Introduction Thienemann (1925) Naumann (1932) Hynes 1975 Kratz and others 1997 Soranno and others 1999 Riera and others 2000 Omernik and others 1981 Osborne and Wiley 1988 Allan 1995 Kratz and others 1997 Soranno and others 1999 Riera and others 2000 Vollenweider 1975 Sverdrup and others 1992 Hornung and others 1995 a priori Essington and Carpenter 2000 We hypothesize that both streams and lakes are strongly linked to the surrounding landscape, and that spatial variation in surface water chemistry is regulated by non-mutually exclusive factors acting on various hierarchical scales depending on landscape type and/or geographic position. Here, we study the effect of regional and local-scale factors on three commonly measured water chemical variables. Acid neutralizing capacity (ANC) was selected to indicate the effect that catchment geology and weathering might have on buffering capacity. Total phosphorus (TP) was selected for its key role in driving ecosystem productivity and because it is biologically active (i.e., is expected to decrease along, e.g., lake chains). Finally, total organic carbon (TOC) was used as a surrogate measure of the importance of allochthonous input from the boreal catchments. The sites used in this study are often natural brown-water systems, with high contents of humic substances. We attempted to (1) identify and quantify possible sources of variation in surface water chemistry of boreal streams and lakes, (2) determine which environmental factors and which spatial scales are most important in determining the surface water chemistry of boreal streams and lakes, and (3) determine similarities/differences in the factors driving stream/lake water chemistry. Methods Study Site Johnson and Goedkoop 2000 Wilander and others 2003 1 2 Wilander and others (2003) Rapp and others (2002) Fig. 1 Location of the 361 lakes and 390 streams used to assess the influence of geographic position, and regional and local scale factors on surface water chemistry  2 2 2 2 Water Chemistry A single, midstream or midlake (approximately 0.5 m depth) water sample was collected in autumn 2000. All water chemistry analyses were done by the SWEDAC (Swedish Board for Accreditation and Conformity Assessment) certified laboratory at the Department of Environmental Assessment, Swedish University of Agricultural Sciences following international (ISO) or European (EN) standards when available. ANC is a measure of the buffering ability of lakes and streams against strong acid inputs. This metric was chosen because it includes humic substances and compensates for their natural variation, i.e., the effect of acid deposition is more pronounced than in other acidification indicators such as pH or sulfate concentration. Independent Variables 1 Table 1 Dependent and independent variables used in RDA Variable Unit N N a) Dependent    Chemistry    Acid neutralizing capacity (ANC) −1 3.36 (0.09−0.74) 0.51 (0.15−0.99) Total phosphorus (TP) −1 13.17 (2−28) 27.42 (2−67) Total organic carbon (TOC) −1 9.13 (2.02−16.6) 10.57 (2.2−21.08) b) Independent  Explained variability  Geographic position  Lakes Streams Latitude Decimal degrees   Altitude m a.s.l. (2) (3) * Dummy variable     Arctic/alpine Dummy variable     Northern boreal Dummy variable     Southern boreal Dummy variable     Boreonemoral Dummy variable     Nemoral Dummy variable   Regional factors      Mean annual discharge (Q) 3 -1  (5) x      Wet & dry non-seasalt Mg deposition    Catchment land use/cover      Urban areas %     Forested areas %     Alpine treeless land cover % (1) (2)   Glacier %     Open freshwater bodies %     Marsh/mires %     Arable land % (3) (1)   Pasture %     Alpine forested areas %   Local factors    Physical properties of sample site      Stream width m M  (4)   Lake area 2 (5)    Water temperature °C   ** Classified 0−3     Boulder (>250 mm) Classified 0-3     Block (200–250 mm) Classified 0-3     Cobble (60–200 mm) Classified 0-3     Pebble (20–60 mm) Classified 0-3     Silt/clay (0.02 mm) Classified 0-3     Coarse detritus Classified 0-3     Floating leaved vegetation Classified 0-3 (4)    Fine leaved submerged vegetation Classified 0-3     Periphyton Classified 0-3     Fine dead wood Classified 0-3   Riparian land use/cover      Deciduous forest Classified 0-3     Heath Classified 0-3     Arable land Classified 0-3     Alpine Classified 0-3     Pasture Classified 0-3     Mire Classified 0-3     Canopy cover Classified 0-3   a) Chemistry variables (n = 3) with mean values and 10th and 90th percentiles in parentheses. b) Environmental variables (n = 38), divided into three subsets, included in the analyses. Also shown are the first five variables (explained variability in %) that could best explain the variability in ANC, TP, and TOC, using RDA and stepwise forward selection with the order of selection shown in parentheses. Note: the middle boreal ecoregion was insignificant in the Monte Carlo permutation test and excluded from the analysis * Nordic Council of Ministers (1984) ** Catchments were classified as percentage land use/vegetation cover according to the same land use categories used for riparian zones. Hence, catchment land use/cover ranged from 0% (all classes) to 100%. Thereby, maximum urban areas in catchments were 10.2% (lakes) and 26.3% (streams), forested areas covered 99.8% in both lake and stream catchments, and alpine treeless cover was very high with 99.7% (lakes) and 99.9% (streams). Glacier areas comprised only 2.3% of total lake catchment areas, but covered 26.6% of stream catchments; other open freshwater bodies in the catchment comprised 19.4% of lake and 28.9% of stream catchments. Maximum marsh or mire land cover was 82.9% for lake and 67.4% for stream catchments, whereas pasture comprised 18.1% (lakes) and 14.2% (streams). Maximum alpine forested area cover was higher in lake (98.7%) than in stream catchments (65.6%), and maximum arable land covered 24.4% of lake and 24.6% of stream catchments. Nordic Council of Ministers (1984) Statistical Analyses First, detrended correspondence analyses were conducted to obtain the gradient length of both the stream and lake chemistry data. Because the gradient lengths were in both cases ≤1.5 SD, the linear method redundancy analysis (RDA) was used to study the effects of environmental variables representing geographic position and regional- and local-scale factors on stream and lake water chemistry. Moreover, preliminary analyses of water chemistry (total phosphorus concentration) and catchment land use (% agriculture) did not reveal any step changes between the northern and southern regions. RDA was performed on a correlation matrix and is a form of direct gradient analysis (like Principal Components Analysis). In a first step in RDA, the entire set of 60 environmental variables was tested to determine the significance of individual variables using a Monte Carlo permutation test (with 999 unrestricted permutations). Variables that were not significantly correlated with the three water chemistry variables or that were found to co-vary with other environmental variables (i.e., variance inflation factors >100) were removed (n = 22) from the data set. Variance Partitioning 2 2 Borcard and others (1992) Økland and Eilertsen 1994 Anderson and Gribble 1998 Fig. 2 Venn diagram (hypothetical model) showing the unique variation, the partial common variation, and the common variation of the three subsets G, R, and L representing the environmental data  Table 2 The procedure of variation partitioning of water chemistry (n = 3) in streams (n = 390) and lakes (n = 361) explained by three sets of environmental variables, geographic (G), regional (R), and local (L) in partial redundancy analysis (pRDA) Run Environmental variable Covariable streams lakes 1 GRL None 0.751 0.651 2 Geo R&L 0.018 0.037 3 R&L None 0.733 0.614 4 R&L Geo 0.173 0.184 5 Geo None 0.578 0.467 6 Reg G&L 0.099 0.078 7 G&L None 0.652 0.573 8 G&L Reg 0.055 0.116 9 Reg None 0.696 0.535 10 Local G&R 0.029 0.058 11 G&R None 0.721 0.593 12 G&R Local 0.270 0.221 13 Local None 0.480 0.430 a 3 Table 3 Calculation of explanatory power of each component in the variance partitioning model Variation explained by factors 2 3 2 streams lakes Geographic G 2 0.018 0.037 Regional R 6 0.099 0.078 Local L 10 0.029 0.058 Geographic & regional GR 12–6–2 0.153 0.106 Geographic & local GL 8–2–10 0.008 0.021 Regional & local RL 4–6–10 0.045 0.048 Geographic, regional & local GRL 7–8–(12–6–2)–(4–6–10) 0.399 0.303 Total explained TotX 1 0.751 0.651 Unexplained UX TotV−TotX 0.29 0.349 Total variance TotV  1.0 1.0 a 2 2 Fig. 3 Sources of variation in lake and stream water chemistry, respectively. Column labels indicate the variation (%) in acid neutralizing capacity, total phosphorus, and total organic carbon accounted for by each subset and their combinations  2 2 3 Stepwise RDA R 2 Ter Braak and Smilauer 1997–1998 Results 3 streams lakes 3 3 4a x Fig. 4 A B x x  4b x Stepwise RDA of stream and lake chemistry as dependent variables and the “single” variables of geographic position and regional and local environmental variables showed that all variables accounted for 65% (lakes) and 75% (streams) of the total variance. The amount of alpine treeless areas in the catchment was the single most important predictor of lake water chemistry (explaining 55.4% of the explained variance). The second variable selected was altitude (18.5%, i.e., the amount of residual variance explained after running the first variable selected, “alpine treeless areas in the catchment,” as a covariable), followed by the amount of arable land in the catchment (4.6%), percent coverage of floating leaved vegetation in the littoral (4.6%), and lake surface area (1.5%). For streams, the five best single predictors of water chemistry were the amount of arable land in the catchment (68%), followed by the amount of alpine treeless areas in the catchment, altitude (2.7%), stream width (2.7%), and mean annual discharge Q (1.3%). Discussion Essington and Carpenter 2000 Allan 1995 Kratz and others 1997 Soranno and others 1999 Riera and others 2000 Quinlan and others 2003 Vannote and others 1980 Kratz and others 1997 Schonter and Novotny 1993 Allan and others 1997 Johnson and others (1997) Hunsaker and Levine (1995) Omernik and others 1981 Cooper 1990 Osborne and Kovacic 1993 Redundancy analysis showed that the variability in both stream and lake water chemistry was explained by the similar regional- and local-scale variables. For example, as discussed above, the proportion of arable land use in the catchments was a strong predictor of stream water chemistry (68%), followed by alpine, treeless land cover (17.3%). For lake water chemistry, the amount of alpine, treeless land cover was a good predictor (55.4%), followed by altitude (18.5%) and catchment arable land use (4.6%). Clearly, several of the variables in different “local” and “regional” components covary. For instance, the amount of alpine treeless land cover in the region/catchment and stream width are presumably correlated with altitude. However, as demonstrated here, regional factors were better predictors of stream and lake water chemistry and thus contribute largely to the explanatory power of the covariation components. limes norrlandicus Nordic Council of Ministers 1984 limes norrlandicus Anonymous 1979 Johnson 1999 Wright and others 1998 Landers and others 1988 Larsen and others 1988 Foster and others 2003 Frissell and others 1986 Tonn and others 1990 Poff 1997 Minshall 1988 Manel and others 2000 Allan and others 1997