Creation of an atlas of lake-sediment geochemistry of Western Labrador and Northeastern Québec

A geochemical database of 21 678 lake-sediment samples, analysed for 53 elements by ICP-MS after aqua-regia digestion, has been created. Statistical tests and plots of the combined data (in some cases, after the application of levelling functions) show that the disharmony that existed previously – between analyses that employed digestions of differing strengths – can be largely alleviated. This allows the creation of integrated geochemical maps of an area of almost 300 000 km2. Many of the large-scale geochemical features revealed by the data are related to the rocks of the Paleoproterozoic ‘Labrador Trough’, within the region termed the New Quebec Orogen. However, those that are defined by an association of Rb, K, Ti, Li, Cs, Mg, Ga, Sn and Th, correlated negatively with LOI, Hg, Se, S, Ag, Cd, Ca and Sr may be related to the relative amounts of organic and inorganic clastic material in the lake sediment, rather than bedrock geology. It is hoped that the merged data will provide assistance to the geological mapping of this large area, as well as providing a stimulus for mineral exploration.

The creation of geochemical maps of large areas of the Earth's surface is inevitably faced with the problem that different jurisdictions apply different analytical protocols; in particular, in the digestion methods that precede spectroscopic or other 'wet-chemical' analysis. This is the challenge to the merging of lake-sediment geochemical data from Labrador and northeastern Québec. The Labrador region (with the exception of the mountainous region north of 58°30′ N, where stream sediment was preferred as a medium) was sampled in its entirety as part of the National Geochemical Reconnaissance (NGR) program of the Geological Survey of Canada (GSC; Boyle et al. 1981;Hornbrook et al. 1983;Friske & Hornbrook 1991;Friske et al. 1993aFriske et al. , b, c, d, 1994. Some additional elements were later analysed by the Geological Survey of Newfoundland and Labrador (GSNL; McConnell & Finch 2012). The Québec samples were collected during a series of campaigns (Beaumier 1982(Beaumier , 1983(Beaumier , 1984a(Beaumier , b, 1985a(Beaumier , b, 1986a(Beaumier , b, c, 1987Madore et al. 1999) and employed sampling, sample-preparation and analytical protocols that were similar, but not identical, to those of the GSC.
To create a harmonized dataset with the aim of delineating largescale geochemical patterns, and assisting in geological mapping and mineral exploration in the region, c. 5000 lake-sediment samples from Labrador were reanalysed by induction-coupled mass spectrometry (ICP-MS) after digestion by a modified aqua-regia (1 HCl/1 HNO 3 ) digestion. This follows the protocol applied to the adjacent samples from Québec: a 0.5 g sample is weighed into a test tube for sample digestion. A modified aqua-regia solution of equal parts concentrated ACS grade HCl and HNO 3 and demineralized water is added to each sample (6 ml/g) to leach in a hot-water bath (c. 95°C) for one hour. After cooling, the solution is made up to a final volume with 5% HCl. Sample weight to solution volume ratio is 0.5 g per 10 ml.
This work is based on results from the Geological Survey of Canada's Geo-mapping for Energy and Minerals Program, Phase 2 (GEM-2) Hudson-Ungava Project -Core Zone Surficial Activity, focused on northeastern Quebec and west central Labrador (see McClenaghan et al. 2017). The GEM program (2013-2020 provides modern public geoscience that will set the stage for long-term decision making related to investment in responsible resource development. Geoscience knowledge produced by GEM supports evidence-based exploration for new energy and mineral resources and enables northern communities to make informed decisions about their land, economy and society.
The first part of this paper describes the study area, database and methods of sample preparation and analysis. Descriptions of the regional geology, physiography and data-processing methods are followed by a description of the problems created by mixing analyses performed after different digestion methods. The verification that the data from Québec and the reanalysed Labrador data are a good match, using nearest-neighbour comparisons, is followed by a description of the levelling applied to certain elements that still exhibited discontinuities when plotted, even within a supposedly homogeneous dataset. After a brief description of the element associations, the areal distributions of certain elements are described and compared with respect to certain large-scale geological and topographic features.

Study area
The study area covers about 215 000 km 2 in northeastern Québec and 85 000 km 2 in northern and western Labrador. It is bounded to the east by the Labrador Sea coastline, with the exception of the SE which is bounded by longitude 62°W; to the south by latitude 54°N, except in the west where the southern boundary of coverage is 57°N ; to the west by longitude 70°W in the south, and 72°W in the north; and to the north by latitude 59°N in the west, and by the Ungava Bay coastline in the east ( Fig. 1A and B).
Each sheet is 1°of latitude high and 2°of longitude wide, with an approximate area, in the region of study, of 14 000 km 2 .
The total number of Québec samples collected within these map areas is 24 261, while the Labrador subset comprises 5510 samples.
The samples were collected in a series of helicopter-supported campaigns between 1979 and 1997. Although a sample density of 6-8 samples per 100 km 2 was maintained over most of the study area in both Labrador and Québec, the sample density over the Québec portion of the Labrador Trough, south of 57°30′ N, was four times as high. To create a dataset of more evenly spaced samples, a subset of 25% of the samples from this latter region was randomly selected for the current study. Sub-selection was done independently within each NTS 1: 50 000 map sheet (of approximate area 875 km 2 ), to avoid regions of over-or under-selection, or 'clumping', but the resulting coverage is still somewhat uneven compared to the remainder of the study area. This resulted in a reduction in the total number of Québec analyses from 24 261 to 16 167. Maps showing the sample density, before and after this adjustment, are presented by Amor (2015).
The raw Labrador data are available as two GSC open file reports (McCurdy 2016;McCurdy et al. 2016) and the raw Québec data can be downloaded from https://sigeom.mines.gouv.qc.ca. Smoothed data were used to create the geochemical maps in this paper (see below); they are available on request from the senior author.

Sample preparation
Sample-preparation methods for the NGR samples are described by Friske & Hornbrook (1991) as follows: when air-drying at ambient temperatures was complete, the samples were crushed or broken down to pieces less than 0.5 cm in diameter, loaded into ceramic ball mills and shaken for 45 min. The minus-80 mesh (<177 µm) fraction of this crushed material was used for subsequent analysis and archived in polyethylene containers. The procedure to which the Québec samples were subjected was similar, except that the drying temperature was 45°C (Beaumier 1985b).

Analysis
Originally, the Labrador samples were analysed for a limited suite of elements by atomic absorption spectrophotometry (AAS) and certain other specialized methods, after reverse aqua-regia (Lefort acid) digestion, and subsequently by instrumental neutron-activation analysis (INAA); both these analytical programs were carried out under the auspices of the Geological Survey of Canada (GSC). The Labrador samples have also been subjected to inductioncoupled plasma optical emission spectrometry (ICP-OES) after multi-acid (HF/HCl/HNO 3 /HClO 4 ) digestion at the laboratory of the Geological Survey of Newfoundland and Labrador (GSNL;McConnell & Finch 2012). The Québec samples have been subjected to a variety of analytical techniques, performed by the provincial Ministry of Energy and Natural Resources; however, it is only the samples analysed by ICP-MS after a modified aqua-regia digestion (see above) by Acme Analytical Laboratories in Vancouver (Maurice & Labbé 2009) that are discussed in this paper.
The differing digestion methods employed by the two government organizations cause problems in direct comparison of results, and make combination of existing datasets problematical. However, the seriousness of the problem varies from element to element.
To harmonize the data and facilitate the creation of a single, atlasscale dataset, a subset consisting of 5510 Labrador lake-sediment samples was recovered from the archive. The samples were reanalysed at the same commercial laboratory (Bureau Veritas, formerly ACME Labs, Vancouver), by the same method (ICP-MS), for the same elements, and after the same digestion method (modified aqua regia), as their Québec counterparts (McCurdy 2016;McCurdy et al. 2016).
The entire geochemical database for the combined regions prior to the current study is described in Table 1. In the current study there is general correspondence between the Québec and Labrador programs in the analysed elements and their lower detection limits (Table 2), although the detection limits for Au and Pt in the Québec dataset are both 1 ppb, compared to 0.2 ppb and 2 ppb in the new Labrador dataset. Also, the Labrador samples were analysed All of the indicated NTS sheets were sampled in their entirety except 14L, which was only sampled south of 58°30′N, 24P and 25A which were only sampled west of the provincial border. Mineral deposits discussed in the text are also indicated. additionally for Dy, Er, Eu, Gd, Ho, Lu, Nd, Pr, Sm, Tb, Tm and Yb, which are omitted from the study since there are no corresponding data for the Québec samples.

Geology
The geochemical atlas compiled by McCurdy et al. (2018) includes a geological map of the study area. The following description was written by D. Corrigan and is taken, with minor modifications, from Amor et al. (2016).
The area covered by the lake-sediment sampling is entirely underlain by Precambrian rocks of the Canadian Shield (Hoffman 1990;Wardle & Van Kranendonk 1996). It consists of a major zone of diachronous accretion and collision between three Archean-age crustal blocks: (1) the Superior Craton (Archean); (2) the Core Zone (mostly Archean with earliest-Paleoproterozoic crust); and, (3) the North Atlantic Craton (Archean), and intervening Paleoproterozoic-age supracrustal sequences and magmatic arcs (Fig. 2).
The boundary between the Core Zone and North Atlantic Craton is the Torngat Orogen, a zone of high metamorphic grade and ductile, mostly dextral transcurrent shear formed at c. 1.87-1.84 Ga (Wardle et al. 2002). The Torngat Orogen preserves an accretionary prism dominated by metapelites (Tasiuyak Gneiss) and the tectonically exhumed deep root of a c. 1.87-1.86 Ga magmatic arc (Lac Lomier Complex). The Burwell Domain, exposed on the northern tip of Labrador, consists of c. 1.89 Ga tonalites and charnockites emplaced in Archean-age crust (Scott & Machado 1995). On its western side, the Core Zone is separated from the Superior Craton by the c. 1.83-1.79 Ga New Québec Orogen (Clark & Wares 2005). This predominantly transpressional orogen consists of greenschist-to lower-amphibolite facies clastic and chemical sedimentary rocks representing: (1) Autochthonous rift-to-drift sequences, as well as tectonically overlying flysch, banded iron-formation and molasse sediments interlayered with minor volcanic rocks (Kaniapiscau Supergroup); (2) Belts composed predominantly of mafic volcanic rocks and sills interlayered with sediments comprised mostly of meta-siltstone and black shale (Baby-Howse and Doublet domains) with the Doublet interpreted as remnant of a back-arc basin (Rohon et al. 1993); and, (3) Mid-to upper-amphibolite facies, Paleoproterozoic-age clastic metasedimentary and minor mafic metavolcanic rocks that are in thrust contact with the Baby-Howse and Doublet zones in basement-involved thick-skinned tectonics (Rachel-Laporte zone).
Voluminous c. 1.88 Ga metagabbro sills of the Montagnais suite intrude rocks of the Baby-Howse and Doublet zones; this assemblage makes up what is generally referred to as the 'Labrador Trough' (e.g. Greene 1974). Based on continuity of regional aeromagnetic patterns, Archean crust occurring in basement windows and east of the Labrador Trough is interpreted to have Superior Craton affinity; that hypothesis, however, remains to be tested. The Core Zone is also host to a c. 1.84-1.81 Ga continental magmatic arc, the De Pas Batholith (Dunphy & Skulski 1996).  Table 3 shows the relationship of the major domains within the Labrador Trough to rock formations as identified by McCurdy et al. (2018), which will be referred to later.
The southeastern portion of the map presented in Figure 2 is intruded by voluminous Mesoproterozoic-age plutons, including the Nain Plutonic Suite, emplaced between c. 1.46 and 1.29 Ga (Emslie & Stirling 1993;Amelin et al. 1999). The latter are economically important, being host to the Voisey's Bay magmatic Ni-Cu deposit, as well as the Strange Lake rare-earth element (REE) and rare-metal (RM) deposit.
The southernmost extent of the lake sediment survey, east of the Smallwood Reservoir, covers the Seal Lake Group, a set of c. 1.25 Ga volcano-sedimentary sequences with low-grade metamorphic overprint that contains numerous Cu occurrences (van Nostrand & Corcoran 2013). The study area is bounded to the south by the Grenville Orogen.

Metallic mineral deposits
The Sokoman Formation of the Kaniapiscau Supergroup is host to a major iron-ore district which hosts (at time of writing) one producer in Québec (Goodwood) and three in Labrador (Kivivic, Howse and Timmins #4), 24 former producers, 78 developed prospects (described as 'tonnage evaluated' in the Québec mineral-occurrence database) and 36 prospects ('worked deposits' in Québec).
The only other current producer is the Voisey's Bay Ni/Cu mine on the Labrador coast, which is an orthomagmatic sulphide deposit associated with troctolite and gabbro. However, the study area also hosts 34 developed prospects and 167 prospects of Au, Cu, Ni, U, Ti, V, Ag, Pb, Zn, PGE and REE/RM. Information regarding the Québec and Labrador mineral deposits was sourced from SIGÉOM à la carte (https://sigeom.mines.gouv.qc.ca) and the Geoscience Atlas of Newfoundland and Labrador (geoatlas.gov.nl.ca), respectively.

Physiography
Most of the major physiographic features trend northwards or northnortheastwards and consist, from east to west, of the Torngat Mountains, the George Plateau, the Whale Lowland, the Labrador Hills and the Kaniapiskau Plateau. Labrador's Red Wine Mountains, in the SE corner of the study area, are a component of the NE-trending Hamilton Upland (Grant & Sanford 1977).
Relief in the study area varies from flat and swampy over much of the central part, to very high over the Torngat Mountains, which occupy most of the Labrador portion east of 65°W, and the adjacent part of Québec. Areas of local high relief in Québec are present over parts of the Labrador Trough, and in the valleys of the Leaf (aux Feuilles), Larch (aux Mélezes), Caniaupiscau and Koksoak Rivers (Fig. 3). Altitude ranges from sea level to 1652 m above sea level at Mount Caubvick (known as Mont D'Iberville in Québec) on the Québec/Labrador border at 58°53′ N, 63°43′ W.
The predominant mapped surficial unit (Klassen et al. 1992) is present to a greater or lesser extent over the entire study area with the exception of the SE (over the Harp Lake Intrusive Suite; Emslie 1980), the Ungava Bay coastline and Labrador coastline north of 56°N, the valley of the Koksoak River and parts of the Torngat Mountains. It consists predominantly of silty to sandy diamicton, although many other sediment types are also present locally. Rogen moraine is well developed over the Labrador Hills and northern Whale Lowland, the area straddling the Québec/Labrador border between latitudes 56°45′ and 54°45′ N, and west of the Smallwood reservoir. Ablation drift, formed during disintegration of the ice sheet, is primarily restricted to the eastern side of the Torngat Mountains. Drumlins tend to be most abundant in areas of relatively low relief and are absent over the Torngat Mountains and the Areas with greater than 80% bedrock exposure, and with till cover mostly less than 1 m thick, are also widespread, and almost universal over the Torngat Mountains north of 56°45′ N and over the Harp Lake Suite. On the other hand, outcrop is scarce in the Whale Lowland, over the Labrador Trough south of 55°30′ and NE of the Smallwood Reservoir.
There is a significant concentration of eskers either side of the interprovincial border north of 56°N, in the north of the Whale Lowland, and in Labrador west of the Smallwood Reservoir, (Cummings et al. 2011). Glaciofluvial deposits (ice-contact fans and deltas, outwash plains and terraces, kames, kame terraces, etc.) are particularly abundant in the southern Torngat Mountains and over the Labrador Trough, and scarce in the Whale Lowland. Glaciomarine and marine deposits (gravel, sand and mud), forming thick blanket accumulations, are present intermittently along the Labrador coast but very extensive along the Ungava Bay coast within the study area, and extend many tens of kilometres inland in the valleys of the rivers draining into the bay.

Nearest-neighbour regression
Nearest-neighbour regression is a method of geochemical levelling (as defined by Darnley et al. 1995, pp. 75-78). This approach to comparing the two datasets is necessary when duplicate analyses of exactly the same samples (at the two laboratories where the analyses from the two territories were performed) are not available for comparison. The current study has involved the selection of pairs of samples, one from each of the Québec and Labrador datasets, separated by a small enough distance that their compositions, if measured by the same method, can be expected to be closely related. Various statistical tests can be carried out on the resulting pairs and a regression carried out to relate them. If the regression statistics indicate significant correlation then results can be used to convert one set of values to another so that the match between the two datasets is improved.
This method was applied to establish similarities and differences between the samples disposed either side of the border between the two territories, and to derive a levelling function where necessary (for Ag). It was also used to alleviate discontinuities within the Québec dataset itself for Cd and Hg, caused by the analysis of samples from projects completed in different years.   Fig. 3. Relief map of study area (defined as maximum-minimum contours intersected in each 10-km square cell).

Quantile regression
In the method of quantile regression, as applied to the levelling of geochemical data, a regression equation is derived between the quantiles of a particular regionalized variable for two adjacent datasets. Its application has been described by Darnley et al. 1995 andDaneshfar &Cameron (1998). This method was applied to the alleviation of an inter-project discontinuity in Zn values within the Québec dataset where nearest-neighbour regression did not produce satisfactory results.

Moving-median smoothing
To extract the 'signal' from what may be rather 'noisy' geochemical data, a 'moving-median' smoothing algorithm was applied. This is similar to that described by Gustavsson et al. (2001), whereby the value plotted at each sample point is the median value of all the samples that fall within a fixed radius of that sample. This has the  effect of smoothing the data but preserves the original sample locations in the resulting plots at the same time. A radius of 10 km proved appropriate for the current study, although very broad regional-scale changes become apparent if larger radii are employed. An example of the application of this smoothing method, as applied to Se data, is shown in Figure 4A and B.

Relationship between aqua-regia and multi-acid digestion
A comprehensive comparison of the analytical results, following multi-acid and modified aqua-regia digestion, in a subset of 1869 Québec samples was reported by Amor (2015). Cobalt provides an example of an element for which the modified aqua-regia recovery is close to 100%, and where no discontinuity is apparent at the boundary between the surveys. In contrast, Ba provides an example where the relationship between the two digestions is poor and where there is a pronounced discontinuity ( Fig. 5A and B).

Results of nearest-neighbour comparison
Sample pairs were selected for comparison and (where necessary) regression analysis if the distance separating them across the interprovincial border was less than 5 km. In view of the differing behaviour exhibited by organic-rich and organic-poor lake sediments (e.g. Garrett et al. 1990), the nearest-neighbour pairs were separated into high-LOI and low-LOI groups. The cut-off between the two groups was identified by examining the relationship of a variety of elements, with increasing LOI content, in the Québec and Labrador samples. Examples of this relationship are illustrated with boxplots in Figure 6. In many (but not all) cases there is either a local maximum, or an inflection point, at c. 18.4% LOI. This corresponds closely to a cut-off arrived at similarly by Trépanier (2007), on lake sediments from the north of the Québec study area only. The areas of high and low LOI, thus defined, are shown in Figure 7.  Relationships were investigated both between nearest neighbours across the territorial boundary and, for comparison, between nearest neighbours within the Québec dataset alone. The strength of the relationships has been quantified and compared by means of the Spearman rank correlation coefficient. These correlation coefficients are summarized in Table 4, and displayed graphically in Figure 8A and B. Figure 8A deals with the internal nearestneighbour correlation coefficients alone, and correlation coefficients for the low-LOI sample pairs are plotted against those for the high-LOI pairs; Differences in correlation coefficients between the low-LOI and high-LOI groups would manifest themselves as the majority of points, each representing an element, being disposed to one side or the other of the 1:1 line. No conspicuous difference between the groups is apparent. In Figure 8B, the internal nearest-neighbour correlation coefficients are plotted against their crossborder counterparts. Again, if the correlation coefficients were conspicuously stronger internally than across the border, then the pairing of nearest neighbours (at least, those separated by 5 km or less) would be invalid for the purposes of comparing the Québec and Labrador datasets; this is not the case.
Examples of the relationships between nearest neighbours' analyses for Sr in high-LOI samples, and Ca in low-LOI samples are shown graphically in Figure 9A and B.
Overall, although the correlation between cross-border nearest neighbours is not consistently strong, it is not significantly weaker than the correlation between internal nearest neighbours; it is, therefore, reasonable to compare the two.  Table 3); correlations for the two sediment types are not conspicuously different. (b) Correlation coefficients for Québec internal nearest neighbours plotted against cross-border nearest neigbours, all with maximum separation of 5 km. Each point represents a different element (See Table 3); correlation coefficients for internal nearest neighbours are slightly higher than for cross-border nearest neighbours. Open squares: LOI ≤ 18.4%; closed squares: LOI > 18.4%  Results for all pairs that pass the paired t-test analyses are marked in bold. The numbers in brackets following the Pass or Fail indication is the average difference between the pairs in log 10 units. The additional number for the all pairs column is the probability associated with the computed t estimate.Elements where more than 10% of values were less than the detection limit are omitted. Results for all pairs that pass the paired t-test analyses are marked in bold. The numbers in brackets following the Pass or Fail indication is the average difference between the pairs in log 10 units. The additional number for the all pairs column is the probability associated with the computed t estimate.Elements where more than 10% of values were less than the detection limit are omitted.

Equivalency of means
Paired t-tests to estimate if the mean differences between the pairs was not significantly different from zero (R-Project 2016), following log 10 transformations, were applied to the cross-border pairs of analyses. More than 10% of the analyses of As, Au, Ge, Hf, In, Re, Sb, Te and W were less than the lower detection limit in all six datasets and are not amenable to statistical calculations of this kind. The results are shown in Tables 5 and 6 An alternate procedure, using the average difference between the pairs in log 10 units is as follows: The average difference is antilogged and the difference from one (equivalent to zero difference between the logarithms) is expressed as a percentage. These figures are shown in the last two columns of Tables 5 and 6. Where the difference is less than or equal to 10%, the entries in the last column are emboldened, and where it lies between 10% and 15% they are italicized. Based on the ≤10% criterion, and as applied to nearest  neighbours separated by 2 km or less, the following elements would be amenable to combination without levelling: Al, Ba, Be, Cd, Cr, Cu. Hg, Mg, Pb, S, Sr, Th, Ti, and U. Relaxing the criterion to ≤15% adds: Ca, Fe, Ga, K, La, Li, Rb, and Y.

Levelling between the datasets
In order to compensate for the mismatch in Ag analyses between the Québec and Labrador datasets (that show the greatest discrepancy) levelling equations were derived by regressing Québec analyses as the dependent variables against Labrador analyses as the independent variables. Two separate levelling equations were derived for high-and low-LOI samples, as defined above. Figure 10A and B show the relationship between Québec and Labrador samples in these two groups, while Figure 11A and B show the areal distribution of Ag values before and after levelling. Despite earlier statistical calculations having indicated that the mismatch between the two groups was greater for Ag than any other element, the perceptible effects of the levelling are very slight. This suggests that re-analysis of the Labrador samples has proven effective, for the most part, in eliminating discontinuities and that repeating this exercise for other elements would not be productive.

Levelling within the Québec dataset
On the other hand, additional nearest-neighbour calculations were necessary within the Québec dataset, to address a conspicuous discontinuity along the 57°30′ N line of latitude; this corresponds to the boundary between samples collected in 1997, and those collected during earlier campaigns, and is probably the result of a slight, inadvertent, calibration shift. The most pronounced discontinuities are displayed by Cd and Zn (where the 1997 values are noticeably higher), and Hg (where they are significantly lower).
Levelling equations were derived by directly regressing pre-1997 analyses against those from the 1997 project, for Cd and Hg. The data were not, in this case, subdivided into subsets of high-and low-LOI samples, because the former greatly outnumber the latter. For Cd, a regression equation derived from all samples within 5 km of the discontinuity provided more satisfactory results, when plotted, than a regression equation derived from a more restricted dataset of pairs separated by less than 2 km. The latter, smaller dataset was, however, suitable for deriving a levelling equation for Hg. The relationships between analyses on either side of the inter-project discontinuity are shown in Figure 12A and B, and plots of the areal distribution of these elements, before and after the application of levelling equations are shown in Figure 13A-D.
A somewhat less pronounced discontinuity is also apparent in Zn values between the 1997 and pre-1997 samples. Nearest-neighbour regression of the individual Zn values, at both the ≤2 and ≤5 km levels, did not produce satisfactory results when the levelled values were plotted; therefore, quantile regression (Darnley et al. 1995;Daneshfar and Cameron 1998) of the Zn analyses for samples within 5 km of the discontinuity was resorted to, with satisfactory results, instead.

Element associations
In the following sections, the areal distribution of certain elements is described and discussed with respect to geology, mineral occurrences and (in some cases) to neither. Maps of the distribution of all of the elements are included in McCurdy et al. (2018). Figure 14 is a correlation matrix for the post-levelling dataset as a whole. It was once again necessary to omit As, Au, Bi, Ge, Hf, In, Mo, Re, Sb, Te and W, as more than 10% of the values of these elements are below the analytical detection limit. The data were subjected to the isometric symmetric coordinate transformation (Reimann et al. 2017) to overcome the effects of constant sum.
The elements in the correlation matrix are listed in an appropriate order for the strongest correlation coefficients to be grouped together. The strongest association is between Rb, K, Ti, Li, Cs, Mg, Ga, Sn and Th, all of which also display somewhat weaker negative correlation with LOI, Hg, Se, S, Ag, Cd, Ca and Sr; these latter elements are positively correlated with one another. Although it is hard to envisage a source rock for sediments in which alkali metals like Rb and K are associated with elements like Ti and Mg, an association of the first six elements (as well as Ba, Hf, Na and, surprisingly, Sr, which displayed opposite behaviour to that in the current study) has been described by Amor (2014) in lake sediments for the whole of Labrador. However, the analyses in that study were of total-element content and analyses of Ga and Sn were not performed. The element association in the Labrador sediments was postulated to be related to the relative content of inorganic clastic material in the sediment. If inorganic clastic material is dominant in the signal, the response of these elements to variations in bedrock composition is likely to be muted. Other element associations consist of La-Ce-Y-U; Cu-Ni; and Fe-Mn.

Relation of lake-sediment composition to mapped rocktypes
The lake sediments have been divided up according to underlying rock units, as described in Table 3 and presented by McCurdy et al. (2018). Box plots of selected elements are shown in Figure 15A-J and Tables 7-11 summarize the distinguishing characteristics of mapped rocktypes in general and in various subdivisions of the more widespread rocktypes. The following general associations (or lack of them) are apparent: • Sediments collected over anorthosite (Table 7) are characteristically low in many elements (As, Ba, Bi, Cr, Cs, Cu, Fe, Hg, K, Li, Mg, Mn, Ni, Pb, Rb, Se, Th, Tl, U, Zn, Zr) and high in only one (Na). • Ultramafic rocks (Table 7) are characterized by enrichment in the overlying lake sediments in Ag, Al, Au (Fig. 15A), Co,  (Table 7) is not characterized by Fe enrichment in the overlying lake sediments; with the exception of apparent depletion in Ti, they are geochemically rather bland. • Felsic volcanic rocks (Table 7) show an apparent overall enrichment in Be, Ce (Fig. 15A), La, Mo, Nb, Pb, Y and Zn, and depletion in Co, Cu, Li, Mg and Se. However, the majority of these 17 samples were from areas underlain by REE-enriched peralkaline rhyolite in the Letitia Lake area (NTS 13L/01 and 13L/02; Miller 1988). Therefore, this signature is not necessarily characteristic of lake sediments overlying felsic volcanics generally. • Lake sediments over the Superior Province granites ( Fig. 15C and Table 8) are relatively enriched in Ag. Granites assigned to the Mesoproterozoic Nain Plutonic Suite are elevated in Be, La, Mo (Fig. 15D), P, and Y, all of which suggest REE mineralization (which is indeed present), as well as W. • Samples collected over the Montagnais gabbro (Table 9) are elevated in Ag, As (Fig. 15E), Bi, Cd, Cr, Cu, Fe, Hg, Ni and Sb. Those collected over the gabbro and ferrodiorite of the Nain province (Nain Plutonic Suite) show enrichment in elements normally associated with REE mineralization (which may be the consequence of glacial dispersion from the adjacent mineralized granites and syenites), as well as depletion in Mg (Fig. 15F). • Of the basaltic rock units (Table 10) in the Labrador Trough, lake sediments overlying the alkali Nimish basalts are elevated in As, Bi, Cd, Fe and Sb, as well as Hf. Those overlying the Willbob basalts are enriched in Ag, Au, Cr, Cu (Fig. 15G) and Ni and depleted in Ba and Mo. Sediments overlying the Murdoch basalts are enriched in only Cr and Ni (and also depleted in Ba and Mo). Lake sediments overlying the Bacchus basalts are not geochemically distinctive. This may be a consequence of the presence of a clastic-sediment component in this basaltic unit, which is absent or insignificant in the others. The Tunulik/Orma basalts have lake-sediment signatures that suggest a contribution by material derived from granite (likely due to glacial displacement). • Sedimentary rocks of the Sakami Formation (Table 11), which overly rocks of the Archean Superior province but are lower Proterozoic in age, are characterized by lake sediments elevated in Ba, Cs, K, Li, Sn, Sr, Th, Tl and W, but not in U, despite the presence of uranium occurrences (McCurdy et al. 2018). • Other than ultramafic rocks in general ( Fig. 15A and Table 7), the only rock units demonstrating significant Au enrichment in the overlying lake sediments are ultramafic rocks in general ( Fig. 15A and Table 7); the Willbob mafic volcanics ( Fig. 15G and Table 10) and the Tasuiyuak metasedimentary gneiss (Fig. 15I and Table 10). • Arkose and quartzite of the Seal Lake Group are characterized by depletion in Al, Bi, Co, Cr, Cs, Cu, Fe (Fig. 15J), K, Li, Mg, Ni, Pb, Rb, Sc, Th, Tl and Zr, and enrichment in Sr alone.

Areal distribution of selected elements
All of the data plotted in the geochemical maps described in the following section consist of analyses that have been subjected to the moving-median smoothing process described above. For most elements, the intervals are represented by five colours (from lowest to highest: greycyangreenyellowred), defined by 20percentiles (quintiles). However, in the case of Re too many analyses were below the analytical detection limit and the number of intervals and representative colours is restricted to three. The distributions of Rb, as a representative of the strong association described above, and of Th, which represents the same association more weakly, are shown in Figures 16 and 17. This strong geochemical association in the data is interpreted to reflect the amount of inorganic clastic sediment in the samples.
The areas where the highest Rb values are concentratedsuggesting a high clastic content and, in turn, a relatively highenergy lacustrine sedimentary environmentconstitute four major zones: (1) In the SW of the map area; bordered to the SW by the Caniapiscau Reservoir, and to the NE by a line extending approximately from Menihek Lake, at the southern boundary of  the coverage, to west of Schefferville along the territorial border, and to the Sérigny River at 55°36′ 3 N/69°23′ W. This feature is not defined by high Th values.
(2) A two-pronged feature in the NW of the map area, bounded to the west by the western boundary of coverage. The northern prong extends to SW of Lac Dulhut at 58°38′ N/70°46′ W, bounded to the SE by the Riviere aux Feuilles; while the southern prong extends to 57°37′ N /70°16′ W, at the Riviere aux Mélezes. This same feature is characterized by high Th values, although the zone defined by Th is not 100% coincident with the Rb-defined zone.
(3) A wedge-shaped zone, 400 km long, west of the territorial border in Québec, extending from Eclipse Lake (59°46′ N/64°19′ W) in the north, where the zone is only about 10 kilometres wide, to Lac Champdoré in the south (55°55′ N/65°53′ W) where the width increases to 100 km. High Th values also characterize this zone although they extend further south to enclose the Misery Lake REE occurrence (55°20′ N/63°58′ W; see Fig. 1B).
(4) A zone along the Labrador coast between Okak Bay, in the south (57°29′ N/62°12′ W) to Saglek Fiord (58°29′ N/62°57′ W) at the limit of coverage in the north. High Th values are restricted to the northern and southern extremities of this feature.
The weaker association of Th appears to be because in addition to the clastic-sediment content, as displayed in the valleys of the Koksoak and Feuilles rivers in the NW, its distribution appears to be related to certain rare-earth or rare-metal occurrences. In particular, Th defines the following: (1) The dispersion train from the Strange Lake REE/RM occurrence (McConnell & Batterson 1987;Fig. 1B); (2) A linear, subparallel zone 40 km to the north of Strange Lake, that may also represent a dispersion train from enriched bedrock; (3) An approximately circular high, centred on 57°19′ N/64°1 1′ W, west of Okak Bay in northern Labrador (NTS 14E), where the results of earlier lake-sediment and lake-water analyses, and detailed aeromagnetic studies carried out by the private sector, suggest the presence of REE mineralization (Amor 2011);and (4) A linear feature extending almost due south for 100 km south of Ungava Bay, immediately west of Kangiqsualujjuaq. This feature is transgressive to the regional strike and underlain by granites, gneisses and minor metavolcanic rocks of the Tunulik Belt.
The co-association of LOI ( Fig. 7) with S ( Fig. 18) is fairly strong. High values occupy a broad zone in the catchment basin of the Koksoak River, which drains into Ungava Bay; this zone is also marked by high values of Ca. An area of very low LOI and S is present in Labrador north of 56°N, and in the adjacent part of Québec. Low values of Ag, Cd and Hg, to some extent, correspond with these low S values. Geological controls are not evident in this pattern as it covers many different rock types. However, despite comparable correlation strength with LOI and S (see Fig. 14), the mapped distributions of Ag (Fig. 11B), Cd (Fig. 13C), Hg (Fig. 13D) and Se (Fig. 4B) appear to be controlled at least in part by geology. For example, Ag is enriched over rocks of the Baby-Howse and Doublet domains of the Labrador Trough, at least as far north as Lac Hérodier (see below), and over rocks of the cupriferous Seal Lake Group in the SE (most of which is outside the area of coverage). Enrichment of Cd, Hg and Se appears to correlate with the Labrador Trough over its entire strike length, and similar enrichment is seen over the Seal Lake Group.
The distribution of La and U is shown in Figures 19 and 20. These elements show an association with Ce, and Y. The axis of the largest feature, which probably incorporates distribution from a number of      . From this point it extends eastward to the limit of coverage north of Harp Lake (55°36′ N/61°59′ W). The feature defined by the uppermost 20-percentile of La is at least 100 km wide over this last stage, although it encloses two local maxima of U. One of these local maxima is more effective than La in delineating dispersion from the Strange Lake REE/RM deposit; indeed, it was the original anomalous U analyses in lake sediment that led to its discovery (McConnell and Batterson 1987). The other local maximum of U represents the westernmost extension of the uraniferous Central Mineral Belt (Sparkes 2017), most of which was not covered by this study. Values of both La and U delineate the zone of enrichment west of Okak Bay, which is also defined by Th (see above); the stronger response here is for La.
In the southeastern corner of the study area, a zone of high values of La, which encloses a number of REE occurrences in the Letitia Lake/Red Wine Mountains area of Labrador (Miller 1988), is disposed slightly to the SE of a zone of the highest U values. The    latter is associated with a number of minor Cu occurrences, but no U significant mineralization has been reported.

Geochemical characteristics of the Labrador Trough
A feature characterized primarily by Cu (Fig. 21), but also by Ag (Fig. 11B), Fe, Hf (Fig. 24A), Ni, Pb, and Zn follows the strike of the Labrador Trough from André Lake in Labrador in the SE (54°3 9′ N/65°41′ W) to Lac Hérodier (57°22′ N/68°41′ W) in the NW. It reappears north of the Koksoak River and extends to the northern boundary of the sample coverage at Tasiujaq on Ungava Bay. The feature appears to be geologically controlled by the Montagnais gabbro and the Willbob, Murdoch and Bacchus mafic volcanic formations; interestingly, the gap in the zone of elevated Cu, between Lac Hérodier, and the Koksoak River, corresponds to the absence of significant Cu occurrences in bedrock (McCurdy et al. 2018) and the absence of rocks assigned to the Bacchus Formation.
A less extensive feature, in strike length and thickness, occupies the core of the feature defined by Cu. Its northern termination is north of Lac Otelnuk, at approximate latitude 56°18′ N. The feature is characterized by elevated values of Al (Fig. 22), Cd, Co, Cr, Ga, In, Li, Mg, Sc, Te and V.
The entire Labrador trough is defined by high values for a range of other elements. This is exemplified by Sb (Fig. 23A) although similar features are displayed by As (Fig. 23B), Bi (Fig. 23C) and Re (Fig. 23D) as well as by Cd (Fig. 13C) and Hg (Fig. 13D).

Dispersion from Strange Lake
The world-class rare earth element (REE) and rare metal (RM) deposit known as Strange Lake -that straddles the territorial border east of Lac Brisson in Québecwas discovered by following up anomalies of U and Pb in lake sediments, which were identified in samples from the NGR program (McConnell and Batterson 1987). The dispersion train from Strange Lake has a multi-element geochemical signature, including enrichment in REE and other incompatible elements (Batterson 1989;Batterson & Taylor 2009). The suite of elements analysed in the current study only includes two REE (Ce and La) and one REE 'surrogate' (Y), whose dispersion trains are not particularly well-defined; at least, when quintiles are used to represent the geochemistry. However, the dispersion train is well displayed by values of other elements, including Hf (Fig. 24A), Nb (Fig. 24B), Sn (Fig. 24C) and Zr (Fig. 24D). Each of these elements displays patterns of interest for different reasons. Tin dispersion from Strange Lake is previously undocumented and presumably indicates enrichment in the deposit itself, which is also not known to have been documented previously. Niobium defines a linear feature parallel to the dispersion from Strange Lake and about 30 km north of it, which might represent a discrete source; and Hf, Sn and Zr apparently extend the dispersion   train about 20-30 km to the west of the Strange Lake deposit into Québec, which might also indicate a second source region.

Geochemical characteristics of the Seal Lake/Red Wine Mountains area
The extreme SE of the coverage area in Labrador is marked by enrichment in a variety of elements. There are two mineralized districts in this area: (1) the Seal Lake area; and (2) rocks of the Seal Lake Group. Both are characterized by numerous small occurrences of Cu mineralization (Van Nostrand & Corcoran 2013), of which only a fraction are covered by the current study. A zone of restricted Cu enrichment at the limit of sample coverage is enclosed to the NW by a much larger zone of elevated values of Sb (Fig. 23A), As (Fig. 23B), Mn, P, Re (Fig. 23D), Se (Fig. 4B) and U (Fig. 18B). The Red Wine Mountains/Letitia Lake area contains REE mineralization (Miller 1988) and shows lake-sediment responses in Ce, Hf (Fig. 24A), La (Fig. 19), Nb (Fig. 24B) and Pb. Elevated values of Ag (Fig. 11B), Be, Cd (Fig. 13C), Mo, W and Y define zones that overlap both the Seal Lake and Red Wine Mountains/Letitia Lake areas.

Application to mineral exploration
Although the approach used in this paper to create most of the geochemical maps, in which data were divided into five equal quantiles, is aimed at the definition of regional patterns, the data have equal potential in the identification of patterns related to mineralization. Such responses would, of course, be expected to be manifest in considerably less than 20% of each element's highest values, so the 5-class approach is not appropriate for mineralexploration purposes. Examples of more focussed maps for Sb and Zr are shown in Figure 25A and B, in which the uppermost decile (green) and 2.5-percentile (red) are indicated . For Sb (Fig. 25A), the maximum values, as defined by the latter colour, can be seen to be present in six zones within the larger zone of Sb enrichment that covers the whole of the Labrador Trough. Five of these overlie clastic sediments of the Menihek Formation and banded iron formation of the Sokoman Formation; the sixth and most easterly overlies clastic sediments of the Baby Formation, intruded by gabbros and ultramafic rocks of the Montagnais Suite.
High values of Zr (Fig. 25B), defined by the same method, define the well-documented glacial dispersion train from the Strange Lake deposit, east of the provincial border and west of Nain. Another linear feature is concordant with the strike of the Labrador Trough but locally contact-transgressive. It extends for 100 km over the sedimentary Menihek Formation, banded-iron formation of the Sokoman Formation and dolomite of the Denault Formation between 55°35′ N, 67°35′ W (Lac Wakuach) and 54°51′ N, 66°44′ W (the town of Schefferville). A local maximum centred about 20 km SE of Tasiuaq overlies clastic sediments of the Baby Formation, gabbro of the intrusive Montagnais Suite, dolomite of the Denault Formation, and schists and amphibolites of the Rachel-Laporte Domain and is unlikely to be directly related to bedrock geology. This is also probably true of the feature in the NW of the study area, whose shape is similar to that defined by high Rb values and may be related to elevated clastic content of the lake sediments.

Conclusion
Reanalysis of c. 5000 lake-sediment samples from northern and western Labrador, by ICP-MS after a modified aqua-regia digestion, has permitted their combination with c. 16 000 samples from adjacent Québec that were digested and analysed in the same way.
The resulting composite database covers an area of almost 300 000 km 2 in northeastern Canada, much of which is of current or historical interest for mineral exploration.
Reanalysis of the Labrador samples was necessary because although geochemical data were previously available for both the Québec and the Labrador samples, the pre-analysis application of an aqua-regia digestion on the Québec samples, and multi-acid digestion (or the alternate total-sample method of neutron activation) on their Labrador counterparts, made them difficult to compare; when the data for some elements were plotted spatially, the interprovincial border appeared as a pronounced discontinuity. Also, the ICP-MS package includes several elements for which analyses were previously unavailable for the Labrador samples.
Because paired analyses of the exact same samples are not available, the match between the two components of the new dataset was tested by comparison of cross-border nearest neighbours. In general, correspondence is good and only one element (Ag) required levelling. However, discontinuities within the Québec dataset caused by slight calibration shifts, for Cd, Hg and Zn, were addressed by the same method, with satisfactory results.
In creating geochemical maps, the data were smoothed through the application of a moving-median algorithm with the search radius set at 10 km, and this allows the delineation of regional patterns in the combined data. Amongst these patterns are several large-scale features that are believed to be related to relatively high amounts of clastic inorganic material in the lake-sediment samples. The geological control exerted by rocks of the Labrador Trough is most strongly exerted on certain chalcophile elements like Sb, As and Bi, although the rocks of the trough influence the dispersion of most siderophile and chalcophile, and some lithophile, elements. The previously documented glacial dispersion train associated with the Strange Lake REE/RM deposit has been shown for the first time to be defined by Sn, and to extend several tens of kilometres west of the deposit.
The new combined dataset should provide additional information for the geological mapping of this large area, as well as providing stimulus to mineral exploration. The approach used in this paper, in which smoothed data for most elements were divided into quintiles, is aimed at the definition of regional patterns. However, the unsmoothed analyses of the elements of interest also have application in identifying the responses to mineralization, which in most cases is of limited areal extent (the dispersion from Strange Lake being a notable exception). This approach would serve effectively as a regional exploration tool.