This is Rob Dekker's Typepad Profile.
Join Typepad and start following Rob Dekker's activity
Join Now!
Already a member? Sign In
Rob Dekker
Recent Activity
Needless to say that I think the conclusions from Ding et al 2017 are flawed. The valid conclusion that I think we can draw from the Ding et al 2017 paper findings is this : 60% of Arctic sea ice reduction is caused by summer-time climate change, while 40% is caused by climate change over the remaining 9 months. Which is a very interesting conclusion by itself.
Al Rogers I think it presents a useful result but that result is presented in such a way that it can be used to suggest that natural variability is responsible for perhaps half of the loss of Arctic SIE recorded over the last four decades - a suggestion which is patently false. You are right, but still it is a challenge to actually show where the paper goes wrong. I think I found the problem with Ding et al 2017 in this comment :
I dropped the same comment on William Connolley's site since Eric Steig (a coauthor whom I respect very much) commented there. Let's see what the response is.
Toggle Commented Mar 21, 2017 on Lowest maximum on record (again) at Arctic Sea Ice
For all the words written in this paper (Ding et al 2017) and all the correlations and experiments they performed it is surprising that they did not even investigate the most obvious test of AGW attribution of them all : A correlation between summer temperature and sea ice extent.
Toggle Commented Mar 21, 2017 on Lowest maximum on record (again) at Arctic Sea Ice
Jim, thank you so much for reporting on the Ding et al 2017 paper. I just read it in detail and would like to report my findings here. The essence of the conclusion (attribution to AGW) of the paper lies in this section : to estimate the anthropogenic contribution to the observed warming and sea-ice reduction in t he Arctic, two additional experiments are conducted. Exp-7 and 8 are equivalent to Exp-2 but we remove t he effects of global warming on the high-latitude winds, which are used to constrain the model in Exp-2 (Supplementary Fig . 8). These results show the same strong geopotential height increases as in Exp-2, with approximately 70% of Arctic low-level warming and sea-ice extent change (north of 70◦N) relative to Exp-2. Hence, these experiments suggest that ∼30% of the anomalous thermodynamic sea-ice extent reduction is attributable to anthropogenic influences on the Arctic circulation. Applying this estimate to the overall circulation-driven sea-ice trend established in Exp-6 (60%), we estimate that about ∼42% (70% × 60%) of the sea-ice decline observed since 1979 in September is due to internal variability. Now, both these fractions (70% and 60%) are questionable. First of all, the 60% number refers to the correlation between sea-ice trend and atmospheric circulation over the Arctic. However, that does NOT say how much atmospheric circulation over the Arctic is influenced by temperature. Since higher temperature means expanding air mass, geopotential height will always increase with increasing temperature, which their own findings in figure 1e of the paper clearly shows (best correlation of geopotential height at 200 mb is with temperature). So that 60% influence of atmospheric circulation can very well be simply caused by atmospheric temperature increase, which can easily be AGW in origin. After all, Z200 is high up in the atmosphere, which means that even during summer it is not much influenced by melting sea ice below. And the 70% (natural variability) refers only to the influence of “high-altitude winds”. Here, again, high altitude winds (such as the jet stream) are caused by geopotential height, which is again caused by temperature changes over the Arctic. If the Arctic warms more than the rest of the planet (due to albedo feedback or increase in moisture or any other reason), the geopotential height over the Arctic will increase more than the rest of the planet, and thus the high altitude winds will be less “cyclonic” than otherwise. That means this “70% (natural variability of the atmospheric circulation over the Arctic)” may very well be completely caused by temperature increase. So both numbers are highly dependent on temperature increase, and since the paper does NOT investigate the correlation of these variables to temperature increase, even though its own analysis establishes that correlation very clearly (fig. 1) its conclusions are not sustained by the evidence they collected. As William Connolly wrote (thank you Jim for reporting on that : they then convince themselves that most of the circulation changes that matter to the ice are not GW forced, and so must be natural variability; and hence the conclusion.
Toggle Commented Mar 21, 2017 on Lowest maximum on record (again) at Arctic Sea Ice
Rascal, thank you for your assessment of the PIOMAS volume data, as it reflects on the upcoming melting season. In short : Yours are intimidating numbers. A bit scary I may add.
Toggle Commented Mar 20, 2017 on Lowest maximum on record (again) at Arctic Sea Ice
Jim said : Rob - Don't forget the reanalyses from ECMWF: .... and NCEP Yeah. Thank Jim, you just tripled the amount of work to be done to find FDDs for past winters :o)
Toggle Commented Mar 16, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Bdwo, from your first link : The ice that survives at least one summer melt season is typically thicker and more likely to survive future summers. My point exactly. That's the ice in the Central Arctic, which tends to survive multiple melting seasons...
Toggle Commented Mar 16, 2017 on PIOMAS February 2017 at Arctic Sea Ice
@P-maker What is missing, in my view, is an indicator of the “Freezing Power” in the Central Arctic Ocean, Actually what is missing is an indicator of the "Freezing Power" in the marginal ice zone : the area of the Arctic that tends to melt out. The Central Arctic is less relevant, since that's where the MYI hangs out ; the ice that does not affect the minimum "extent" in September that much. @Bill Thanks for the additional notes on area "weighted" temperatures, versus unweighted DMI numbers.
Toggle Commented Mar 13, 2017 on PIOMAS February 2017 at Arctic Sea Ice
There probably is a correlation between 2m temps and 925 hPa temps. Slater has both temps posted. The tricky part appears to be to retrieve these temps from CFS(v2). It looks like we need at least a GRIB2 parser and then need to figure out exactly which files (and at which index) contain the temp data, 2m or 925 hPa :
Toggle Commented Mar 12, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Wait a minute. Slater also reports 2m temps : with the following note : The +80N T2M images on this site are similar to those shown by the Danish Meteorological Institute (DMI) except that I use the NOAA model (vs. ECMWF) and area weighting is applied. That is encouraging. Let me dig a bit further into that data. Also, Nico (Tealight) posts FDDs on Neven's graph page : which suggests he uses the DMI data.
Toggle Commented Mar 11, 2017 on PIOMAS February 2017 at Arctic Sea Ice
P-maker, indeed. Even if we work out these irregularities in the DMI data, it still does not serve as a good indicator of the ice thickness in the melting zone (70-80 deg), because the data is not area-weighted. For example, in the DMI temperature graph (on a 0.5 deg grid) attaches 40x the significance to the temperature within 0.5 deg of the NP as compared to the significance of a similar area at the 80 deg North lateral. And since area/distance around the NP is a quadratic function, the DMI graph is for 50% determined by the temperature between 87.5-90N, which is only 25% of the area North of 80deg. And Slater's data seems to be area-weighted, but as you correctly point out it is 925 hPa, which is too far from the surface to be indicative of ice growth during winter. I can (and will once I have the time) still run the correlation between these data sets and the September sea ice extent, but I think it is simply is too far from the melting zone to be significant. I have more hope for the PIOMAS ice thickness south of 80 deg, but for that I need access to gridded PIOMAS data, which requires some work (and more importantly : time, which I chronically lack).
Toggle Commented Mar 11, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Thanks P-maker, For the moment, I'm still confused about that DMI temperature data (as a source to calculate FDDs from past years). I'm taking with Jim about this starting here : One thing that is revealing already is that that DMI 80N temp graph that we use here at Neven's so often : appears to be not "area weighted". Specifically the following declaration from DMI is important : However, since the model is gridded in a regular 0.5 degree grid, the mean temperature values are strongly biased towards the temperature in the most northern part of the Arctic! Therefore, do NOT use this measure as an actual physical mean temperature of the arctic. Humbling thoughts...
Toggle Commented Mar 10, 2017 on PIOMAS February 2017 at Arctic Sea Ice
For Frank : Two over-due comments : (1) I just realized that the correlation table I presented in this post : is the correlation of absolute sea ice 'area' in September against any of the (June) variables mentioned. The correlation between Ice MELT between June and September and the variables mentioned (which refers to the "feedback" of the variables) is smaller. When used as a 'predictor', the standard deviation of the residuals of course remains the same : About 340 k kM^2 for the June->September prediction. (2) Regarding Petty et al, thanks for the link ! It seems an extension to the work done by Schroder, who is also a contributor to the SIPN Arcus network. The issue I have with their method is that the standard deviation over their predictions is not much better than a simple linear extrapolation of sea ice melt. They produce 500 k km^2 standard deviation over the June->September prediction, while my (simple) statistical method sits at 340 k km^2. Yet their predictive methods for months before June (like March and April) exceed my method.
Toggle Commented Mar 9, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Thanks guys, DMI North of 80 FDD numbers are worth checking out (for correlation with Sept extent). But I realized something : What we are really after is thickness of FYI. After all, the majority of ice that melts out each year is FYI. So rather than assuming a super-simple model (taking the SQRT of a measure (FDD) of freezing) from an area that contains mostly MYI, why not find a much better model that produces FYI thickness : PIOMAS ! What I'm after for that experiment is the sea ice thickness from PIOMAS for the areas that are known to be FYI : The areas outside of the prior year September sea ice minimum. So I need to go to a gridded source of PIOMAS.
Toggle Commented Mar 9, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Thank you Neven. That is the best overview of the differences between PIOMAS and CryoSat-2 that I ever read. And let me add that I share your opinion that PIOMAS is probably closer to the truth. It is much harder to determine freeboard from 150 miles (?) high with cm accuracy than it is to model sea ice growth if you input atmospheric data (temperature, wind etc).
Toggle Commented Mar 9, 2017 on PIOMAS March 2017 at Arctic Sea Ice
Bill, yes, FDD would be a nice variable to include in the prediction method. To follow physical behavior it should be included as SQRT ( FDD ) in the formula, since FYI thinkness roughly (theoretically) relates to the square root of FDD. The main issue is : Where do I find FDD for (many) Arctic winters past ?
Toggle Commented Mar 8, 2017 on PIOMAS February 2017 at Arctic Sea Ice
you guys have me seriously confused : Bill said to wayne : I'm not sure how you arrived at the 286 kJ/kg figure wayne said (in this post) : Sea ice latent heat of fusion = 3.37e17 j/km3 US fisheries and agriculture department Which is 337 kJ/kg. Did wayne just change that number, Bill ?
Toggle Commented Mar 7, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Frank said : since "year" is such a strong factor, should it not be included in the analysis? To reflect, as you say, the long-term trend. If we include the "year" variable, then indeed the correlation goes up a little bit more. However, time by itself does not increase heat in the Arctic. What I wanted to do is to see how we can explain the decline in Arctic sea ice using PHYSICAL variables only. Specifically the ones that we KNOW have an effect on heat absorption : sea ice "area", water close to ice (extent-area) and land snow cover. And with these, with the proper weighting factors, we already obtain R=0.94, explaining (R^2) 88 % of the variability between June and September. Which is pretty darn good. I've contemplating using other physical variables (like temperature during winter; which relates to the thickness of FYI and thus how fast it will melt out in summer) but I don't want to include too many variables, since as the saying goes : "With four parameters I can fit an elephant, and with five I can make him wiggle his tail".
Toggle Commented Mar 6, 2017 on PIOMAS February 2017 at Arctic Sea Ice
For those of us not familiar with statistics, the standard deviation of the residuals of a linear regression between June data and September data is simply a measure of the uncertainty in the prediction.
Toggle Commented Mar 5, 2017 on PIOMAS February 2017 at Arctic Sea Ice
RobertS, are you talking about the "residuals" ? The noise that is left over after linear regression ? If so, you are right that "demons that are lurking in the error bars", but in our case it really seems to be noise (most likely due to variability of summer weather. The standard deviation of the residuals (1992-2015 data regression with my formula) is 340 k km^2. And no, it does not have a 'trend', since that would have been taken out already by linear regression.
Toggle Commented Mar 5, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Frank, I wrote my little program in 'awk' (remember that one? :o). For anything more complex than linear regression, I know I have to switch to something better. I really like R with all its statistics support but did spend enough time with it to feel comfortable coding in R. I'll take a look at that Python package, but I'm not fluent in Python either. I may resort to C/C++, since that is my life. It would be great if you could give it a shot. For the 'extent' and 'area' for all months I use NSIDC (on Neven's 'graph' page) : For land snow cover I use Rutgers Snow Lab monthly data : It would be great if you could confirm my findings for June data as a predictor for September ice "extent" (R=0.94 with my formula). And correlation is even better for ice "area". Here are the correlation factors against ice melt (Sept 'extent' - earlier "area") if we use only individual variables in prior months as melt formula F : (using 1992 - 2015 data). April May June F = area R=0.19 R=0.54 R=0.90 F = extent R=0.10 R=0.20 R=0.76 F = snowcover R=0.73 R=0.81 R=0.90 F = extent-area R=0.25 R=0.5 R=0.69 F = year R=0.87 R=0.87 R=0.87 Note that land snowcover in June is a good projector for how much ice will melt out until September, and so is "area" in June. But for earlier months, the best predictor is "year" which is simply reflecting the linear trend of reducing sea ice. Explicitly note that ice "area" and "extent" in earlier months (April) are very poor predictors for ice melt until September. I think someone upthread asked about.
Toggle Commented Mar 5, 2017 on PIOMAS February 2017 at Arctic Sea Ice
zebra said If it's OK for people "on our side" to make speculative arguments with no physical basis, and dismiss the "rules" that got us to where we are in science and technology, Please, zebra. Nobody made any speculative arguments with no physical basis. My analysis specifically confirms the albedo feedback effect : the darker the Arctic, the more energy is absorbed by the surface. I made an educated guess on the fractions of absorption and the regression results on actual ice melt confirm these. And neither did anyone dismiss any "rules" of science and technology.
Toggle Commented Mar 4, 2017 on PIOMAS February 2017 at Arctic Sea Ice
Thanks for the nice words, guys ! Very much appreciated. Frank said : Now S = a0 + a1.M1 + a2.M2 + a3.M3 From my point of view I find it clearer to see it this way. We would find the best fit coefficients using "least squares", that is, minimizing the sum of the squares of the residuals. This is multivariate linear regression, and would be a more standard approach. Yes, I realize that multivariate linear regression would have been a more standard approach. For one, I would not have to manually tweek the parameters to obtain the best correlation, which was rather time consuming as you can imagine :o) I choose simple linear regression on a single variable (melt function F) simply because I had a program written already that runs flawlessly on the data at hand. I contemplating using Principal Component (PC) analysis, but since statistics is not my main area of expertise, I did not feel comfortable writing a new program that may or may not contain bugs. Do you know of any publicly available implementation that can do multivariate linear regression (or PC analysis) that can be used to solve this problem ? Either way, thank you much for your feedback, and I hope you will 'unlurk' more often. Your insight (as a mathematician) contributes to the quality on this fine blog.
Toggle Commented Mar 4, 2017 on PIOMAS February 2017 at Arctic Sea Ice
wayne said : Not one single chance that there is 50% constant SW to surface, cloudy not cloudy, sorry that concept is extraterrestrial. Not observed by me in any way. Did you measure downwelling SW yourself ?
Toggle Commented Mar 3, 2017 on PIOMAS February 2017 at Arctic Sea Ice