IMPROVEMENTS TO THE CLASSIFICATION PERFORMANCE OF RDA

Regularized discriminant analysis has proven to be a most effective classifier for problems wheretraditional classifiers fail because of a lack of sufficient training samples,as is often the case in high-dimensional settings.However,it has been shown that the model selection procedure of regularizeddiscriminant analysis,determining the degree of regularization,has some deficiencies associated with it.We propose a modified model selection procedure based on a new appreciation function.By means ofan extensive simulation it was shown that the new model selection procedure performs better than theoriginal one.We also propose that one of the control parameters of regularized discriminant analysis beallowed to take on negative values.This extension leads to an improved performance in certain situations.The results are confirmed using two chemical data sets.     

The surface geometry of the Last Glacial Maximum ice sheet in the Andøya-Skånland region, northern Norway, constrained by surface exposure dating and clay mineralogy

Blockfields, weathering boundaries and marginal moraines have been mapped along a longitudinal transect from northern Andøya to Skånland in northern Norway. The degree of rock-surface weathering above and below glacial trimlines, clay-mineral assemblages and surface exposure dating based on in situ cosmogenic ¹⁰Be have been used to reconstruct the vertical dimensions and timing of the Last Glacial Maximum (LGM) of the Scandinavian Ice Sheet in this region. The cosmogenic exposure dates suggest that the lower blockfield boundary/trimline along the Andøya-Skånland transect represents the upper limit of the Late Weichselian ice sheet, with an average surface gradient of c . 9.5 m/km. The surface exposure dates from Andøya pre-date the LGM, suggesting that the LGM ice sheet did not reach mountain plateaux at northwest Andøya. The results thus support evidence from lake sediment records that the northern tip of Andøya was not covered by the Scandinavian Ice Sheet during the LGM.     

Spatial and temporal stability of the climatic signal in northern Fennoscandian pine tree-ring width and maximum density

The spatial and temporal stability of the climatic signal in pine tree - ring width and maximum density is examined using data from four sites in northern Sweden and Finland. Moving-window multiple linear regression, using monthly and daily climate data, indicates that ring widths at all sites have been strongly controlled by July temperatures throughout the past century. The relationship between maximum density and temperature is stronger but much less stable across both space and time. Shifts in the hottest part of the summer do not explain large shifts in the period most strongly influencing density. It is concluded that palaeoclimate reconstructions based on northern Fennoscandian pine tree - ring width chronologies should be restricted to the temperature of midsummer (July), whereas maximum density should be used to reconstruct the temperature of a longer growth season (June to August). They thus provide different and compl e mentary palaeoclimate signals. At all four sites, the correlation between maximum density and June to August mean temperature is lowest in the latter half of the 20th century, but split sample tests with strong verification statistics (RE and CE) show that this represents a quantitative change in the strength of the correlation with climate, rather than a qualitative change in the nature of that relationship, and thus does not invalidate climate reconstructions.     

The vertical extent of ice sheets in Nordfjord, western Norway: measuring degree of rock surface weathering

Degree of rock surface weathering was measured on sites in Oldedalen and Brigsdalen, where dates of deglaciation have been estimated. and on an altitudinal transect on the slopes of Skåla. representing one of the highest supra-marine reliefs in western Norway. The Schmidt hammer is useful only for distinguishing sites deglaciated during the Little Ice Age from those deglaciated during the Lateglacial and early Holocene. Degree of roughness of granitic augen gneiss bedrock surfaces was quantified from profiles measured in situ using a micro-roughness-meter and profile gauge. There is a significant increase in surface roughness above a clear trimline at c. 1350 m a.s.I. but no significant increase above a higher trimline previously proposed as the vertical limit of the last ice sheet in this area (c. 1560 m a.s.I.). The roughness of boulder surfaces on the summit blockfield does not direr significantly from the roughness of bedrock surfaces downslope as far as the lower trimline. These unexpected results suggest that bedrock surfaces between the two trimlines were not glacially abraded during the Late Weichselian, so that the upper trimline is unlikely to represent the vertical limit of ice during either the Late Weichselian or a subsequent readvance. Preliminary results of ¹⁰Be dating of surface quartz samples from above the lower trimline support the proposal that the site was not abraded during the last glaciation. The results can be interpreted in two ways: (1) The upper trimline represents the vertical limit of a pre-Late Weichselian advance. During the Late Weichselian the mountains were completely covered but surfaces down to the lower trimline were protected by cold-based ice. (2) The lower trimline marks the vertical limit of the Late Weichselian ice and the upper limit an older and more extensive glaciation.     

Nunataks of the last ice sheet in northwest Scotland

High-level weathering limits separating ice-scoured topography from an upper zone of frost-weathered detritus were identified on 17 mountains in NW Scotland at altitudes of <600 m to< 900 m, and a further 6 peaks were found to support evidence of ice scouring to summit level. Weathering limits are most clearly defined on Torridon Sandstone, which is resistant to frost shattering, but can also be mapped on Cambrian Quartzite, Lewisian Gneiss and Moine Schist. Contrasts in degree of rock surface weathering above and below the weathering limits were evaluated using measurements of joint depth and rock surface hardness, and through X-ray diffraction analyses of clay mineral assemblages. The results indicate significantly more advanced rock and soil weathering above the weathering limits. Widespread persistence of gibbsite above the weathering limits suggests that they represent the upper limit of Late Devensian glacial erosion, and the regularity of the decline in weathering limit altitude along former flowlines eliminates the possibility that it represents a former thermal boundary between protective cold-based and erosive warm-based ice. The weathering limits are therefore interpreted as periglacial trimlines defining the maximum surface altitude of the last ice sheet around former nunataks. Calculated basal shear stresses of 50–78 kPa are consistent with this interpretation. The altitude of the trimlines implies that the former ice shed lay at 900–930 m in the Fannich Mountains and descended gently northwards, and that the ice surface descended NW from the ice shed to >500 m over the extreme NW tip of Scotland and to 700–730 m at the head of Little Loch Broom.     

THE LAST ICE SHEET IN NORTH-WEST SCOTLAND: RECONSTRUCTION AND IMPLICATIONS

Recent models of the last Scottish ice sheet suggest that nunataks remained above the ice surface in areas peripheral to the main centres of accumulation. This proposition has been investigated on 140 mountains over an area of 10,000 km² in NW Scotland. Outside the limits of the later Loch Lomond Readvance in this area there is evidence for a single high-level weathering limit that separates glacially eroded terrain from higher areas of in situ frost debris. This limit occurs at altitudes ranging from 425 to 450 m in the Outer Hebrides to >950 m on the mainland, and is best developed on lithologies that resisted breakdown after ice-sheet downwastage. Interpretation of this weathering limit as a periglacial trimline cut by the last ice sheet at its maximum thickness is supported by: (1) joint-depth and Schmidt hammer measurements that indicate significantly more advanced rock breakdown above the weathering limit; (2) a much greater representation of gibbsite (a pre-Late Devensian weathering product) in the clay fraction of soils above the limit; (3) cosmogenic isotope dating of the exposure ages of rock outcrops above and below the limit; (4) the sharpness of the limit at some sites and its regular decline along former ice flowlines; and (5) shear stress calculations based on the inferred altitude and gradient of the former ice surface. Reconstruction of the ice surface based on trimline evidence indicates that the mainland ice shed lay near or slightly east of the present watershed and descended northwards from >900 m to ca. 550 m at the north coast. Independent dispersion centres fed broad ice streams that occupied major troughs. On Skye an ice dome >800 m deflected the northwestwards movement of mainland ice, but the mountains of Rum were over-ridden by mainland ice up to an altitude of ca. 700 m. The Outer Hebrides supported an independent ice cap that was confluent with mainland ice in the Minches. Extrapolation of the trimline evidence indicates that most reconstructions of ice extent are too conservative, and suggests that low-gradient ice streams extended across the Hebridean Shelf offshore. Wider implications of this research are: (1) that blockfields and other periglacial weathering covers are not all of the same age or significance, depending on the resistance of different lithologies to frost weathering; (2) that the contrasting degree of glacial modification in the Western and Eastern Highlands of Scotland may reflect a former cover of predominantly warm-based ice in the former and predominantly cold-based ice in the latter; and (3) that the approach and techniques developed in this study have potential application for constraining ice-sheet models, not only in areas peripheral to the main centres of ice accumulation in Britain and Ireland, but also in other mountain areas where nunataks protruded through warm-based Late Pleistocene ice masses.     

Contemporary terminal-moraine ridge formation at a temperate glacier: Styggedalsbreen, Jotunheimen, southern Norway

Terminal-moraine ridges up to 6 m high have been forming at the snout of Styggedalsbreen for two decades. Based on intermittent observations during this period, combined with a detailed study of ridge morphology, sedimentary structures and composition during the 1993 field season, a model of terminal-moraine formation that involves the interaction of glacial and glacio-fluvial processes at a seasonally oscillating ice margin is presented. In winter, subglacial debris is frozen-on to the glacier sole; in summer, ice-marginal and supraglacial streams deposit sediments on the wasting ice tongue. The ice tongue overrides an embryonic moraine ridge during a late-winter advance and a double layer of sediment (diamicton overlain by sorted sands and gravels) is added to the moraine ridge during the subsequent ablation season. Particular ridges grow incrementally over many years and exert positive feedback by enhancing snout up-arching during the winter advance and constraining the course of summer meltwater streams close to the ice margin. The double-layer annual-meltout model is related to moraine formation by the stacking of subglacial frozen-on sediment slabs (Krüger 1993). Moraine ridges of this type have a complex origin. are not push moraines, and may be characteristic of dynamic high-latitude and high-altitude temperate glaciers.     

Degree of rock surface weathering on fjell summits in northern Finland: implications for the thermal regime of the last ice sheet

Measures of degree of rock surface weathering on three fjell summits in Finnish Lapland provide an independent test of the extent of continuously cold-based ice during the Late Weichselian. The surface hardness of exposed surfaces is compared with that of surfaces that have remained covered and protected from subaerial weathering. Roughness measurements from granite gneiss surfaces are compared with results from similar lithologies that are known to have been eroded during the Late Weichselian. The results suggest that the summits have been covered and actively eroded by warm-based ice and are difficult to reconcile with suggestions that large parts of Fennoscandia have not been glacially abraded since the Early Weichselian.     

The age and origin of Neoglacial moraines in Jotunheimen, southern Norway: new evidence from weathering-based data

This paper considers the controversial issue of the existence of pre-'Little Ice Age' Neoglacial moraines in southern Norway. Schmidt hammer rebound values are combined with measures of boulder roundness and weathering rind thickness in an attempt to isolate moraines that include weathered boulders. A critical approach is used in distinguishing sites where boulders have weathered in situ from those where previously weathered clasts have been incorporated into relatively young moraines. The results confirm that possible pre-'Little Ice Age' Neoglacial moraines seem to be restricted to small, high-altitude glaciers in eastern Jotunheimen. It is concluded that at these glaciers a particularly large response to a short-lived earlier Holocene climatic event is more likely to explain the survival of such moraines than a particularly subdued response to the climatic deterioration of the 'Little Ice Age'. More refined dating techniques are required to determine the age of formation of the anomalous moraines, but before the palaeoclimatic significance of such dates can be assessed, a critical test is required to establish whether the moraines mark former ice-front positions, and therefore reflect lowering of equilibrium line altitudes, or whether they have been displaced forwards by later and more extensive glacier advances.