 Okay, thank you. Before I start, I would like to acknowledge my supervisors, Reto Knutting and Derek Fisher, who contributed to this work. I would like to open this presentation showing you this figure taken by a paper by Dazer et al. We're actually already a similar figure, I think on Monday during another talk, so I'll try to be relatively quick on this one. What the authors did in this work was analyzing a 40 member initial condition ensemble obtained by running CCSN3 under the A1B forcing scenario. And what they showed was that internal variability can have a substantial effect on the definition of future trends of both temperature and precipitation. So here we are looking at temperatures and on the top left panel, you can see the mean winter temperature trends, well the mean winter temperature change actually in the period 2005, 2016, as computed by averaging across the 40 runs. This can be, of course, regarded as our best estimate of the force response of temperatures, but as soon as we look at individual realizations of the model, we see that they can be very different. On the two panels below, we can see the warmest and the coolest among the runs. On the right hand side panel, we see different mean winter temperature anomaly time series as computed from the warmest and the coolest among the runs in red and blue respectively over three cities, that corresponded to three different grid points of different latitudes. And you can see, we can actually see the different effect of internal variability of different latitudes. And by aggregating temperature data over the United States and the whole globe. So when it comes to the global mean temperature, we see that internal variability doesn't really have, doesn't really play much of a role in defining future trends. But even if we look at a pretty huge region like that of the United States, we see that internal variability can actually enhance the future change in temperature or almost completely hide it. So the key point here is that different realizations of the very same climate can be very different. And this has to be taken into account also when comparing model simulations with observations. So how can we better compare models and measurements? One of the possible answers could be offered by dynamical adjustment. Dynamical adjustment is a tool, is a method which allows to remove the component of variability of a certain quantity of interest throughout this presentation, it's gonna be temperature. That is attributable to atmospheric circulation. So it consists of three steps, which I'm going to describe very quickly. As a first step, we need to sample internal variability of atmospheric circulation. And this is typically performed by applying an empirical orthogonal function analysis on certain fields. In this case, we are computing the two leading empirical orthogonal functions on monthly winter 500 hectopascalogic potential height over this Euro-Atlantic domain. As a second step, we can regress the associated principal components upon the values of temperatures, the time series of temperatures, grid point by grid point, then basically compute the mean effect of these modes of circulation and temperatures. And as a third step, we can dynamically adjust our measurements, our fields, by applying this formula. But TDA here are the dynamically adjusted temperatures which are computed by removing from the original temperatures, the contribution of the first and modes of circulation which are computed by multiplying the regression maps by their associated principal components. So this method can be applied to a pretty huge number of scales, both special scales and temporal scales. As a first case study, I would like to show you this temperature time series. This is the mean winter, well, this is the time series of mean winter temperature anomalies as computed over Switzerland in the period 1960, 2015. What's striking is the pretty strong cooling that Switzerland has been undergoing since 1988 amounting to approximately minus 0.037 degrees per decade and in contrast with the 0.25 warming in the period 1960, 2015. So what we can do is sampling circulation over the very same domain that we're showing you in the previous slide and estimating the effect of each mode on this very time series, on this data. And we end up with this new time series in Roshon and Rad, which of course has features wiggling of smaller amplitudes. That's almost by definition as we are removing part of the variability of the signal. But in particular, it's characterized by this warming trend of 0.27 degrees per decade, a pretty much in line with the 0.25 measured in the period 60, 15. As a further step, we can compare this observed temperature trend with the very same quantity as simulated in the five ensemble. So this is the observed value at the histogram of the mean winter temperature trends in the period 88, 15 as simulated by 40 simit five models. And although we see that there are some models actually showing a cooling, the simit five mean and the observed value are pretty much for a part. And this discrepancy can be accounted for looking, well, removing the contribution of temperature to temperature related to atmospheric circulation. And this is of course the dynamically adjusted value of the measured temperature trend. So we can apply the very same method, actually a very similar version of the method over the European domain. These are on the left you see the 1988, 2012 mean winter temperature trends in over Europe as estimated by the Berkeley data set. The map shows basically cooling everywhere with an average value of minus 0.42 degrees per decade. And after accounting for circulation, we end up with this new map with an average value of 0.44. I was saying a similar method because in this case we actually estimated the modes of circulation by performing an empirical autogonal function analysis on sea level pressure fields, but still pretty much the same. And again, we can compare the original and the dynamically adjusted value of the, well, the mean trend over the region with the very same distribution of the very same quantity as simulated by the summit five ensemble by 40 members. And also in this case we see that we can somehow reconcile the discrepancy between the absurd value and the summit five mean. So these results are of course sensitive to the number of modes we include in the when dynamically adjusting, but still they're not very much sensitive. So what else? Yeah, as a further step, we applied this dynamical adjustment method to test to what extent the observed atmosphere, the observed boreal winter cooling of a land masses in the period 1998-2012 could be explained in terms of atmospheric circulation. So the idea originally came from what was highlighted by Cohen and co-authors in a paper published in 2012 in which they showed that the global warming hiatus as at least as estimated in the spirit 1998-2012 was basically an asymmetric phenomenon. So characterized by a very strong cooling in the Northern hemisphere winter, especially over land masses, while in contrast more or less significant warming trends could be measured, detected elsewhere and in other seasons. So I think there's two panels pretty much make the point. These are the annual mean and DJF mean temperature trends estimated from era interim in the period 1998-2012 and we see that much of the cooling comes from the North America and Eurasia, especially over Eurasia. So what we did was applying, was sampling atmospheric circulation from sea level pressure variability in the domain 20, 90 degrees North and we saw that atmospheric circulation could in fact explain much of the observed cooling in boreal winter. So these are the temperature trends from era interim analysis in winter. In the top panel, we see the original trends and in the bottom panel, the dynamically adjusted trends. What we also noticed actually, was that atmospheric circulation couldn't really explain the whole iatres. Explaining by saying explaining the iatres, I mean I just mean reconcile the seasonal and annual temperature trends in this period 1998-2012 with the long-term trends. And an additional factor we found to play a quite an important role is coverage bias, basically meaning missing observations. So what we did was testing the effect of coverage bias as represented in the Hart-Cruity IV data set on the estimation of temperature trends on the seasonal and annual temperature trends in the iatres period, computed from five different reanalysis data sets. So this was actually done in a pretty simple way. What we did was simply, so here you see the January 2012 temperature anomalies in Hart-Cruity IV and in era interim. What we did, very simply was reinterpolating the reanalysis data set to the Hart-Cruity IV resolution and masking out on a monthly basis, all the grid points corresponding to missing observations in our measurements. So simulating the coverage bias in reanalysis. And we saw that this actually played a very important role in the estimation of the iatres across all the fiber analysis. So we saw a very similar response of the five data sets to this little trick. In particular, the effect of coverage bias was found to be related to the under-assembling of the Arctic region, which would have led to an underestimation of the effect of Arctic amplification on the temperature trends in this period. So before concluding, I just would like to ask this question to what extent is atmospheric circulation affected by anthropogenic forcing? This is, of course, a question which is of great interest by itself, but it also represents a caveat when it comes to applying the dynamical adjustment method I just described. This is simply for the reason that if atmospheric circulation is affected by anthropogenic forcing, as we remove circulation, we remove part of the force response of the system. We, this is actually pretty much work in progress. We have been performing a number of analysis, both working on several data sets, and especially on the Simit-5 ensemble. What we found is very short, that all the results I've been showing you so far are not affected by this issue, but that this issue might actually become a problem when applying this dynamical adjustment method in the future, in future simulations. So what we found, analyzing the Simit-5 ensemble, was basically a weak but statistically significant response of atmospheric circulation to anthropogenic forcing. In the second half of the 21st century and under the RCP 8.5 forcing scenario, so we're pushing it hard. So yeah, in conclusion, I hope I could show that this dynamical adjustment method could serve as a pretty useful tool for the estimation of the anthropogenic contribution to past trends, and it could also serve the purpose of a better comparison between models and observations. Furthermore, we could show using dynamical adjustment that accounting for both atmospheric circulation and the effect of coverage bias, as sampled from Hart-Cruity 4, allows to reconcile the hiatus seasonal and annual temperature trends to the long-term counter parts. So thank you for your attention.