The investigations of population structure using Principal Component Analysis (PCA) showed that only one morph inhabited the landlocked localities Røde Elv (69° 16'N, 53° 29'W) and Kangarssuk (69° 16'N, 53° 50'W) [see Additional file 1]. Therefore, only individuals sampled from these two populations were used in the analysis. All individuals from both localities had parr marks characteristic of fish that have not smoltificated. The individuals from Røde Elv were small, ranging in size from 82 to 169 mm, and had large eyes and ventrally placed mouths. Despite their small size, a number of fish in Røde Elv were ready to spawn, a feature common to both populations. Specific for all fish sampled at Kangarssuk were large external, infectious wounds. Fish in Kangarssuk were large, ranging in size from 220 to 367 mm and a considerable number were ready to spawn. All individuals caught at Kangarssuk were more than 5 years old, while ages of individuals in other populations were uniformly distributed. All had eaten chironomids and additionally a large number of individuals had preyed on benthic ostracods and Lepidurus arcticus.

Detailed results from multiple linear regressions are depicted in table 1. Regressions were carried out with growth rate residuals (GR _t ) as response variable and one year delayed autoregressing (GR _t-1), mean annual temperature (T_a), mean summer temperature (T_s), mean winter temperature (T_w) and mean annual precipitation (P_a) as predictor variables. None of the four weather parameters showed significant interdependence. Analyses of cohorts caught in 2004 were based on weather data from Qeqertarsuaq, while analyses from 1990 cohorts were carried out using data from Aasiaat.

Analyses of the Røde Elv individuals caught in 2004 (n = 14) showed a negative effect of T_a on the annual GR _t together with a non-significant GR _t-1 (R² _total = 0.92; T_a: p = 0.05; GR _t-1: p = 0.11; Fig 1a). The otolith data from this sample allow regression with more than two predictor variables in only two cases but neither yielded any significant results (table 1). Otolith data from individuals caught in Røde Elv in 1990 (n = 14) by contrast, allowed simultaneous modelling of several predictor variables and GR _t . These data showed a significant positive relationship with T_a and a significant negative relationship with P_a(R² _total = 0.83; T_a: p = 0.03; P_a: = 0.05; GR _t-1: p = 0.78; Fig 1b).

Data on annual GR _t from the Kangarssuk individuals caught in 2004 (n = 14) allowed simultaneous multiple linear regression with several predictor variables and the analysis yielded several significant relationships between GR _t and T_s and T_w, respectively (se Fig 2a, 2b and table 1). For example, multiple linear regression with GR _t and the predictor values T_s, T_a and GR _t-1 showed a positive significant connection (R² _total = 0.94; T_s: p = 0.04; T_a: p = 0.18; GR _t-1: p = 0.66), while a similar regressing with GR _t and the predictor variables T_w, T_a and GR _t-1 showed a negative significant relationship (R² _total = 0.98; T_w: p = 0.04; T_a: p = 0.81; GR _t-1: p = 0.36). Thus, GR _t are significantly related negatively to T_w and positively to T_s, without any significant GR _t-1.

Data on annual GR _t from individuals caught in Kangarssuk in 1990 (n = 10) allowed multiple linear regressions incorporating several predictor variables. A model incorporating T_a, P_a and GR _t-1 showed a negative effect of P_a on GR _t (R² _total = 0.77; P_a: = 0.07; T_a: p = 0.21; GR _t-1: p = 0.09), while a model incorporating T_w instead of T_a further documented the relationship (R² _total= 0.8; P_a: = 0.07; T_w: p = 0.17; GR _t-1: p = 0.09; se also Fig 2c).

Otolith derived changes in mean values of fork length estimated means (F_em), and thereby growth, were assessed from the periods 1982–1989 and 1995–2003, respectively. An ANOVA test was performed to evaluate the differences between the time series. The estimated rates of growth from the Røde Elv samples from 1982–1989 were significantly larger compared with the 1995–2003 sample (p = 0.008; d.f. = 7; Fig 3a). The opposite was the case among fish from Kangarssuk. Here, rates of growth were larger in the period 1991–2003 compared to 1982–1989 (p = 0.0004; d.f. = 7; Fig 3b).

The study was carried out on Disko Island (69° 15'N, 53° 31'W) off the west coast of Greenland. Six study sites (four lakes and two rivers) were chosen known to have populations of Arctic charr. Data on wind, temperature, precipitation and albedo were obtained from records at the weather survey station at Qeqertarsuaq on Disko (1990 to 2001) and similar from the Danish Meteorological Institute's 18 climate station in Aasiaat on the west coast of Greenland approximately 65 km from Qeqertarsuaq (1981 to 1989).

Water temperatures were measured with depths at all study sites using a Hydrolab multisonde. Fish were caught in July and August 2004 with multi mesh gill net (Faarups Specialnet, Denmark) of a type proposed for monitoring use in Nordic countries (referred to as 'biological survey gill nets'; Nordic Norm) designed to catch representatively across a population. The nets were 40 meters long and composed of 12 mesh sizes: 29, 35, 5, 15.5, 24, 12.5, 8, 55, 10, 6.25, 19.5 and 43 mm. Sampling was carried out using randomized pelagic and littoral settings for each locality. The pelagic nets were placed parallel to the shore, while the littoral nets were set perpendicular to the shore. Specimens from the two rivers were obtained by angling (Nipisat Elv) or by hand nets and multi mesh gill nets (Røde Elv).

Populations suitable for climate studies were selected using principal component analysis (PCA) conducted in PCord-4 to examine the homogeneity following standard procedures detailed in 19 with respect to morphological quantitative characters (fork length, weight, age, mouth depth, mouth depth relative to fork length and condition factor 20 ) and categorical characters (colour, sex, gonad maturity and stomach content).

Otoliths (sagittae) were sampled from all fish and subsequently mounted on a glass slide using resin at 140°C with the sulcus facing down and polished to enhance the annular ring structure. The age of the fish was estimated to the nearest year by counting of annual rings and by making transects to quantify the distance between annular rings using digital video microscopy with the software Image Pro Plus 5.0 (MediaCybernetics). Transects were placed as close to a downward angel of 90° to rostrum as possible in all examinations.

Measurements of otolith radius, fork length (F) and age at time of capture were used in constructing an Individual-Character matrix for each locality. We applied a recent model for growth back-calculation developed by Morita and Matsuishi 21 which takes into account that growth of fish (fork length) may be allometrically uncoupled from growth of their otolith. Multiple regression analyses were conducted using SYSTAT 8.0 (Systat Software Inc.) and back-calculated F-values were applied for calculation of growth rates using the simple relationship between body size at succeeding years:

G _t = F _t+1 - F _t

where G _t is the estimated growth rate at age t, and F _t+1 and F _t are fork lengths estimated by back-calculation at age t and t+1. To allow detection of signals from climate conditions on growth rates, the effect of ontogenetic development in every individual was subtracted from the G-values. We assumed a simple additive relationship between the effects of ontogenetic development and environmental conditions on the realized growth rate of the individual:

G _t = f(O) + g(E)

where O is ontogenetic stage, E is the sum of environmental conditions and f and g are mathematical functions. Thus, to reveal the effects of climate, which are inherent in E, we subtracted the O-impact from the estimated G _t and obtained what was termed annual growth rate residuals (GR _t ):

GR _t = G _t - f(O)

where the functional values of O were computed for each locality and for the total pooled data set by fitting a logarithmic function to the age-determined growth-tendency for the individual sample. This was done by logistic curve-fitting using the least-squares-method [22]. The resulting GR _t values for each individual were assigned to the concurrent calendar year, thus obtaining the final time-individual matrix consisting of the growth rates for each individual. The mean growth tendency in the population was calculated annually and contrasted to variation in annual temperatures and precipitation.

Three annual levels of analysis were conducted. First, the constructed time series for annual fish growth and climatic variation, assessed by the parameters T_a, T_s, T_w and P_a, were depicted graphically for visual tendency interpretation. The Product Moment Correlation Coefficient r was calculated for opposed time series of fish growth and climate variables.

Secondly, multiple linear regressions were carried out (Systat Software Inc.) regressing GR _t for each locality on T_a, T_s, T_w or P_a, respectively, incorporating one year delayed autoregressing (GR _t-1):

GR _t = α + yGR _t-1 + x _t β

where x _t is the mean climate parameter for temperature and precipitation and α, β and γ are regression constants.

Finally, to evaluate the impact of long-term climate change, we compared estimated growth curves between the two sampled time periods within each of the two landlocked populations. The growth curves formed a continuum from respectively 1982–1989 and 1995–2003. Data from 1990 and 2004 could not be used since the growth season had not finish at the time of sampling. The population means of the individual fork length estimates were tested by an ANOVA test [22] and plotted with corresponding standard errors of the estimate for each year. A statistical level of significance of 5% (p ≤ 0.05) was chosen for all analyses.