Strong coherence between cloud cover and surface temperature

Strong coherence between cloud cover and surface temperature variance in the UK E. W. Mearns* and C. H. Best * School of Geosciences, University of Aberdeen, Kings College, Aberdeen AB24 3UE Independent Scientist Understanding the future trajectory of climate change on Earth depends upon complex global circulation models that aim to integrate natural and manmade forcing of the climate system 1. Uncertainties remain in the magnitude of climate feedbacks 2 and the impact of natural Earth cycles such as changes in the pattern of ocean circulation 3,4 and of global cloud cover 5,6. Here we show that temperature variance observed in the UK since 1956 is closely linked to changes in sunshine hours which we interpret as an inverse proxy for cloud cover. Our data come from 23 UK Meteorological Office weather stations that have lengthy concordant sunshine and temperature records. From 1956 to 2010 the correlation between total annual sunshine hours and mean monthly maximum temperature is 0.8 (R 2 determined on 5 year means). We use a sunshine-surface temperatureforcing model combined with a CO 2 radiative forcing model to convert variance in sunshine and CO 2 to changes in surface temperature. The model is optimised with Net Cloud Forcing factor (NCF) set to 0.54 and Transient Climate Response (TCR) to 1.3 C. This provides a correlation

between model and observed temperatures of 0.85 (R 2 ), m=1.000 and c=0.01. Our empirically constrained TCR of 1.3 C is identical to the value reported by Otto et al 7. Most of the temperature variance in the UK can be explained by variance in sunshine/clouds and CO 2 forcing alone. The apparent absence of theoretical net positive feedbacks is perhaps an explanation for the pause in warming this century and the over-estimation of recent temperatures by global circulation models 8. Global circulation models (GCMs) have failed to predict the slowing of lower troposphere warming since the year 2000 8. This has given rise to debate about where the predicted excess heat, has gone 9. The greatest uncertainty for GCMs lies in understanding feedbacks to the climate system (from manmade and natural forcing) and in natural cyclic variations in global and local cloud cover 1. Global cloud cover variation measured by a number of satellites under the guidance of the International Satellite Cloud Climatology Project (ISCCP) is subject to uncertainty linked to data acquisition methods 10. An alternative approach to gathering information on cloud cover is to use surface sunshine records 11,12. This is an imperfect inverse proxy for actual cloud cover but may actually be an ideal proxy since sunshine at surface is controlled by line of sight absence of cloud between the point on the surface and the Sun. The impact that cloud cover has on surface temperatures is complex but for daylight hours it is widely accepted that low clouds prevalent in the UK have a net cooling effect through reflecting a portion of incoming solar radiation back into space 13. Since sunshine can only be used to estimate cloud cover during daytime our analysis has focussed on daily maximum temperature (Tmax). However, in our data set,

daily minimum temperature (Tmin), which invariably occurs at night, is closely correlated with Tmax with R 2 =0.93 (calculated on 5y means), thus conclusions drawn for Tmax may equally apply to Tmin or daily mean temperatures. Our data set comes from 23 UK Meteorological Office (Met Office) weather stations that have lengthy concordant sunshine and temperature records 14. These stations span the length and breadth of the UK with over 10 latitude of separation (Figure 1). At the beginning of the twentieth century only three of the selected stations were recording sunshine but by 1933 that number had grown to 13 providing representative nationwide cover. By the 1980s, there were 23 operational stations but that number has since declined to 19 due to station closures (Figure 1). Our analysis therefore covers the time interval 1933 to 2010 using centred 5-year means, which capture data from 1931 to 2012. Only those parts of records that report both sunshine and temperature are used. The Met Office reports total hours sunshine per month and mean daily Tmax per month. As is always the case with climate records, intermonth and interannual variability obscures trends. We therefore began our analysis by calculating annual means from the monthly data and then calculated 5-year means on the annual mean for each station in order to reveal decade-scale cyclic variability in the data. Doing so and plotting the aggregate sunshine and Tmax data as a time series reveals a pattern of strong co-variance between these two variables (Figure 2a).

On closer examination the co-variance since 1956 is much stronger than for the period 1933 to 1955. The correlation coefficient from 1956 to 2010 is 0.80 (Figure 2b). In contrast, from 1933 to 1955 there is no correlation (R 2 =0.00). Since sunshine variance and Tmax variance are expected to correlate, the lack of correlation for this period is unexpected. We suggest that this can be explained by the introduction of clean air legislation to the UK in 1956 that restricted the burning of coal for home heat and moved power stations away from the cities 15. Looking at only the summer months, June, July and August (JJA), when burning coal would be at a minimum, the correlation between Tmax and sunshine holds good for the whole interval 1933 to 2010 (R 2 =0.67 based on unsmoothed, JJA annual data) (Figure 2c and d). Since the pre-1956 annual data may be affected by air pollution, the remainder of this discussion will therefore focus only on the period 1956 to 2010. In order to evaluate the extent to which changes in sunshine hours may account for changes in surface temperatures we developed a model for the forcing of surface temperature by sunshine (see methods). The model accounts for daylight, insolation and NCF differences at the latitude and longitude of each station and aggregates these variables over the annual cycle. The model is calibrated by using measured Tmax in 1956 as the starting point and then adds or subtracts the calculated incremental change in temperature (dt) along the time series to provide temperatures calculated solely on variance in sunshine and cloud. Since our approach depends on comparing the trend of dt calculated from Sunshine variance with recorded dt, it is essential at the outset to have a good correlation between temperature and sunshine in order to test the model. These variables

show excellent co-variance on quarterly means for June, July and August (JJA; R2=0.67) and for 5Y running means (R2=0.8) but less good for mean annual data (R2=0.45). We therefore elected to use the 5Y means, but can show that similar conclusions may be drawn from the quarterly JJA data. NCF is used to describe the combined influence of cloud transmissibility and the radiative warming effect of cloud cover and associated water vapour. Output from the model using NCF values of 0.3 to 0.6 are displayed in Figure 3a, which shows that much of the structure of variance in Tmax can be explained by variance in sunshine and cloud cover. By way of reference, NASA report mean cloud transmissibility of 0.4 for the latitude of interest 16, NCF values greater than 0.4 combine cloud transmissibility with the radiative warming properties of cloud. To refine our analysis we introduce a CO 2 radiative forcing model (described in methods) with outputs for TCR of 1 to 4 C shown in Figure 3b. None of the CO 2 model outputs can explain the cyclic variance in observed Tmax but a TCR of ~2 C can explain the net rise in temperature in the period of interest. We combine the two models in order to optimise the fit of model output to observed Tmax using a combination of R 2, m, c and sum of residuals as measures of fit. Output for a range of permutations is shown in Table 1 that shows the sum of residuals are minimised with NCF factor set close to 0.5 and TCR close to 1. The optimal fit is for NCF factor = 0.54 and TCR = 1.28 C (Table 1; Figure 4) with m=1.000, c=0.01 C, R 2 =0.85 and sigma residual = -0.71 C uniformly spread along

the time series. The NCF factor of 0.54 compared with an expected cloud transmissibility of 0.4 16 provides an empirical control on the warming effect of cloud relative to occluding, cooling effect of transmissibility at this latitude. The relative contributions of cloud cover and CO 2 forcing on surface temperature variance may be considered in two separate ways. Net dt equals the rise in T from 1956 to 2010 while gross dt equals the total variance in temperature along the time series. Our model computes dt for each of 23 Met Office stations but not all stations were collecting data for the full period of our analysis (1956 to 2010, Figure 1) and this imparts some structure to the model. The effect of such discontinuous data series is small but significant and accounts for 19% of the observed net dt and 7% of the gross dt (Figure 3c). Variance in CO 2 accounts for 49% of the net dt but only 5% of the gross dt. Variance in cloud cover accounts for 32% of net dt but 88% of gross dt. In other words, variance in cloud cover accounts for nearly all the structure variance in UK temperature but somewhat less than half of the total temperature rise since 1956. The UK data suggest that TCR lies close to 1.3 C per doubling of CO 2 from preindustrial levels, similar to the best estimate of Otto et al 7 and significantly lower than the CMIP5 multi-model mean TCR of 1.82 C reported by Forster et al 17. Furthermore, most of the temperature variance in the UK can be explained by variance in sunshine hours and CO 2 alone and does not require the existence of theoretical net positive feedbacks that we conclude are likely absent. GCMs provide valuable and perhaps accurate tools for forecasting the future trajectory of surface temperature and climate on Earth but so far all these models have

failed to reproduce the observed temperature change this century 8. One reason for this failure may be the incorporation of positive net H 2O feedbacks, for which there is little direct physical evidence. We expect that substantial reduction of feedbacks, so that TCR~1.3 C, would improve the predictive power of GCMs. The cause of decadal to multi-decadal scale cyclic change in UK cloud cover remains unknown. We can find no clear links to recognised oceanic cycles such as the Pacific Decadal Oscillation or to atmospheric cycles like the North Atlantic Oscillation. Nor is there any link to sunspot cycles. There is however a coherence between variance in UK cloud cover and variance in global cloud cover as measured by the ISCCP with R 2 =0.37 (on global mean cloud amount 18 ) that is the subject of our on going research. It remains to be seen if the ground based results for the UK discussed here are replicated from ground based climate stations around the globe. Methods Summary Data The used portions of MET office climate station records used in this study are >99.5% complete. Missing data were normally patched using the equivalent data from the preceding year. Where there was no preceding year the succeeding year s data were used. Preliminary results have been accepted as final. No adjustment has been made to temperatures to account for the variations in altitude. The sunshine surface temperature-forcing model

The recorded annual sunshine for each station was used to derive an effective (daytime) cloud cover CC for each year, where CC= (4383 - total hours sunshine)/4383. We define the NCF factor as the resultant balance between increased albedo and LW warming effects for clouds in the UK. The radiative energy balance at each station is then given by 1 where, S 0 is the annual mean top of atmosphere insolation at the latitude for each station from NASA 16. Therefore for each year, y, the incoming net insolation is 1 and the calculated temperature change for T calc (y) is. where 3.5 Wm -1 C -1 is the Planck response DS/DT which is the increase in IR radiation for a 1 C rise in surface temperature. We initialize the model by setting T calc(1956) = T max(1956) and then calculate all subsequent temperatures based only on measured changes in CC. 1

The CO 2 radiative forcing model The change in CO 2 forcing for year (y) is calculated using the formula 19 5.3 ln 1 where CO 2(y) is the seasonally averaged concentration of CO 2 in the atmosphere for year y. CO 2 values are taken from a fit to Mauna-Loa data 20 smoothly extrapolated backwards to 1956. 1. Solomon, S. et al. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. (Cambridge Univ. Press: Cambridge [u.a.], 2007). 2. Bony, S. et al. How Well Do We Understand and Evaluate Climate Change Feedback Processes? Journal of Climate 19, 3445 3482 (2006). 3. Stouffer, R. J. et al. Investigating the Causes of the Response of the Thermohaline Circulation to Past and Future Climate Changes. Journal of Climate 19, 1365 1387 (2006). 4. Bond, G. Persistent Solar Influence on North Atlantic Climate During the Holocene. Science 294, 2130 2136 (2001). 5. Randall, D. A. Cloud Feedbacks. Frontiers in the Science of Climate Modeling (2006).

6. Randall, D. A. & Wood, R. A. Climate Models and Their Evaluation. (Cambridge Univ. Press: Cambridge [u.a.], 2007). 7. Otto, A. et al. Energy budget constraints on climate response. Nature Geoscience 6, 415 416 (2013). 8. Christy, J. R. et al. What Do Observational Datasets Say about Modeled Tropospheric Temperature Trends since 1979? Remote Sensing 2, 2148 2169 (2010). 9. Trenberth, K. E. & Fasullo, J. T. Tracking Earth s Energy: From El Niño to Global Warming. Surveys in Geophysics 33, 413 426 (2011). 10. Rossow, W. B. & Schiffer, R. A. Advances in Understanding Clouds from ISCCP. Bulletin of the American Meteorological Society 80, 2261 2287 (1999). 11. Muneer, T. & Gul, M. S. Evaluation of sunshine and cloud cover based models for generating solar radiation data. Energy Conversion and Management 41, 461 482 (2000). 12. Gul, M. S., Muneer, T. & Kambezidis, H. D. Models for obtaining solar radiation from other meteorological data. Solar Energy 64, 99 108 (1998). 13. Ramanathan, V. et al. Cloud-Radiative Forcing and Climate: Results from the Earth Radiation Budget Experiment. Science 243, 57 63 (1989). 14. Met Office: Historic station data. (2013).at <http://www.metoffice.gov.uk/climate/uk/stationdata/> 15. Clean Air Act. 4 & 5 Eliz. 2 52, (1956). 16. Kusterer, J. M. NASA Langley Atmospheric Science Data Center (Distributed Active Archive Center). (2008).at <https://eosweb.larc.nasa.gov/index.html>

17. Forster, P. M. et al. Evaluating adjusted forcing and model spread for historical and future scenarios in the CMIP5 generation of climate models. Journal of Geophysical Research: Atmospheres 118, 1139 1150 (2013). 18. Catalog. (2009).at <http://www.ncdc.noaa.gov/thredds/catalog/isccp/catalog.html> 19. Myhre, G., Highwood, E. J., Shine, K. P. & Stordal, F. New estimates of radiative forcing due to well mixed greenhouse gases. Geophysical Research Letters 25, 2715 2718 (1998). 20. Keeling, C. D. et al. Atmospheric carbon dioxide variations at Mauna Loa Observatory, Hawaii. Tellus 28, 538 551 (1976). Figure 1 Map showing the locations of the 23 Met Office weather stations used in this study. Annotations show: Station name, latitude N, longitude W, elevation in m. E=longitude E. The histogram shows the number of active stations that were recording temperature and sunshine data used in this study. Figure 2 Sunshine and temperature co-variance for UK weather stations. a Total annual hours sunshine and mean annual Tmax (both plotted as centred 5 year means). Note that co-variance is obvious from 1956 to 2010 but clearly absent pre-1956.

b Correlation between total annual hours sunshine and mean annual Tmax from 1956-2010. c Total sunshine hours for the months of June, July and August (JJA) plotted with mean Tmax for the same months. No further averaging or smoothing. Note how there is good visible co-variance between these variables for the period 1933 to 1955 that is absent in the annual data (Figure 2a). d Correlation between total sunshine hours for the months of JJA and mean quarterly Tmax for the same three months. The high amplitude, high frequency variability (Figure 2c) results in 5-year means that have lower correlation than the annual data. Figure 3 The three model components that combine to make our optimised model output shown in Figure 4. a Surface temperature variance for four different NCF factors. NCF values of 0.3 and 0.4 give rise to model temperature amplitude variations that are larger than the observed at the front end of the time series but provide a decent fit towards the back end. Values of 0.5 and in particular 0.6 provide a better fit at the front end but produce model temperatures lower than observed at the back end of the time series. b Surface temperature variance for four different TCR values. CO 2 forcing can explain part of the net rise in temperature but much less of the cyclic gross variance in temperature.

c As station records come and go (Figure 1) this imparts a bias to the observed data and to the model outputs shown here as the discontinuous data input residual. Figure 4 The optimised combined model output a Observed variance in Tmax for the UK compared with the optimised output of our combined sunshine and CO 2 forcing model (Tcalc) with NCF set to 0.54 and TCR set to 1.28 C. b The correlation for data plotted in Figure 4a showing gradient close to 1, intercept close to zero and R 2 of 0.85. c The residuals from Figure 4a, where the modelled temperature variance is deducted from the measured temperature variance. Table 1 Outputs for various experimental inputs to the combined sunshine and CO 2 forcing model compared with the observed variance in Tmax. NCF factors <0.5 require TCR values <1 to achieve a decent fit. TCR values >1.5 have net negative residuals. All permutations provide high correlation coefficient, hence R 2 alone is not a good guide to model fit. The optimised model (bottom line) has an excellent fit according to all parameters. R 2 = correlation coefficient, m=gradient, c=intercept, sigma res = sum of residuals, all outputs based on Microsoft XL for Mac 2011.