Geophysical Research Letters

Cloud response to summer temperatures in Fennoscandia over the last thousand years



[1] Cloud cover is one of the most important factors controlling the radiation balance of the Earth. The response of cloud cover to increasing global temperatures represents the largest uncertainty in model estimates of future climate because the cloud response to temperature is not well-constrained. Here we present the first regional reconstruction of summer sunshine over the past millennium, based on the stable carbon isotope ratios of pine treerings from Fennoscandia. Comparison with the regional temperature evolution reveals the Little Ice Age (LIA) to have been sunny, with cloudy conditions in the warmest periods of the Medieval at this site. A negative shortwave cloud feedback is indicated at high latitude. A millennial climate simulation suggests that regionally low temperatures during the LIA were mostly maintained by a weaker greenhouse effect due to lower humidity. Simulations of future climate that display a negative shortwave cloud feedback for high-latitudes are consistent with our proxy interpretation.

1. Introduction and Background

[2] Cloud cover is one of the most important factors controlling the radiation balance of the Earth because clouds modulate the surface temperature response to external forcing, partly by reducing shortwave radiation (sunshine) received at the ground surface. The response of cloud cover to increasing global temperatures represents the largest uncertainty in model estimates of future climate [Intergovernmental Panel on Climate Change, 2007; Trenberth and Fasullo, 2009; Clement et al., 2009] because the cloud response is not well-constrained in General Circulation Models (GCMs) [Bony et al., 2006; Soden and Held, 2006; Dai et al., 2006]. Clouds may diminish in the tropics (a positive feedback) and increase at high-latitudes (a negative feedback) under 21st century warming but with a wide range of spatial responses [Trenberth and Fasullo, 2009; Soden and Held, 2006]. Recent attempts to better constrain model uncertainties by analyzing instrumental cloud cover data [Clement et al., 2009; Lindzen and Choi, 2009] are burdened by the limited length of the observational record. Records of sunshine hours, which are inversely related to cloud cover data, suffer from similar constraints. Inter-annual weather variability, and the limited length of cloud records, confound analysis of the forcing-and-response relationship, over the last few decades.

[3] In the context of future climate change it is the sustained, long-term, response of cloud cover to changes in temperature that is of interest, because clouds will modulate the radiative perturbation caused by increasing greenhouse gases into the 21st century. A new millennial length treering stable carbon isotope (δ13C) series from Finland is interpreted here as a record of summer near-ground solar radiation and calibrated using local records of summer sunshine hours. When compared with the regional temperature history, based on treering densities, this new proxy record offers the chance to critically reduce the uncertainties in the cloud response to temperature anomalies.

2. Materials and Methods

[4] We developed a reconstruction of mean summer sunshine hours based on treering δ13C ratios from Scots pine (Pinus sylvestris L.) at a site known as Laanila, in the northern Boreal forest zone of Finnish Lapland (Table S1 of the auxiliary material). A modern δ13C chronology (AD 1740–2001) from this site has been described previously [Gagen et al., 2007] and was built from living trees. Here we extend the Laanila δ13C series to AD 886 using sub-fossil trees preserved in lakes. Both δ13C chronology sections are unusually well replicated with a minimum of five trees (see Table S1). Trees were processed in offset 5-year blocks such that the final chronology represents a 9-year centralized moving average built from eight overlapping cohorts (Figures 1 and S2). Since the offset in absolute δ13C values between trees is often large, compared to the inter-annual variability, absolute δ13C was fixed by sampling a much larger number of trees at the join-points between cohorts. Methodological details are given in the auxiliary material. We compare our δ13C series to a reconstruction of past April–August temperatures based on a maximum latewood density (MXD) record from Torneträsk (Figure 2) in northern Sweden, ∼400 km west from Laanila [Grudd, 2008].

Figure 1.

The δ13C chronology from Laanila, centralised 9-year (thin blue) and 30-year (thick blue) moving averages. Standard deviation is also given (n = 5). An insert panel shows the simple linear correlation between daily summer sunshine hours at Sodankylä and the Laanila δ13C chronology (1958–2001), both centralized 9-year moving averages.

Figure 2.

Daily summer sunshine hours reconstructed from Laanila δ13C 9-yr mean (dark blue) with uncertainty estimates (light blue), observed summer sunshine (July–August, 9-yr mean) (black) and April–August temperature reconstruction at Torneträsk (MXD) (pink). 30-year moving averages for the reconstructions are shown (thick blue and red lines respectively) and the LIA and MWP are indicated by dark grey shading. The mean for each series is indicated (black horizontal lines).

[5] The influence of climate on carbon isotope fractionation in trees is well understood [Farquhar et al., 1982] and the perturbing effect of tree age has less influence on δ13C than it has on other treering parameters, so that low-frequency climate information can be retained with less uncertainty [Gagen et al., 2007] providing the offset between trees is addressed. Over the industrial period (since AD 1850), trees have been exposed to large changes in the amount and isotopic composition of atmospheric CO2, for which corrections are necessary [Saurer et al., 1997; McCarroll et al., 2009]. Prior to this, treering δ13C records changes in the internal concentration of CO2 (ci) in the leaves or needles, reflecting the balance between stomatal conductance to incoming CO2 and photosynthetic assimilation rate [Farquhar et al., 1982; McCarroll and Loader, 2004].

[6] At cool, moist sites the dominant control over ci is assimilation rate, which can be limited either by enzyme activity, and thus photon flux, or by enzyme production, and thus leaf temperature [Beerling, 1994]. At Laanila summer temperatures are too high for this latter mechanism to be important [Luoma, 1997] and experimentation with Pinus sylvestris in eastern Finnish Lapland confirms that growing-season sunlight, not leaf temperature, explains 90% of photosynthetic rate variations [Hari et al., 1981]. Experimental evidence also suggests that tree leaf temperature is highly stable across a very wide geographical area [Helliker and Richter, 2008] such that temperature limitation of ci is unlikely to be the norm.

[7] The potential for treering δ13C from Boreal sites to thus record sunny summers has recently been identified [Gagen et al., 2007; McCarroll et al., 2003; Young et al., 2010]. Boreal treering δ13C results often correlate as well with instrumental summer temperatures as with records of sunshine hours or percentage cloud cover over the 20th Century, due to the co-variance of the two climate variables at high frequency but δ13C based temperature reconstructions tend to fail against long instrumental temperature records [Young et al., 2010; Hilasvuori et al., 2009] at multi decadal timescales. The assumption that the strong co-variance between summer sunshine and temperature seen in the calibration period is stationary at multi-decadal timescales, seems to be invalid and treering δ13C chronologies from high latitude tree line sites are better interpreted as records of summer sunshine or cloudiness [Young et al., 2010; McCarroll et al., 2003]. If the relationship between temperature and cloud cover has varied through time, δ13C series from northern Boreal pine will track variations in summer sunshine not temperature [Young et al., 2010].

[8] Meteorological records, available from the Finnish Meteorological Institute's Sodankylä station, 150 km south of Laanila (see Table S1), reveal that summer sunshine hours and summer temperature correlate highly (r = 0.72) between AD 1958 and 2001. In the common period, the correlations between Laanila δ13C and summer sunshine hours and temperature are comparably high (Table 1). Correlations between the Laanila δ13C series and summer % cloud cover (AD 1908–2001) are also strong but weaker than those between δ13C and temperature/sunshine hours (Table 1). The correlations with cloud cover also appear to be somewhat less time stable, which may be a reflection of notorious error in cloud cover measurements [Dai et al., 2006].

Table 1. Correlations Between Sodankylä Climate and the Laanila δ13C 9-Year Smooth Seriesa
Laanila δ13CCorrelation (r =), Temp Sodankylä, (July–Aug)Correlation (r =), % Cloud Cover. Sodankylä, (July–Aug)Correlation (r =), Sunshine Hours, Sodankylä, (July–Aug)
  • a

    Significance values (P < 0.05, in bold) were estimated using Monte Carlo simulations for the 9-year smooth series. Simple linear correlations for the annual Laanila δ13C series are shown in brackets. Statistical significance is not indicated for the correlations with sunshine data due to the low number of degrees of freedom, * = 2001–1958.

2001–19080.61 (0.74)0.56 (0.58)n/a
2001–19550.82 (0.79)0.66 (0.75)*0.84 *(0.74)
1954–19080.67 (0.71)0.66 (0.47)n/a

[9] A linear regression model was used to reconstruct summer sunshine hours (Table S2) at Sodankylä using the Laanila δ13C series as the predictor (Figure 2 and auxiliary material). The calibration of the Laanila δ13C record, to estimate Sodankylä sunshine hours, was accomplished using ordinary least squares regression, whereby the annual Laanila δ13C record [Gagen et al., 2007] was used as predictor and the Sodankylä sunshine hours meteorological record as predictand. The target meteorological variable was mean daily sunshine hours averaged over July–August and the calibration period was dictated by the available data at Sodankylä (AD 1958–2001). Diagnosis of the regression residuals indicated that they were not significantly serially correlated (Table S2). To obtain the reconstruction of the meteorological variable throughout the series length, the same regression slope and intercept as obtained in the annually measured calibration period were applied to the regression model used to develop the 9-year smoothed reconstruction. The uncertainty bounds for the whole period were estimated assuming the same lag-1 autocorrelation of the regression residuals as obtained in the calibration, and assuming that the sampling procedure of the δ13C record is indeed equivalent to a 9-year running average smoothing [Briffa et al., 2002].

3. Results and Discussion

[10] The Laanila summer sunshine reconstruction contrasts markedly with the Torneträsk growth season temperature reconstruction (Figure 2). The two proxy records are sensitive to slightly different parts of the growing season and we make the assumption that the relationship between April–August climate (Torneträsk reconstruction) and July–August (Laanila) would be the same in the past as now (highly correlated). The coldest century of the LIA in the Torneträsk reconstruction (17th), which is supported by other evidence to suggest cold summers at this time across the region [Gouirand et al., 2008], is also the sunniest century. During the warmest phase of the Medieval at Torneträsk (AD 950–1100), δ13C values were low, suggesting unusually cloudy summers. This presents a picture of a negative cloud feedback at this latitude.

[11] An explanation requires the regionally cool periods of the LIA to have also experienced enhanced summer sunshine. Support for a sunny Little Ice Age in Northern Fennoscandia is provided by past glacial dynamics in northern Sweden [Holmlund, 1986], which suggest low precipitation. A further δ13C series is also available from the Atlantic side of the Scandes Mountain range at Forfjorddalen in Norway [Young et al., 2010]. Laanila and Forfjorddalen are on opposite sides of the Scandes and thus do not correlate well at high frequency. However they do show a similar response to the LIA with high δ13C values also at Forfjorddalen [Young et al., 2010].

[12] According to our interpretation summer temperature and summer sunshine in Fennoscandia co-vary at interannual timescales. This is supported by the station data and by the plausible physical reasoning that at this location less cloud cover causes higher temperatures at inter-annual timescales. However, at multidecadal timescales, the proxy records indicate that cloud cover may change as a response to temperature variations caused by external forcing. This time-scale dependent behavior is reproduced in a millennial simulation with the global atmosphere-ocean model ECHO-G (see auxiliary material for model details) [von Storch et al., 2004; Zorita et al., 2005]. Figure 3 displays the multidecadal evolution of the simulated summer temperature, incoming radiation, and specific humidity, over Fennoscandia. The modeled inter-annual variations of Fennoscandian incoming shortwave radiation (“sunshine”) and temperature are positively correlated (r = 0.47, similar to the observed correlation), but the multi-decadal evolutions of incoming shortwave radiation and temperature are different. Although the Late Maunder Minimum (AD 1680–1710) is a period of weak external solar irradiance, stronger downwelling solar radiation at the surface is indicated over Fennoscandia, indicative of sunnier weather. Markedly lower temperatures are partly due to the lower atmospheric water content, which causes a diminished greenhouse effect (reduced incoming infrared radiation). Thus, at longer time scales, regional surface solar radiation and regional temperature can be decoupled due to the other factors becoming important at these time scales.

Figure 3.

ECHO-G GCM simulated time series (31-year running mean, anomalies from the 20th century mean) of (a) incoming shortwave surface(grey), incoming infrared surface (black) and total incoming (thick grey) surface radiation averaged over Fennoscandia. (b) Near-surface temperature (red) and specific humidity at 500 mb height (blue). (c) IPCC AR41 models, scenario A2, downwelling shortwave radiation July–August, deviations from the 2000–2010 mean, zonal mean (60°N–80°N) and (d) for the Laanila region (20°–40°E; 60°–70°N), 21-year centralised moving average.

[13] In the period between AD 1600 and 1700 (approx.) the simulated shortwave radiation does not diverge as clearly from the simulated temperature as the Laanila δ13C record does from the temperature reconstruction. Furthermore, in contrast to the proxy correlation between sunshine and temperature, the correlation between model shortwave radiation and temperature over the whole millennium at multidecadal timescales is close to zero. The time evolution of the modeled surface solar radiation is dependent on the representation of the regional details in the coarse-resolution global model and on the parametrizations of cloud microphysics, which are still imperfect in climate models. Therefore, a close quantitative agreement over the last centuries between modeled radiation and the Laanila record cannot be expected for one particular model. Nevertheless, the model results do provide a plausible mechanistic explanation for multidecadal-scale decoupling of regional summer temperature and incoming shortwave radiation in spite of their coupling at inter-annual time scales. This behavior represents a target for the forthcoming simulations of the past centuries in the next IPCC report, particularly if further sunshine proxy records can be retrieved from other sites worldwide. As an illustration of this type of combined proxy-model analysis, Figure 3 also presents deviations of incoming shortwave radiation in nineteen future climate model simulations from the IPCC AR4 archive, driven by the scenario SRES A2. Although the noisier regional long-term trends are not as consistent through time, most models do show less incoming shortwave radiation than present in the high-latitude zonal average and thus are consistent with the palaeo interpretation of an anticorrelation between temperature and shortwave radiation, in this region, at long time scales.

4. Conclusions

[14] Under the assumption that the cloud response at high latitudes is driven by the mean near-surface temperature, it would be expected that the behavior of cloud cover in a warmer future climate should be opposite to that of the Late Maunder Minimum. If, as indicated by the Laanila summer sunshine reconstruction, cloud cover was less in colder periods, and therefore incoming shortwave radiation greater, cloud cover at this latitude in a warmer climate should be greater and therefore incoming shortwave radiation less than at present. Most future climate simulations from the IPCC AR4 suite do display this behavior.

[15] Treering stable carbon isotope ratios from both northern Finland and northern Norway indicate a negative shortwave cloud feedback in this high-latitude region in summer in the LIA with the Laanila series presented here also showing a negative feedback in the Medieval, in line with model findings for the 21st century [Trenberth and Fasullo, 2009]. We hypothesize that information on past changes in sunshine could be extracted from treerings in other areas where assimilation rate, rather than stomatal water regulation, dominates the δ13C signal, which includes many tropical regions. Treering δ13C may thus provide a critical test of the ability of climate models to deal with the most uncertain feedback mechanism that still hampers accurate future climate projections [Bony et al., 2006].


[16] We thank Kevin Anchukaitis, Anders Moberg and Valerie Trouet for critical comment on the manuscript and the Millennium consortium for support. This research was funded by the European Union Millennium project (017008). MG is supported by a Research Councils United Kingdom Fellowship. NJL is supported by the British Natural Environment Research Council. Tarmo Aalto, Pekka Närhi and Reino Vierelä are thanked for field sampling and dating. Thanks to Nicola Jones for help with figures.

[17] Noah Diffenbaugh thanks two anonymous reviewers.