Calibration data set and ph transfer function
A total of 266 diatom taxa representing 34 genera were identified in samples from the 45 subarctic north-eastern Lapland sites, comprising predominantly benthic and periphytic species. Acidic sites were mainly characterized by Brachysira brebissonii Ross in Hartley, Brachysira vitrea (Grun.) Ross in Hartley and Frustulia rhomboides var. saxonica (Rabenh.) De Toni, whereas more alkaline sites showed a dominance of Cyclotella rossii Håkansson, Achnanthes minutissima Kütz., Achnanthes levanderi Hust., Aulacoseira italica f. crenulata (Ehrenb.) Ross in Hartley, and particularly Fragilaria spp.
After deleting highly collinear variables (latitude, conductivity, DIC, TOC) on the basis of their high variance inflation factors (Weckström, Korhola & Blom 1997b), forward selection and associated Monte Carlo permutation tests indicated that four of the remaining 22 environmental variables made statistically significant contributions to explaining the variance in the diatom species data. These four variables were Ca, pH, Si and maximum lake depth, which together accounted for 26·3% of the total variance in the diatom data.
The results of the variance partitioning are summarized in Table 3 and 4. Following the guidelines given in Okland & Eilertsen (1994), only the subset of significant variables was included in the partial CCAs. First, the variance in the diatom data was partitioned amongst the chemical and physical components. The three most meaningful chemical variables (pH, Ca, Si), as determined on the basis of the forward selection procedure (see above), independently accounted for a statistically significant (P≤ 0·05) proportion (18·0%) of the variance in the diatom data, whereas the most influential physical variable (lake depth) independently captured a statistically significant proportion of 7·5% of the variance (Table 3). There was only a small conditional effect between the two sets of variables that contributed an additional 0·8% of the variance. We thus conclude that the major chemical variables affecting the diatoms in the data set are not significantly confounded by the existing physical gradients.
Table 3. Results of partial canonical correspondence analysis (CCA) partitioning the total variance in the calibration set diatom data between the most influential chemical (pH, Ca, Si) and physical (lake depth) gradients. P = significance level of Monte Carlo permutation test (99 unrestricted permutations)
|Source of variation||Variance explained (%)||P|
|Chemical vs. physical variables|
| Independent contribution of pH, Ca and Si||18·0||0·01|
| Independent contribution of lake depth||7·5||0·01|
|Covariance between chemical and physical variables||0·8||–|
Table 4. Results of partial ordination (CCA) of diatom assemblages in 45 north-eastern Lapland lakes. P = significance level of Monte Carlo permutation test (99 unrestricted permutations)
|Variable||Covariable||Variance explained (%)||P|
Secondly, we partitioned the explained variance between the three most significant chemical components. This analysis indicated the significant and unique responses of diatoms to pH, Ca and Si, respectively, regardless of the covariables used in each analysis (Table 4). On the basis of all the ordination analyses, pH, Ca and Si were therefore identified as strong predictor variables in explaining the diatom composition in our 45-lake calibration data set. Although reliable inference models could, at least in theory, be developed for each of these three chemical variables, we present in this connection only the prediction model for reconstructing trends in lake water pH. When interpreting the results, however, one should bear in mind that, although there is an independent and statistically significant response of diatoms to lake water acidity, pH is at least partially confounded by other chemical variables, as indicated by Table 4 (e.g. almost half of the variance explained by pH is conditional on Si). Subsequent reconstructions for pH can therefore not be considered totally independent of other chemical gradients.
In relation to pH, diatom data had a gradient length of 2·17 standard deviation (SD) units. In a DCCA with pH as the sole constraining variable, pH explained 10·0% of the variance in the diatom data, whereas the second unconstrained axis explained 9·6%; the gradient length of the second unconstrained axis was 2·02 SD units. These results support our view that pH is a strong explanatory variable, but there also exist strong secondary gradients and hence much variation in the diatom data that is not related to this particular variable. We nevertheless conclude that the relationship between the diatom assemblages and lake water pH is strong enough for developing a statistically robust and biologically meaningful predictive model from these data for lake water pH.
Results of WA-PLS regression for lake water pH using the 45-lake data set are shown in Fig. 2. The transfer function indicated a close agreement between measured and diatom-inferred pH (r2 = 0·73), and the model had a low RMSE of 0·17 pH units (Fig. 2a). After leave-one-out cross-validation, the first component WA-PLS model provided a jack-knifed r2 of 0·60 and a RMSEP of 0·20 pH units (Fig. 2b). As indicated by the residuals, the model typically tended to over-estimate values slightly at the low end of the pH range and under-estimate the high values, the reasons for which are discussed, for example, by Lotter et al. (1997). No improvement on the predictions was achieved by using further components in WA-PLS or by deletion of unusual samples. In general, the statistical performance of the predictive model compares well with the other previously developed diatom–pH transfer functions (Hall & Smol 1995; Korsman & Birks 1996).
Figure 2. Relationship between measured and diatom-inferred lake water pH in 45 north-eastern Lapland lakes using a one-component weighted-averaging partial least squares (WA-PLS) model. (a) Inferred pH (apparent relationship) without leave-one-out cross-validation (r2 = 0·73; RMSE = 0·17 pH units). (b) Predicted pH with leave-one-out cross-validation (r2 = 0·60; RMSEP = 0·20 pH units). Distribution of residuals is also shown.
Download figure to PowerPoint
The LOI, β-activity and SCP profiles for Lake 222 are shown in Fig. 3a, together with the previously published 137Cs activity profile as well as the age–depth curve obtained by 210Pb dating. The SCP profile showed trends similar to those noted in sediment cores previously taken from north-western Finnish Lapland (Sorvari & Korhola 1998; Korhola et al., in press) and lakes close to the pollution sources on the Kola Peninsula (Rose 1995). There is first a slow but steady increase in concentration from approximately 7·5 cm until a period of more rapid increase at approximately 4·0 cm. This is followed by a peak at 1·5 cm, then a decline, and another concentration peak right at the core top. Sediment metal records from lakes in Kola Peninsula document increases above background levels around the 1920s and 1930s (Norton et al. 1992). The increase in SCP at 4·0 cm may thus date from this time (Sorvari & Korhola 1998; Korhola et al., in press). Maximum concentrations of Ni and Cu in the sediments occurred in the latter half of 1970s (Norton et al. 1992), coinciding with the period of maximum emissions of sulphur dioxide from the area (Traaen et al. 1991). This date may also correspond with the SCP concentration peak at 1·5 cm. If the increase observed in β-activity at 0·5 cm is interpreted as being caused mainly by the nuclear reactor failure in Chernobyl in 1986, then the SCP and β-activity chronologies match well with each other. However, the activity peak at Lake 222 is rather weak, indicating either that the Cs in this lake has not been effectively removed from the water and fixed to the sediment or that the activity measurements are actually detecting radionuclides other than 137Cs. Nevertheless, the established chronology based on SCP and β-activity is in good agreement with the 137Cs activity profile and 210Pb dates obtained from a parallel core in the lake (Fig. 3a).
Figure 3. (a) Depth profiles of loss-on-ignition (LOI), spheroidal carbonaceous particles (SCP) and β-activity for Lake 222 core. 137Cs activity and the age–depth curve obtained by 210Pb dating another sediment core from the site are also shown [137Cs and 210Pb results adapted from Vartiainen et al. (1997)]. (b) The beta count rates for the sediments from Lake Sarvijärvi (SJ). (c) The beta count rates for the sediments from Lake Pieni Kokkoselkä (PK).
Download figure to PowerPoint
The β-activity profiles for SJ and PK are shown in Fig. 3b,c. At PK, the β-activity rises from a sediment depth of 12 cm and reaches its highest value at 4·0 cm, suggesting a possible Chernobyl signal in 1986. At SJ, the activity starts to rise at 3 cm, with the activity maximum at the top of the core (0–0·5 cm). As we found only small compaction, we assumed that the sedimentation rates of the upper part of the cores (0–3 cm) were only slightly higher (10%) than those of the lower parts, a view that is supported by the 210Pb dating of Lake 222 (Fig. 3a). Thus, average sedimentation rates can be estimated for each of the three study lakes ranging from a low of 0·5 mm year–1 and 0·6 mm year–1 (Lake 222 and SJ, respectively), to a high of 4·0 mm year–1 (PK). In general, sediment accumulation rates in lakes vary according to the input of both organic and inorganic material from the lake catchment and biogenic material from the lake itself. In oligotrophic systems, like the lakes studied here, material derived from the catchment usually dominates. The extremely slow sedimentation rates observed at Lake 222 and SJ are typical for arctic sites with rocky catchments and low primary production (Douglas, Smol & Blake 1994; Sorvari & Korhola 1998). The considerably faster sediment accumulation rate at PK is most probably related to the location of the site at a lower elevation (148·0 m a.s.l.) and the resulting higher percentage of organic soils and denser vegetation cover in its catchment. The sediment of PK also had much higher LOI values (70–80%) than the sediments of lakes 222 and SJ (20–30% LOI). In all, the data demonstrate that a continuous stratigraphic record is present in each of the cores over the period of interest.
Each of the studied cores featured surprisingly monotonous diatom stratigraphies (Fig. 4). The diatom floras of SJ and PK were dominated by acidophilous, benthic species, such as Brachysira brebissonii and Frustulia rhomboides var. saxonica as well as various Eunotia and Navicula species. Lake 222's development was dominated by small centric Cyclotella spp., particularly C. rossii, which constituted approximately 60% of the total flora throughout the core. The high abundance of Cyclotella is typical for deeper lakes in Finnish Lapland (Weckström, Korhola & Blom 1997a,b; Sorvari & Korhola 1998). Diatom-inferred pH (DpH) was more or less stable in each core. No major response to the onset of operations at the Kola smelters and an associated increase in acid deposition was observable either in species compositions or in the inferred DpH (Fig. 4). At each site, the DpH for the uppermost subsample corresponded relatively well with the known pH value of the lakes (7·0, 6·5, 6·8 and 6·7–6·9, 6·6, 6·5–6–6, respectively), suggesting that the pH reconstruction technique used is valid.
Figure 4. Relative frequency diagrams of the most dominant diatom taxa recorded in the sediments of the three study lakes and pH reconstructions (DpH). (a) Lake 222, (b) Pieni Kokkoselkä (PK), (c) Sarvijärvi (SJ). The commencement of operations at the Kola smelters in the early 1940s is indicated in the diagrams by a solid line.
Download figure to PowerPoint
Although the diatom stratigraphies look fairly featureless, there might still be changes that are not observable by visual inspection. The potential ‘hidden’ trends in the diatom assemblages can be analysed more sensitively using community ordination techniques that allow the interrelatedness of the samples to be detected. The technique used here was CCA, a direct gradient analysis method where the ordination axes are constrained to be linear combinations of environmental variables (ter Braak 1986). If the fossil assemblages are entered passively in a CCA of the modern diatom–environment data set used for pH reconstruction, changes in the diatom compositions can be compared directly to the environmental gradients within this data set (Birks, Juggins & Line 1990; Allott, Harriman & Battarbee 1992). In Fig. 5 it can be seen that the stratigraphic samples (i.e. diatom assemblages) of each core are clustered closely around each other in the CCA space and are not chronologically aligned with the pH or any other environmental gradient. The overall CCA results thus suggest that no major shifts in abundance of the diatom taxa are present in the records to indicate recent acidification of the lakes, leading us to conclude that the studied ecosystems have not so far responded biologically to the increased acid loading.
Figure 5. Canonical correspondence analysis (CCA) ordination diagram showing the positioning of the samples of the three studied cores in relation to environmental gradients in the calibration data set used for pH reconstruction. Core samples were treated passively in the ordination analysis.
Download figure to PowerPoint
To assess further the sensitivity of the selected lakes to acidification we calculated their baseline, site-specific critical sulphur load by means of an empirical diatom-based palaeolimnological model calibrated using sites and data from the UK (Battarbee et al. 1996). This model is based on a dose–response function that can be used to set critical load values for a site from a knowledge of the ratio of Ca2+ of the water to the modelled S deposition at the site. Such a test revealed that for our reference lake (SJ) the critical load has not yet been exceeded (critical load: 0·37 keq S ha–1 year–1; current deposition: 0·19 keq S ha–1 year–1). However, Lake 222 has already reached its critical load limit (0·59 keq S ha–1 year–1, 0·62 keq S ha–1 year–1, respectively), whereas in the case of PK the level is exceeded by more than twofold (0·19 keq S ha–1 year–1, 0·44 keq S ha–1 year–1, respectively). The lack of any pH decline in PK is therefore most revealing regarding the discrepancy between our palaeolimnological evidence and the empirical diatom model based on the diatom–water chemistry data from the UK. These results suggest that the Ca : S ratio of 94 : 1 used to predict critical load exceedance for sites in the UK (Battarbee et al. 1996) is probably not directly applicable to assess the acid sensitivity of lakes in Finnish Lapland. The results further highlight the need to calibrate the diatom model specifically for north-eastern Lapland lakes by means of a larger data set of inferred acidification profiles from dated lake sediment cores throughout the region. Such work is currently in progress.
Because of the importance of organic matter to Finnish lakes, the influence of organic anions on lake acidity and calculations of critical loads should consequently be given special attention (Forsius, Kämäri & Posch 1992). In the Barents Sea region, the most acid lakes are found on the Russian side of Kola, where the flat topography favours the occurrence of peatlands and where the organic carbon content of surface waters is highest (Henriksen et al. 1997; Blom et al. 1999). In contrast, the northernmost areas of the Finnish Lapland are characterized by less organic soils, and the concentrations of humic substances in waters are much lower. In our northern Finland lake calibration data set, the median value for TOC was 4·8 mg l–1 (n = 98) and the median value for DOC was 3·0 mg l–1 (n = 45) (Blom, Korhola & Weckström 1998), whereas the median TOC for all Finnish lakes under regular monitoring is 12·0 mg l–1 (n = 987) (Forsius et al. 1990). In addition to soils, this decreasing south to north trend in TOC values in Finland apparently reflects variations in climate, primary production and decomposition rates (Kortelainen 1993).
Organic carbon may buffer lakes against acidification, but may also add acidity to surface waters (Kullberg et al. 1993). It has been shown that in humic lakes the effect of strong mineral acids is superimposed on organic acid contributions to acidity (Gorham et al. 1986; Brakke, Henriksen & Norton 1987). As a consequence, organic acids make humic lakes generally more sensitive to acidification than clear-watered lakes (Brakke, Henriksen & Norton 1987). Biological communities and processes are also affected by humic substances, either directly through interfering with metabolic processes or indirectly by altering the bioavailability of nutrient or toxicants (Petersen 1991; Kullberg et al. 1993). The biological response on acid loading in clear-water lakes may therefore differ substantially from that observed in humic lakes, which may, at least in part, explain the greater ability of northern lakes to resist acid deposition. Separate, geographically restricted dynamic models for defining critical loads and predicting ecological impacts in clear-water ecosystems in northernmost Lapland are therefore needed.
The observed stability in pH development during the 20th century is not unique to the three cores studied. We have obtained a similar pH record from an arctic lake that we have studied in western Finnish Lapland (Sorvari & Korhola 1998). In contrast, diatom floras in sediments of some mountain lakes of the Kola north have clearly undergone changes during recent times, providing evidence of the development of water acidification (Moiseenko, Dauvalter & Kagan 1997). This is not surprising per se, because the prevailing north-easterly and northerly winds transport higher concentrations of acid compounds to the areas north of the Kola smelters, particularly during winter (Tuovinen et al. 1993). On the other hand, the interpretations of the increased acidity in the lakes of the Kola north is based predominantly on the knowledge of diatom ecology from more southerly locations, including the reconstruction of lake water pH by means of the ‘index B’ developed for the lakes of Sweden (Moiseenko, Dauvalter & Kagan 1997). The transformation of such information to northern algal communities and lake situations can, according to our own experiences, be highly misleading (Weckström, Korhola & Blom 1997b).
The currently available palaeolimnological data indicate that the watershed and in-lake alkalinity-generating processes are still effectively opposing the acidification in many of the clear-water lakes in Finnish Lapland. This is in agreement with other palaeolimnological data showing that there is often a long time-lag between the onset of contamination by atmospheric pollution and its first effects on the biology and chemistry of surficial waters (Kingston et al. 1990). However, the present study is based on three lakes only, from which only two would be expected to respond to the current levels of acid deposition. Clearly, further data are needed to make effective generalizations about the acidification trends and status of the numerous small lake basins in the study area. This includes, for example, additional palaeolimnological studies on lakes in the region, as well as the development of the local ‘diatom model’ to assess critical loads of acidity for these unique subarctic lakes.