 I'm Phil Jones, I'm in the Climatic Research Unit at the University of East Anglia in Norwich, in the United Kingdom. I did a first degree in Environmental Sciences and then did a Masters and a PhD in Hydrology. I finished in 1976 and in the UK that was the height of the famous drought, and so the water companies were not doing any hiring in that year. And I was keen to get a job and I was applying for lots of jobs and by chance I got a job at the Climatic Research Unit here at the University of East Anglia on a three-year research contract. Then 15 more years of short-term research contracts I was finally taken on as a reader within the university system and a professor a few years later. Now there was no planning in that, I was just a job at the time and it has worked out quite well I think, all things considered. Okay, I came here in 1976 and the first bit of my work was not anything to do with the temperature record, it was something else producing a quarterly magazine called Climate Monitor that we produced for another 20 more years. But at the time we'd just been in contact with several people at NCAR in Boulder, Colorado and learnt that someone in the US digitised something called World Weather Records which is a masses of volumes of just temperature data and rainfall data from stations around the world. So we managed to get a magnetic tape and I had a program for my PhD which did interpolation of data across, this was just for a small part of Britain and I was to apply that same program to the rest of the world and we thought it was a good thing to do because we had some gridded data sets of other variables like pressure and we were looking at changes in pressure patterns around the world we wanted to see how they impacted temperature and we were going to move on to precipitation later and so the first time we produced the land global temperature record was just interpolating the station data there was no check on the homogeneity of the data, the station quality we were just using the available data and we wrote up the work for a paper in a monthly weather review then and we just did the Northern Hemisphere and we just did the land and one simple way of displaying the results was to produce a large scale average for the Northern Hemisphere land areas but the whole aim was to produce a gridded data set to look at the patterns of change and relate those to patterns and other variables and it's still been that it's just that this one series has become a little bit iconic and later we moved on to the Southern Hemisphere land again and in the middle of the 1980s we knew we'd not done anything with the marine data so we added that in jointly with some colleagues at the Met Office and that was then the first sort of truly global temperature record combining both hemispheres and the marine parts of the world now people had done this before, I mean at the same time I'm being known to ask Jim Hansen was doing something similar at KISS, it was land only though but we actually reviewed who'd done it earlier and several people had done it earlier, there was a Russian data set at the time there was a guy called Murray Mitchell in the United States had done some work on this in the 1960s there was an even earlier work in the late 1930s someone called Guy Stuart Callender produced the data set and he also looked at carbon dioxide measurements before the record at Mona Loa started and if you go way back you find that a guy called Vladimir Kirpen who also developed the Kirpen classification of climate had produced a series in the 1880s and in some more recent work we've actually gone back and digitised some of the earlier series when people were clearly just working by hand I think Murray Mitchell in the 60s probably had some sort of computer calculator doing it but Callender did it all by hand and Kirpen certainly did so we digitised their data and they agree amazingly well I had a little paper out with Ed Hawkins a few years ago because it was 75 years since Callender's paper in 1938 last year and it was an amazing agreement with what Callender did and Callender, this was just for the Northern Hemisphere or the land areas but Callender was just using about four or five hundred stations around the world and he just had annual averages and we had sort of five thousand stations and we were getting pretty much the same results if we sort of look into the future and how global temperatures might go and the whole idea of climate change and the influence of humans on the climate system when we started the work in the early 1980s the temperature series then that we produced finished in 1981 actually showed cooling from the sort of late 1950s so the 1960s were and early 70s were quite a cool period there were some initial warmth coming in in the couple of years in the early 80s but we didn't really capture that in that data set so when we started doing it we didn't start it to look at that sort of longer term change we were trying to look at the reasons why you were getting some warm years and some cold years and the patterns of change around the world and could we relate those to the circulation obviously having looked at the data a lot more now you found that obviously the very warm years are often almost always El Nino years and the very cold years are often La Nina years and the really cold summers in the Northern Hemisphere are volcanic years when we're responding to big eruptions like Pinatubu so but from the mid 80s onwards the science moved to more climate change global warming and models and climate models but you always need the observations to provide some way of checking the models it wasn't just the models in isolation you had the observations checking how they're doing and then that moved into detection attribution we looked at this in a bit more detail because I realised in doing the original work that when we updated it we got some different extra stations there was people digitising stations even in the 1980s getting access to more data putting more stations in we realised that if you're just interested in the global average there had to be a finite number of series that you could use and get away with but we were always interested in the gridded product and if you wanted to improve that then you got to improve the number of stations everywhere so there was always a lot of stations in Europe and parts of the United States and parts of Australia and Japan and places but obviously there was never that good coverage in parts of Africa, South America and some parts of Asia so we've always concentrated on trying to get more stations for the less well covered areas of the world but in 97 I did some work with Tim Osborne where we tried to sort of quantify what the number was that you needed so in terms of, I came up with this concept of the effective number of spatial degrees of freedom so this is how many you need to actually produce an answer that's within the statistical error and really for the land it's probably less than 100 stations well situated around the world but obviously you don't have them all like that through time so you don't have sort of many long records to choose from in Africa or South America and you certainly don't in the Antarctic but that's the number, if you did have a well distributed set of stations then about 100 stations would produce you the answer for the global average and it would be indistinguishable from the numbers we produce now and in the latest data set in terms of combining with the marine data we're taking this concept a stage further we're using this bit of statistical theory to produce 100 realizations of the gridded data set and the hemispheric and global averages so where you've got very few stations those realizations will differ more from each other so over Europe and North America those realizations are almost the same because they're really robust so it's a way of quantifying the error in the data sets spatially as well and that's really useful for some of the people working on detection attribution issues so that they can then compare with the climate model output they've both got multiple models so they're used to using multiple realizations of climate simulations and they've got multiple realizations of the observations as well to work with and so they can use that way of displaying the error structure of the data in terms of those realizations the data set is a gridded product on a 5 degree latitude-longitude grid now if you have sort of in some parts of Europe and North America you've got 20 or 30 or more stations in a box so that average is very well constrained so there'll be a value everything's done in terms of anomalies from 1961 to 1990 so issues of elevation and distance from coast, etc. sort of disappear so the more stations you have the smaller the error will be and when you've got fewer stations the error is larger obviously you've got one station in the box you've only got one measure it's a large area, an individual box so what you do is you know for that box the standard deviation so you can draw samples from that which are the realizations about how temperature might have varied in the past and you know that how you draw them is just basically random there's very little sort of autocorrelation from one month to the next in most parts of the world I mean most people who now collect the data just collect what they might think of as the best estimate which is the 50th percentile of those realizations but it's very useful for the climate modelers to compare their observations with they know they have a range of possibilities of how temperatures have varied in the past so they can use that concept that we've added to the data set in their analyses well we don't do any of the measurements ourselves all the measurements are made by mostly the Met Services of the world or there are other organizations that do it in some countries and we get access to that data now there are a number of issues with the land stations the marine data is far more interesting and more important but the land stations there are two main issues with the land stations and most of these differ from station to station so they're not consistent problems from site to site so as one particular site you may have several moves of that site from different places in the town but they will be different from other places nearby also sometimes the observation times at stations change and they might also be different from place to place so what you need to do is you need to take into account these issues of site changes and observation time changes because they can make important effects on the data so for example if you've been measuring temperatures and you've been measuring them three times a day which was a common way of doing it in the 19th century so the common thing was to start reading at sunrise sometime at lunchtime about one or two o'clock in the afternoon and at sunset so you have three observations this was the common practice in Europe in the 18th, there in 19th century there were a lot of other issues related to that because Europe was not on, no place was on common time until the railways came we didn't get common time in Britain until about the 1830s when you needed to have common time when you had a railway timetable you didn't need it before then so solar time, you were on solar time everywhere so places were some way, in the larger countries places were some way out from measuring on a common time schedule but just think of measuring at sunrise, one o'clock and sunset it's clearly going to vary during the year you've got the seasonal cycle but if you then suddenly switched from doing that to moving to measuring it as daily maximum minimum temperatures when that thermometer became available in the middle of the 19th century you probably have a one or two degree difference which might differ from month to month there'd probably be a seasonal cycle to the difference so you've got to take into account these different observation schedules and some of them are much more complicated than the simple example I've given you there can be ones where you've got measurements every three hours and then suddenly they decide that in that country that they've got these maximum minimum thermometers they want to use those and they just measure that once a day they just need the observer to go out once a day rather than every, than eight times every three hours so that's the sort of observation time one the other problem is that over the years a lot of sites have moved to outside of towns often to airports now, a lot of their readings taken at airports and so you've got potential jumps in records when sites have moved from city centre sites to airports and you've got to take that into account well we call the process of making sure we've got just the impact of climate and weather on the observations we don't want the effects of human change in the schedule of measurements or where we've taken the measurements or even the screens that have been built around the thermometers we call that homogenisation and there's a well-known definition by a pair of climatologists in the 1940s called Conrad and Pollock who said a climate series is homogenous if it's only affected by the vagaries of the weather and climate so we're actually, by analyses we're making sure that these records are homogeneous one or two people have this belief that there's somehow this master data set of temperature data or precipitation data out there which we draw on well there isn't we have accessed the data initially from where the records as I was saying at the beginning we have then searched in archives particularly the Met Office archives because they had a lot of archives from the British Empire over the years measurements taken in many distant lands so we've got those then we've had contacts with other scientists and other Met Services to try and get additional data and there was a big impetus of getting extra data in 1950 after the Second World War so the coverage improved a lot then but there's still a lot more data out there that could be digitized in the 19th century and the first part of the 20th century which is coming along there's been a big emphasis recently to try and get more data digitized at the daily timescale and sub-daily timescale particularly the pressure data so that it can feed into new reanalysis products but that also has helped in getting some of the daily temperatures and rainfall also digitized okay so the different groups have got different data sets so we have exchanged our data in the past with NCDC and Nashville and that is the main American group in terms of there are three American groups in NCDC and Nashville and GISS part of NASA in New York and this new one called BEST with the somewhat contrived acronym Berkeley Earth Surface Temperature now GISS as far as I understand from reading the papers uses the same station data that NCDC uses and they apply an additional adjustment for urbanization but essentially the data set is the same as what NCDC produces BEST take a number of data sets from NCDC now having got the basic data together they then NCDC and BEST do some homogeneity assessment of that quality of the data and provide adjustments to the station data we're a bit different we did some work on that in the 1980s and we realized that the best people to do that were the MET services themselves so we've encouraged MET services to do it and more and more of them are doing it it's more the developed nations that do it than the developing nations hardly any countries in Africa do it for example so once you've got the basic data the other difference is how you combine that into a grid or hemispheric average and GISS uses one method NCDC uses another method which is similar to ours now and BEST uses a statistical interpolation scheme involving cryging but I don't think that aspect doesn't make much difference at all GISS and BEST managed to do it without having to have a base period for the station so they somehow get over the problem of stations being at different elevations and also measuring temperature in different ways but we use anomalies so we have to have data the station has to have enough data from the 1961 to 1990 period so if we have a station that's only got 30 years in the 19th century then we don't use it because it hasn't got the 1961 to 1990 base period the other techniques seem to be able to use that data but I don't think there's much of that data the biggest issue with the global temperature series and I'm talking here of the global temperatures of land and ocean is from the marine data if you go through the answers I've given you so far in terms of the land a lot of the issues are different from station to station so they're not common from station to station urbanization may be a slight factor in some regions of the world but we think that's relatively small and so if you average enough stations together and they are reasonably reliable the land record will agree quite well the land is always going to be much more noisier as well than the marine part and with the marine data, when the work we've done on the marine data and particularly the work the Met Office has done on the marine data there are a number of key changes to the way temperatures were measured at sea in the past first of all the land data is all air temperatures measured one and a half to two meters above the ground and so to measure air temperatures measured by ships the Met Office found that the ship, the data during the daytime was just not not very reliable it was affected by the heating up of the ship particularly when you had sort of modern ships well steam ships from the late 19th century onwards so they removed the daytime data and so that's halved the data set to start with so you've only got the nighttime ones and so what we've always done is try to go to the sea temperature data and the reason for that is that sea temperatures sea doesn't change much from day to day so in a given square of the ocean you don't need too many observations so on a land station you need observations twice a day at least to pick up the journal cycle and you need observations every day because it varies a lot from day to day in most parts of the world but in the marine part of the world you can probably get away with three or four observations in a month it will give you the average sea surface temperature for that bit of the ocean because it doesn't change much from day to day you're not measuring the immediate skin temperature of the ocean you're measuring it some way about anywhere from the surface down to about five or six meters so the top sort of top layer that top layer of the ocean doesn't change too much but anyway the way that measurement of sea temperature has been made has changed over the years and particularly it changed around the start of the Second World War what's done, what's started the Second World War and continued since is measurements of sea temperature taken with engine intakes so ships, steam ships take on or powered ships in some way take on water to cool the engines slightly bit in the nautilus to how a car takes on air and by putting some thermometers at the intake at the sea intake pipe then you can have those measuring directly on the ship's bridge and it's much easier to do so the captain or the mate can fill in the logbook without having to go and take a real sample of seawater but prior to the Second World War most countries weren't doing that and the measurements were taken with a bucket so there was a bucket of various designs obviously with a rope attached to them and you throw the rope over the ship's side and with the bucket on and you brought up some water and you put a thermometer in and you measured the temperature and the rules, the recommendations were left that thermometer in the bucket for a few minutes for the thermometer to equilibrate with the water or depending on the bucket design that water is going to cool because normally the air temperature is cooler than the seawater that's being sampled over most of the ocean it's not everywhere but it is in many places and so there will be, there was a cooling the bucket temperatures were tended to be somewhere between about 0.3 and 0.7 degrees Celsius cooler and that jump, that change took place around 1940-41 and so there's a massive jump in the sea temperatures if you don't adjust them for that homogeneity problem in the sailing ship period in the 19th century they were using buckets as well and they tended to be wooden buckets and they're better insulators so they had slightly less of an evaporative cooling that came with steam ships so if you took the marine data and put it all together without doing these adjustments you'd find that there was a massive temperature increase because the bucket data from about 1890-1940 is about 0.5 degrees Celsius colder than it should be and it's slightly warmer by a few tenths in the late 19th century and then you've got the modern stuff which is nearer the true temperature so if you didn't make that adjustment you would have a massive warming much greater than you see over the land in the marine data so the biggest adjustments to any of the components of the global temperature data is the sea temperature adjustments around the beginning of World War II and if those adjustments were not made then the air-sea temperature differences would be fine after World War II and then completely wrong before World War II so you cannot use unadjusted data so I put it in another context there's been a lot of... I've read things about the New Zealand temperature record and people claiming that the New Zealand Met Service Niwa have made adjustments to data going back into the 19th century well this is to account for the site changes and so you end up with a record in New Zealand which shows relative warming throughout much of the 20th century and that agrees with sea temperature measurements from the adjusted sea temperature data set around the coasts of New Zealand if you didn't make the adjustments to the land data the land data would have no warming if you didn't make the adjustments to the marine data the sea temperature would have about twice the warming it currently has and so everything in New Zealand before the Second World War would just be completely ridiculous because you'd have air-sea temperature differences that are totally wrong compared to the period after the Second World War roughly because of the big adjustments to marine data so really the really important change the biggest change to any of the components of the global temperature record is in the marine data for the change from buckets to intakes and this is really important also the marine data changes are more important also in recent times too because the numbers of ships taking measurements around the world has reduced slightly there are a number of reasons for that some shipping companies no longer want to do it there's always been issues with fishing fleets not knowing not wanting their competitors to know where they are and the other thing is that ships think that some companies think that sending out the data with the ship's core sign tells pirates off Somalia where they are so in order to improve though before that in order to improve weather forecasting in the Southern Hemisphere and parts of the Northern Hemisphere people realised that improvements to sea temperatures were really vital and so since the late 1970s a lot of buoys have been deployed around the world particularly in the Southern Hemisphere and the tropics and this provides these buoys provide measurements of air temperature sea temperature and pressure it turns out that these buoy data the Americans call them buoys tend to record temperatures slightly cooler than the ships of somewhere in the range of one to two tenths of a degree Celsius when they've compared co-located measurements so over the last 20 or so years we've gone from a maybe about the last 25 years since about 1990 we've gone from a marine measurement marine measurements coming almost entirely from ships to one where about 80% of them are coming from these buoys and so they're now allowing for that in the data sets because you've got to take into account that the buoys have a slightly different absolute temperature which is probably nearer the true one than the ships it may be that the ship intake measurements are probably about one to two tenths of a degree Celsius warmer than they should be which is what a lot of people have said in the past it doesn't make any difference this slight difference in absolute temperature doesn't really make too much difference to us because we're using temperature anomalies it's only if you want to try and go back to the absolute measurements in terms of a real degree Celsius rather than just temperature anomalies yes so we did some work on this sort of the 1940s has always been an interesting period in terms of the course of temperature change on a global basis but what you've got to bear in mind is that the first part of the 1940s was the Second World War and the number of observations was markedly reduced from what was available during the 1930s and certainly what was available in the 50s or even in modern decades as well so the marine data has a greater error range and in our realizations of global temperatures or gridded temperatures that I talked about earlier that is encompassed that greater uncertainty of the marine data is encompassed within that it doesn't help also with the there was a major El Nino event in 1940-41 which we would like to know a bit more about but we've tried the Met Office and others have tried really hard in trying to digitize as many ship observations that we can find, ship logbook observations so almost all the extra British ships during the Second World War have been digitized in the last 10 years or so so they have improved coverage but mainly in the Atlantic and the Indian Ocean there weren't too many British ships in the Pacific although we did notice one intriguing thing with the British ships that at the start of the Second World War Britain Churchill made a deal with Roosevelt in that for giving the Americans leases on bases around the world we got a number of old American ships and these ships we hadn't got the parts for a lot of these ships so when they needed repairs you suddenly see these British ships in the South Indian Ocean suddenly shoot off to San Francisco or Seattle for repairs and you can see this in the logbooks and so there is some British data in the tropical Pacific but very little it's just when they had to go off running repairs during the Second World War in terms of interesting questions it's probably for me it's nothing much to do with the global temperature record I do wonder sometimes that when I did that in the 1980s I should have moved on to something else but it's obviously become quite a thing to keep going one aspect is trying to get more data in but knowing that it's only really going to help in certain parts of the world we have looked at other ways of combining the data we're also doing a lot of work on trying to do similar data sets for precipitation which is clearly a quite different variable to temperature so when I said that 100 stations would produce a reasonable global land temperature estimate for temperature then for precipitation you would need thousands and thousands because it's much more spatially variable than temperature and trying to get access to that sort of quantity of data is quite difficult so you can only really produce reasonably reliable precipitation grids in the data dense parts of the world so you can get ideas of what might be happening in Africa but it's got such large ranges attached to it so we have produced another data set which people say that the crew data set doesn't have gaps where there's no stations that's true but there's another data set we have that a lot of people don't know about an infill data set where we do this infilling and we also do that for temperature and a few other variables too and that data set gets widely used by many people particularly the climate modelers to assess how their climate models are doing in an absolute sense because we put that back to absolute degrees Celsius too and rainfall in millimetres so they compare how well their model is doing both over time and sort of things on the seasonal cycle and many other aspects of the sort of agreement between temperature and rainfall and pressure to see if the model reproduces the same sort of patterns that the real world has If I'm giving a talk in town X it would be useful to know what the temperature record for town X looks like so that question was always asked and then there would always be a question about urbanisation and people think urbanisation causes much of the warming but urbanisation doesn't cause warming of the marine data and the trends are pretty similar between land and marine data Another question people always asked about is the issue we've already talked about is how many stations you need to produce reliable records Now the fact that I said that you can get away with about 100 stations that's at the monthly timescale so the number you need is dependent on the timescale so if you're looking at daily data for temperature then you will need many more stations to do it reliably and precipitation is a different variable entirely it's much less spatially conforming it's more spatially variable so you need more stations and as you go further back in time to the pre-instrumental period when you start using proxy data from ice cores, corals, trees, etc if that number was significantly more than 100 then you wouldn't be able to produce these reconstructions in the past and obviously as you go further back in time on longer timescales onto the 1000 year and 10,000 year timescales that we think about with ice ages then it seems people are quite happy to accept one record from Greenland and one record from Antarctica about the course of ice ages over the last million years they're not worried about what happens elsewhere so it's the actual timescale dependence of a number of these metrics and particularly this number of stations you need is crucial and I think that's one of my best papers that I've ever done in 1997 with Tim on that issue and the other groups do not they're aware of that but they've never really incorporated the error component in their data the best people have but I think their errors are too small so the climate is going to continue to warm but it's not going to warm year on year because it hasn't done that in the past there have been periods when it's warmed there have been periods when it's cooled slightly and there's large changes from year to year so the climate change component from greenhouse gases is really on the decadal and longer timescales you shouldn't see it on the individual years so what's really influencing the individual years is the circulation which is where we started and it's really the tropical circulation you only know as La Nina's, that's the dominant one but in different parts of the world there are components from the North Atlantic oscillation in Europe and North America and you've got the southern annular mode in the higher latitudes of the southern hemisphere so how that circulation changes is a big component on the temperatures on the sort of inter-annual timescale but that's just year to year we wouldn't expect La Nina events and La Nina events to contrive sort of long-term warming that really has to be down to greenhouse gases because we know that the sun has varied but the variations in solar output are relatively small and they should have caused the world to cool slightly since the late 1950s but in fact that hasn't happened