase in carbonate contents (white pigment) as measured by . All tundra gleys above AH 4 have distinct local peaks in carbonate (c. 3e8% more compared to surrounding layers), which agrees with observations of Meyer-Heintze et al. (2018) from a sequence close-by and can be explained by capillary enrichment of CaCO 3 during reduced  (Nigst et al., 2014, redrawn), the Nussloch reference LPS (Moine et al. (2017, redrawn) with GSI plot from P4 and loess events (LE) indicated (Antoine et al., 2009), NGRIP dust record ; note a change of the age scale at 30 ka), and the German loess stratigraphy . Age of Eltville Tephra from Zens et al. (2018). Sand contents of KW2015 and Krems-Wachtberg East (Meyer-Heintze et al., 2018) reflect the geomorphic response of these profiles to paleoenvironmental changes. Note the disagreement in correlation of MPG tundra gleys to GI, i.e. the Nussloch model (TG ¼ GI; Moine et al., 2014) and Willendorf model (TG s GI; Nigst et al., 2014) discussed in the text.
conditions (Cilek, 2001). The tundra gley below AH 4 does not show an increase in carbonate and the dataset of  does not reach pale horizons below. In the Krems region both exist, thin pale carbonate rich tundra gleys and thicker blueish ones without increase in carbonate (Sprafke, 2016). Overall, there is no clear connection between tundra gley formation and carbonate dynamics in Central Europe (Antoine et al., 2009;Krauss et al., 2016), but possibly high loess carbonate contents result in calcite enriched tundra gleys, whereas tundra gleys in loess with low carbonate contents have reduced calcite contents.
The characteristic feature of tundra gleys are lower a* and b* values due to iron reduction, but changes in L* seem to depend mainly on the presence of carbonate. We do note, however, a steady increase in L* in the lower 2 m of the sequence, which likely relates to slight reduction processes due to higher pore water contents close to the bedrock, resulting in a slight overprinting of the color record.
All these qualitative observations suggest that color variations need to be carefully interpreted, before they can be used as paleoenvironmental proxies. Relative variations of color and further parameters (e.g. field observations, GS distributions) have to be considered for stratigraphy and a process-based interpretation of the pedosedimentary evolution of KW2015 (section 5.3). RGB tunings coupled with L*, a*, and b* variations are useful to delimit and classify stratigraphic units. Caution is required to not overinterpret strongly tuned color plots, therefore comparison to field observations is required (Sprafke, 2016). RGB tuning around the profile average works well for KW2015 (Fig. 5), but sequences with well-developed paleosols require the tuning around a standard loess color. To date there are no unified routines for loess colorimetry to provide robust color tuning as a tool to facilitate loess stratigraphy, but our results indicate a high potential for this approach.
Environmental magnetic parameters from the LPS Krems-Wachtberg (Hambach, 2010;Groza et al., 2019) are sensitive to most stratigraphic units of KW2015 (Fig. 8). Tundra gleys of units I, VII and X are represented by considerably lower ARM/MS ratios indicating a higher intensity of pedogenesis (Hambach, 2010). However, the weak unit IX paleosols have no clear correspondence in the environmental magnetic record and several stratigraphic units have ambiguous responses to these data. Connecting the environmental magnetic parameters to the KW2015 stratigraphy may strongly enhance the potential of these proxies.

LPS evolution and paleoenvironmental inferences
KW2015 sand contents inform about local variations between morphodynamic activity and stability, but not in detail about the complex interplay of aeolian sedimentation, pedogenesis, and reworking in the context of the local paleoenvironmental evolution. Complex terrestrial records require a reconstruction of the forming processes based on the integration of complementary parameters (Meszner et al., 2013;Sprafke, 2016;Meyer-Heintze et al., 2018). Unit-based reconstructions benefit from additional paleoenvironmental proxies (Kukla, 1977;Haesaerts et al., 1996Haesaerts et al., , 2007, whereas our interpretation is entirely based on reconstructing the pedosedimentary evolution. Fig. 8 (left side) summarizes the semi-quantitative paleoenvironmental conditions during the formation of KW2015 reconstructed from forming processes of stratigraphic units. The scheme developed for the LPS Willendorf by Heiri et al. (2014) was slightly adapted to better include representative paleoenvironments. It differentiates dust deposition taking place in cold steppe (without permafrost) or in tundra (with permafrost), a classification that depends on the local stratigraphic context, i.e. the paleosols developed therein and further paleoenvironmental information. At KW2015 we note a marked change from terrestrial soils and smooth GS variations (unit IX) to the predominance of tundra gleys and more variable GS pattern (units I-VIII), which we interpret as major climatic transition that goes along with the onset of permafrost and more active floodplains as local deflation centers (Moine et al., 2017). Tundra gleys in this part likely represent milder conditions, with reduced dust deposition (Antoine et al., 2009). The tundra gley of unit X likewise indicates a phase of permafrost presence.
On the cold end of the paleoenvironmental scale (Fig. 8) we introduce a new category without dust accumulation or pedogenesis under the presence of permafrost, which corresponds to a polar desert (cf. Sirocko et al., 2016). This is an open landscape with almost no vegetation cover, with no relevant dust deposition, but episodes of reworking by slope wash. In the geological setting of Krems-Wachtberg these processes result in the admixture of coarse material of upslope weathered gneiss and Neogene gravels.
KW2015 rests on crystalline rock that is partly covered by fluvial gravels, which are reworked and mixed with initial aeolian dust. The thick tundra gley of unit X formed from loess under reducing conditions due to the presence of permafrost impeding drainage (Antoine et al., 2009;Terhorst et al., 2015). Unit IX represents two weak paleosols with brownish pigmentation, interrupted by dust deposition. The lower paleosol shows an advanced development involving clay mineral neoformation as indicated by an enhanced DGSD clr value and positive deviations of EAL_19 in the mSi (medium silt) fractions. The latter is likely due to the pedogenic formation of stable micro-aggregates. Unit VIII represents a strong increase in dust deposition, interrupted by weak pedogenesis, with unit VIIId as weak tundra gley. The slightly darker pigmented unit VIIIf has a higher sand content and is therefore interpreted as short scale (maybe aeolian) reworked incipient topsoil, rather than an insitu humification.
Unit VII encompasses cultural layer AH 4, surrounded by two well-developed tundra gleys that indicate the presence of permafrost and negligible dust deposition. For AH 4 itself, we assume the presence of permafrost and dust deposition as dominating process (H€ andel, 2017). Note that KW2015 does not assess the in situ occupation layer AH 4.4/4.3 as this unit did not extend so far. Topographically, the sampling locations are positioned downslope of the occupation layer, and the assessed part of AH 4 (AH 4.11) contains archeological material redeposited shortly after the site was abandoned, indicated by the scatter of radiocarbon ages related to AH 5 and AH 4.4/4.3 ( Fig. 3; H€ andel, 2017).
Unit VI consists of reworked material and can be attributed to a dry open landscape with sparse vegetation cover facilitating surface processes. Unit V encompasses two further tundra gleys, interrupted by dust deposition and a phase of reworking indicated by coarse material admixture. Unit IV formed by increased dust deposition with at least one minor interruption in the lower part (unit IVf) and a major interruption marked by a tundra gley at the top (IVa-b). The two well-visible brownish horizons in unit III with higher shares of coarse material result from intensified slope processes in a dry open landscape and were interrupted by a phase of dust deposition before a short period of stability (unit IIIb). Later dust deposition (unit II) was interrupted by incipient tundra gleys and an episode of reworking (IIa). Unit I consists of loess, partly with signs of reworking in an open landscape (Fig. 8).

Chronological framework and correlations
Detailed stratigraphic logs of LPS are prerequisite to understanding profile evolution in the context of local paleoenvironments (Haesaerts et al., 1996(Haesaerts et al., , 2007Antoine et al., 2009Antoine et al., , 2013Terhorst et al., 2015). A chronological framework is required to differentiate process intensity from process duration and to allow for correlations to other paleoenvironmental records that lead to an improved understanding of paleoclimatic evolution and regional effects of climatic changes (Sprafke, 2016). The available geochronological dataset from Krems-Wachtberg was outlined in section 2.2 and is shown in Fig. 3. The age model of Hambach et al. (2008) used by  relies on variations of magnetic parameters and their correlation to the NGRIP oxygen isotope variations. Partly inconsistent behavior of magnetic parameters compared to our stratigraphy and a 1e3 ka younger age of AH 4 based on radiocarbon dating suggest a modified age model. Luminescence ages are the only absolute age control from the entire LPS Krems-Wachtberg (Lomax et al., 2014;Groza et al., 2019), but relatively large error bars and methodological inconsistencies do not allow for precise correlation with millennial scale paleoclimatic reference records, e.g. the NGRIP dust record (Fischer et al., 2007;Rasmussen et al., 2014). The 19 radiocarbon ages with smaller errors seem reliable, being well within the dating range of this method (Nigst et al., 2014).
The available radiocarbon ages from AH 4 and surrounding horizons can be used as references to tentatively evaluate the reliability of the luminescence ages as calibration insecurities are far smaller than OSL errors (Fig. 3). Overall, the ages of the three luminescence laboratories involved in the study of Lomax et al. (2014) agree within error, but quartz fine grain optically stimulated luminescence (OSL) ages appear as best match to the radiocarbon ages. Polymineral fine grain infrared stimulated luminescence (IRSL) ages appear systematically younger (although no fading was detected in the laboratories) and quartz coarse grain OSL ages seem slightly too old (cf. Lomax et al., 2014). Despite these insecurities, the data indicate a quasi-continuous sedimentation between c. 40e20 ka and suggest a stratigraphic comparability to well-dated LPS of the same period for the upper and lowermost parts of KW2015, with less robust age control. Fig. 9 compiles stratigraphic information from the nearby LPS Krems-Wachtberg East (Meyer-Heintze et al., 2018), the LPS Willendorf c. 25 km upstream the Danube in the Wachau Valley (Haesaerts et al., 1996;Nigst et al., 2014), the LPS Nussloch close to Heidelberg as Central European reference LPS and the German standard loess stratigraphy Zens et al., 2018). The 30 ka line is used as a common reference for all profiles. This reveals overall similarities in the well-resolved stratigraphic successions and regional/local modifications. Especially when KW2015 is compared to the well-dated LPS Nussloch (Moine et al., 2017) we note some striking similarities and characteristic deviations. The incipient paleosols in unit IX of KW2015 correlate well with the Lohne soil at Nussloch. Both records contain a comparable number of tundra gleys, which agree in their relative stratigraphic position and intensity. Pronounced loess events (LE), marked by GSI peaks (Antoine et al., 2009), seem to correlate to reworked horizons of KW2015 marked by higher amounts of sand (section 5.2, Fig. 8). These features suggest to largely adapt the Nussloch chronostratigraphy for those parts of KW2015 with limited absolute age control, with modifications discussed in the following.

Local environment vs. catchment morphodynamics
KW2015 records millennial-scale paleoenvironmental responses to northern hemispheric climatic changes with specific successions of loess, paleosols and, reworked units. The low weathering degree of unit IX compared to upper MPG soils in the Rhine catchment indicated by the DGSD clr (Zens et al., 2018) can be explained by the continental location of KW2015 at the eastern margin of the Bohemian Massif. Beyond in-situ alteration, the influence of local topography and catchment morphodynamics requires consideration (Smalley et al., 2009).
Reworked layers at KW2015 and Nussloch LE are local signals for the coldest conditions during the UPG, which correlate with NGRIP dust peaks (Moine et al., 2017) and are partly related to Heinrich events ( Fig. 9) (Sanchez Goñi and Harrison, 2010;Rasmussen et al., 2014). During these coldest conditions, maritime moisture of Western Central Europe likely supported Western Alpine glacier activity and enhanced silt production, dynamic braided rivers in the Rhine catchment that transport and temporarily expose silt, and a vegetation cover to efficiently trap dust (Pye, 1995;Smalley et al., 2009). Additional dust at Nussloch originates from Western Europe and was brought in by frequent dust storms; a system coupled to large scale dust dynamics recorded in the NGRIP dust record (Antoine et al., 2009;Moine et al., 2017). Krems at the eastern margin of the Bohemian Massif receives overall less maritime moisture and is located within the Danube catchment, which mainly drains the more continental Eastern Alps. During the coldest periods of the UPG the scarcity of vegetation facilitated reworking and reduced the local potential to trap aeolian dust, which was likely less efficiently produced in the catchment and transported downstream before aeolian entrainment.
Nussloch UPG tundra gleys separate LE and represent milder conditions, correlating to and partly even outnumbering GI (Antoine et al., 2009;Moine et al., 2017). At Willendorf, a record with higher resolution during the MPG, tundra gleys are thought to represent the coldest conditions during the last glacial period (Haesaerts et al., 1996;Nigst et al., 2014). These contrary interpretations of tundra gleys are based on LPS with detailed stratigraphic records and good age control. For KW2015, chronological information from AH 4 and surrounding units, together with stratigraphic reasoning are in strong support for the age model of Nussloch for the UPG and Lohne soil equivalent (units I-IX), whereas the tundragley below (unit X) is problematic. OSL quartz fine grain ages, which are reliable in tundra gleys around AH 4, suggest a sedimentation age of 39.1 ± 3.1 ka for the basal tundra gley parent material, matching Heinrich event 4 (Lomax et al., 2014). A correlation with Nussloch Gm3 (GI 10e11) requires a generous interpretation of the OSL age error and implies an absence of Heinrich event 4 in KW2015. The key question is, does unit X represent a milder phase (e.g. GI 9 and 10) of an MPG dominated by permafrost, with few oxidized paleosols representing the longest GI (Rousseau et al., 2017a), or does it represent climatic Semi-quantitative paleoenvironmental information (cf. Fig. 8) from the LPS Krems-Wachtberg. C: Same as B for the LPS Willendorf with Palaeolithic techno-complexes added (modified from Heiri et al., 2014;Nigst et al., 2014). D: Main stratigraphic units of Central European loess stratigraphy . E: Alpine foreland pollen records from Austria (A) and Switzerland (CH) as compiled by Heiri et al. (2014). F: S Alpine piedmont lobes Garda and Tagliamento from Monegato et al. (2017). G: Radiocarbon ages (calibrated with IntCal20; Reimer et al., 2020) Luetscher et al., 2015) and H€ olloch (Moseley et al., 2014). L: ELSA landscape evolution zones based on proxy evidence from Eifel Maar lakes (Sirocko et al., 2016). M: Global stack of marine oxygen isotope variations by Lisiecki and Raymo (2005) with MIS 3/2 transition marked based on Andersen et al. (2006). N: Summer insolation at 60 N from Berger and Loutre (1991). O: NGRIP oxygen isotope record as paleoclimatic reference for last glacial NW European suborbital paleoclimate fluctuations . deterioration (Heinrich 4) within an MPG otherwise largely free of permafrost?
It is reasonable to correlate paleosols of more advanced development to longer GIs (Rousseau et al., 2017a), but reduction above permafrost and oxidation under well-drained conditions are opposing pathways of pedogenesis. An almost constant presence of permafrost during the entire Pleniglacial and thawing during the longest GI (time-dependence), gradually leading from reducing to oxidizing conditions, could be one explanation (TG ¼ GI model). On contrary one may assume a Pleniglacial with sporadic permafrost, especially during the coldest conditions, represented by tundra gleys (TGsGI model; Fig. 9). Both interpretations may depend on the geographic location of the LPS studied and on the stratigraphic context.
KW2015 records a marked transition from oxidized paleosols representing GI 8 to 7 to reduced paleosols representing later GI, in agreement with Central European paleoenvironmental and global paleoclimatic records (e.g. Lisiecki and Raymo, 2005;Sirocko et al., 2016), which will be discussed in the following section. Based on chronological and stratigraphic evidence lined out above, we assume the tundra gley of unit X represents an episodic occurrence of permafrost in an overall milder MPG (TGsGI model), related to climatic deterioration during Heinrich event 4. Fig. 10 compiles paleoenvironmental information from Willendorf and KW2015 using the TG ¼ GI model for the cold UPG and the TGsGI model for the milder MPG.

Implications for the Central European paleoclimatic evolution
Based on the absolute chronological information and stratigraphic reasoning, we present the paleoenvironmental information available from KW2015 (Fig. 8) on a time scale between 45 and 20 ka (Fig. 10B). Note that with increasing vertical distance from the radiocarbon ages within KW2015, chronological insecurity increases, which is indicated by lighter bars in the paleoenvironmental record. Reconstructions from the LPS Willendorf 23 km upstream the Danube (Haesaerts et al., 1996;Nigst et al., 2014) are added ( Fig. 10C) with adaptions related to the paleoenvironment classes from Heiri et al. (2014).
Both loess records agree in their general stratigraphic pattern and indicate a comparable paleoenvironmental evolution. Together these LPS record millennial-scale paleoenvironmental changes from 50 to 20 ka, a period poorly reflected by other records in the Eastern Alps and their forelands (Starnberger et al., 2011;Heiri et al., 2014;Mayr et al., 2019;Moseley et al., 2020;Stojakowits et al., in press). In addition, both localities are important archeological sites and document regional environmental evolution and cultural adaption from the arrival of AMH until the LGM (Nigst et al., 2014;Staubwasser et al., 2018). A total of nine cultural layers at Willendorf are attributed to the Aurignacian (43.5e35 ka) and the Gravettian (35e29 ka); the Upper Paleolithic sequence at Krems-Wachtberg relates to the Gravettian, with an age of c. 31 ka for occupation layer AH 4.4/4.3. The regional technological change from Aurignacian to Gravettian at around 34 to 36 ka (Fig. 10C) coincides with pronounced environmental changes recorded in KW2015 as part of a general trend in climatic deterioration, along with reducing summer insolation ( Fig. 10O; Berger and Loutre, 1991).
A general increase in oxygen isotope values of deep-sea foraminifera since c. 45 ka (Fig. 10M) indicates increasing global ice volume, culminating in the global LGM around 20 to 24 ka (Lisiecki and Raymo, 2005). The boundary between the milder MIS 3 to the colder MIS 2 has long seen ambiguous definitions (Sanchez Goñi and Harrison, 2010). Conventionally, the end of MIS 3 is correlated with the end of GI 3 in Greenland ice cores, which is dated to c. 27.5 ka Svensson et al., 2006). In Central European loess, the Lohne soil is identified as final paleosol of the milder MPG. Chronological constraints and regional paleoenvironmental variations lead to deviating ages for the Lohne soil and thus the MPG/UPG transition ( Fig. 10D; Kadereit et al., 2013;Terhorst et al., 2015;Sauer et al., 2016). At the well-resolved and tightly dated LPS Nussloch, this boundary dates to c. 34e35 ka (Moine et al., 2017), comparable to other records from western Germany (Zens et al., 2018). Correspondingly, our evidence from KW2015 indicates the onset of the cold UPG around 34e35 ka, i.e. at the end of GI 7 (Fig. 9). Sirocko et al. (2016) suggest the use of paleovegetation records from Eifel maar lakes as Central European reference for paleoenvironments since 60 ka, and group detailed information to landscape evolution zones (LEZ; Fig. 10L). A transition from boreal forest (LEZ 7) to steppe (with trees during interstadials; LEZ 6) around 37 ka, to tundra (LEZ 5) around 28 ka and to polar desert (LEZ 4), corresponds to overall increasingly cold conditions from 45 to 20 ka (Sirocko et al., 2016). Alpine foreland pollen records (Fig. 10E) indicate the onset of tundra vegetation at around c. 36 to 38 ka, whereas open spruce forest still predominated from 43 to 45 ka (Heiri et al., 2014 and references therein). The LPS Willendorf (Fig. 10C) records boreal conditions before 44 ka and for the following 10 ka oscillations between stadials with cold steppe/ tundra ecosystems to Schwallenbach interstadials (SB Ia þ b, II, III) with steppe vegetation and partly trees (Nigst et al., 2014). The two weak paleosols from KW2015 unit IX most likely correspond to SB II and III, which predate the onset of continuously cold conditions from 34 ka into the LGM.
There is remaining dispute about the onset and duration of the Alpine LGM and the timing of major glacier advances (Starnberger et al., 2011;Gaar et al., 2019). Sp€ otl et al. (2013) report a marked environmental change from the Baumkirchen site in the Inn Valley close to Innsbruck, at around 32e33 ka, which marks the onset of the Eastern Alpine glacier advances in the Upper Würmian (UPG; Fig. 10G). First advances of the Rhine glacier into the foreland reportedly occurred around 32 ka (Fig. 10H, from a compilation by Heiri et al., 2014) or considerably later around 28 ka ( Fig. 10I; Preusser et al., 2011), consistent to advances of the Garda and Tagliamento piedmont lobes in Northern Italy ( Fig. 10F; Monegato et al., 2017). The MIS 2 speleothem oxygen isotope record from Siebenhengste Cave in Switzerland (Fig. 10K) supports that the maximum glaciation occurred around 24 ka (Luetscher et al., 2015), but the drivers and extents of earlier last glacial ice advances are unclear. A modelling study by Seguinot et al. (2018) suggests an early Alpine glacier advance culminating around 30 ka and a maximum advance around 24.5 ka (Fig. 10J).
Detailed chronologies for the onset and dynamics of foreland glaciations in the Danube catchment are scarce. Reuther (2007) report exposure ages of 15e17 ka from the Lake Starnberg region around (Ivy-Ochs et al., 2008), whereas Swiss perialpine lakes were ice-free at 20 ka ( Fig. 10H; Heiri et al., 2014). KW2015 data indicate the presence of a polar desert representing the coldest and driest conditions around 25e23 ka, in agreement to authors that suggest LGM glacier maxima during this period (Preusser et al., 2011;Monegato et al., 2017). Chronologically robust, KW2015 indicates comparable paleoenvironmental conditions around 29e30 ka, suggesting a correspondence to earlier Alpine glacier advances, consistent with the evidence from Baumkirchen (Sp€ otl et al., 2013), the Nesseltalgraben (Mayr et al., 2019), and the scheme from Heiri et al. (2014). Taking KW2015 reworked layers as reference, we may assume considerable Alpine glacier advances around 29e30 ka, followed by a retreat, another major advance peaking around 24e25 ka and a final moderate readvance around 20e21 ka. These fluctuations largely correspond to Ca 2þ maxima of the NGRIP dust record (Fischer et al., 2007) and indicate a coupling to hemispheric paleoclimatic evolution with considerable support of glaciations by Heinrich events 3 and 2 and ice volume reductions during GI 4, 3, and 2.
Greenland ice cores do not clearly show the global cooling from MIS 3 to MIS 2 (Lisiecki and Raymo, 2005) but are essential references for the timing of suborbital millennial scale climatic fluctuations, i.e. Dansgaard-Oeschger (DO) and to some extent Heinrich events . Proxy variations of Alpine speleothems ( Fig. 10K; Moseley et al., 2014;Moseley et al., 2020) and Central European LPS (Schirmer, 2012;Moine et al., 2017) indicate a strong link to these millennial-scale paleoclimate variations that appeared largely synchronous on a global scale (Corrick et al., 2020). Tundra gleys at Nussloch that formed between 35 and 20 ka are even larger in number than GI of the same period, which agrees to the information available from KW2015. Nussloch IG5 and KW2015 IVb have no GI analogue in the NGRIP dust and oxygen isotope records (marked with X in Figs. 9 and 10A þ P).
By providing absolute age control for UPG tundra gleys, the chronostratigraphy of Nussloch overcomes circular arguments between climatic cause and paleosol type for the UPG in Central Europe (Moine et al., 2017;Rousseau et al., 2017a), supporting a chronological framework for detailed reconstructions from KW2015 (section 5.5). For the MPG, a TG ¼ GI model requires further age control and regional differentiation. Recognizing a milder MPG and colder UPG for Krems/Willendorf is suggested to reconstruct stratigraphic features of last glacial LPS. Our results underline the coupled significance of local topography, regional paleoenvironment, catchment morphodynamics, and large scale paleoclimate in interpreting marked differences in local loess stratigraphies. Additional studies of spatially distributed LPS are required in order to regionalize past climate changes and to understand how local and catchment environments respond to these changes. These reconstructions require detailed stratigraphic information from field logging supported by high-resolution color and GS data and a careful use of paleoenvironmental proxies.

Conclusions
A robust stratigraphy for the LPS Krems-Wachtberg has been produced by detailed field observations and supported by spectrophotometric data. Color variations are sensitive to changes in pigmentation, which is mostly related to in situ pedogenesis or to sediments pigmented by pedogenesis. Designations of pedological horizons adequately capture the stratigraphic variability of the studied KW2015 section. High-resolution granulometry data differentiate aeolian dust from units affected by pedogenesis and reworking along the slope. Single parameters derived from GS, color or MS are unable to describe the formation of the profile and the paleoenvironmental evolution. Reconstructions of the interplay of dust accumulation, pedogenesis, and slope processes based on the stratigraphic log can be semi-quantitatively linked to paleoecosystems. Published luminescence ages indicate no major discontinuities and radiocarbon ages are important chronological tie points with considerably smaller errors. Chronostratigraphic interpretations strongly benefit from similarities of KW2015 to the tightly dated high-resolution MPG/UPG reference LPS of Nussloch, Germany.
The reconstructions of the forming processes of the LPS Krems-Wachtberg indicate the general climatic deterioration from MIS 3 to MIS 2 and close links to suborbital paleoclimatic variations recorded in Greenland ice core oxygen isotope and dust records. Paleosols that formed under oxidizing conditions are only present in the lower part of KW2015 and correspond to the Lohne soil at the LPS Nussloch, marking the rather abrupt end of the MPG around c. 34e35 ka. The UPG sequence is subdivided by tundra gleys indicating several short phases of climatic improvement, higher in number than GI. This confirms observations from West Central Europe, where the NGRIP dust record (Ca 2þ ) appears as more appropriate reference for characterizing the paleoclimatic oscillations during the last glacial. A characteristic deviation from Western Central Europe is the presence of reworked units that we attribute to the local prevalence of a polar desert during phases of maximum dust accumulation recorded in the NGRIP dust record and the LPS Nussloch. This deviation in local paleoenvironmental conditions likely relates to the more continental location of the study region and the upstream Eastern Alps. This study highlights the need for detailed reconstructions from LPS to regionalize paleoclimate and understand the impact of these changes to the landscape system. A detailed stratigraphy and thorough reconstructions of LPS development is prerequisite for using downprofile parameter variations as paleoenvironmental proxies.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.