Guest post by Dr. Walt Meier
Steve Goddard has written several contributions on sea ice lately, particularly on the PIPS model, and as expected there has been much discussion about sea ice as we’ve entered another summer melt season. I can’t possibly comment on everything, but I will provide some information on a few points. In this post, I’ll tackle the PIPS and PIOMAS model issues.
In a following post, I’ll address the other three issues. I include several peer-reviewed journal references for completeness and to give a sample of the amount of research that has gone into investigating these issues. Note that as usual, I’m speaking only for myself and not as a representative of the National Snow and Ice Data Center or the University of Colorado at Boulder.
PIPS 2.0 and PIOMAS
When I saw PIPS being mentioned, it brought back fond memories for me. I haven’t worked with PIPS recently, but several years ago, I was a visiting scientist at the U.S. National Ice Center (NIC). NIC is a joint Navy, NOAA, and Coast Guard center whose primary duty is to provide operational support for military and civilian ships in and near ice-covered waters.
NIC is the primary customer of the PIPS model outputs, which they’ve used the operational forecasts to help produce their operational ice analyses. As a researcher at NIC, one of the projects I was involved with a project to evaluate the operational forecasts. I was a co-author on a couple peer-reviewed journal articles (Van Woert et al., 2003; Van Woert et al., 2001), where we found that the operational forecasts showed some skill at predicting ice edge conditions over the following 1-5 days, but the forecasts had difficulty during times of rapid ice growth or melt. (Steve referenced one of the papers – thanks Steve!). So I can perhaps clarify and explain some issues about PIPS and its applicability for studying climate and its appropriateness for studying climate compared to PIOMAS. Here are some relevant points:
1. As mentioned above, PIPS is an operational model. It is run to forecast ice conditions over 1-5 day intervals. The basic model physics is the same for any sea ice model – ice grows when it is cold, melts when temperatures are above freezing, and moves around due to winds and other factors. However, model details and how each type of model is implemented and run differ depending on the application. Similarly, climate and weather models include the same basic underlying physics, but you wouldn’t run a climate model to forecast weather or vice versa.
2. Validation of PIPS (see references above) has been done for sea ice extent, concentration, and motion near the ice edge (an important factor in the day-to-day changes in the ice edge). This is because the ice edge is the area of operational interest – i.e., the focus is on providing guidance for ships to avoid getting trapped in the ice. Very little validation was done for ice thickness estimates, particularly in the middle of the ice pack.
PIOMAS has been specifically validated for ice thickness using submarine and satellite data (http://psc.apl.washington.edu/zhang/IDAO/retro.html). Of course, the PIOMAS model estimates are not perfect, but they appear to capture the main features of the ice cover in response to forcings over seasonal and interannual scales.
3. PIPS 2.0 was first implemented in 1996 using model components developed in the 1970s and 1980s. These components capture the general physics of the ice and ocean well, but are basic by today’s standards. This provides suitable simulations of the ice cover, especially for short-term forecasts (which are most sensitive to the quality of the atmospheric forecast that drives the model). There has been a lot of sea ice model development since the 1980s, which according to a recent abstract for a conference presentation at the Joint Canadian Geophysical Union and Canadian Meteorological and Oceanographic Society 2010 Meeting, will be implemented in the next generation PIPS model, PIPS 3.0. However that is not yet being run operationally and thickness fields on the website are from PIPS 2.0. The primary references for PIPS 2.0 are Hibler (1979), Hibler (1980), Thorndike et al. (1980), and Cox (1984).
PIOMAS includes much more up-to-date model components (developed during the late 1990s early 2000s) with significant improvements in how well the model is able to simulate the growth, melt, and motion of the ice cover. In particular, the model do a much better job at realistically moving the ice around the basin and redistributing the thickness (i.e., rafting, ridging) in response to wind forcing. Thus, the thickness fields are likely to be more realistic than PIPS. The primary references for PIOMAS are: Zhang and Rothrock (2003), Zhang and Rothrock (2001), Winton (2000), Zhang and Hibler (1997), Dukowicz and Smith (1994).
4. The PIPS website has very limited information about the model or the model output products; it contains only image files; there are no raw data files, no documentation, no source code, no citation of peer-reviewed journal articles. A few articles can be found online elsewhere, and there are a few journal articles, but overall the information is quite sparse. This isn’t a big issue for PIPS, and I don’t fault those who run the PIPS model, because it has a small, focused user community who are familiar with the model, its characteristics, and its limitations.
The PIOMAS website contains detailed documentation including several peer-reviewed journal articles describing the model; it also contains model outputs, images, animations, and source code. Of course, the amount of documentation doesn’t say anything about the quality of the model outputs. But I think most people today agree that for climate data being widely-distributed and which is being used to make conclusions about climate change, it is a good idea to have data and code freely available.
So, which model results do I trust more? For operational forecasts, I might use PIPS. And PIPS probably does capture some aspects of the longer-term changes. But for the reasons stated above, I would trust the PIOMAS model results more for seasonal and interannual changes in the ice cover. I very much doubt that anyone familiar with the model details would unequivocally trust PIPS over PIOMAS.
But what about the PIOMAS volume anomaly estimates? How can they be showing a record low volume anomaly when there is less of the thinner first-year ice than in previous years as seen in ice age data? Doesn’t this mean that PIOMAS results are way off? Well, first, it is quite possible that the model may currently be underestimating ice thickness. No model is perfect. However, there is a possible explanation for the low volume and the PIOMAS model may largely be correct.
The areas that in recent years have been first-year ice that are now covered by 2nd and 3rd year ice will increase the volume – in those regions. However, compared to the last two years, there is even less of the oldest ice (see images below – I also included 1985 as an example of 1980s ice conditions for comparison). The loss of the oldest, thickest ice may more than offset the gain in volume from the 2nd and 3rd year ice. Also, it’s been a relatively warm winter in the Arctic, so first-year ice is likely a bit thinner than in recent years. Finally, the extent has been less than the last two years for the past couple of months. So the PIOMAS estimate that we are at record low volume anomaly is not implausible.
Early May ice age for: 1985 (top-left), 2008 (top-right), 2009 (bottom-left), and 2010 (bottom-right). OW = open water (no ice); 1 = ice that is 0-1 year old (first-year ice), 2 = ice that is 1-2 years old (2nd year ice), etc. Images courtesy of C. Fowler and J. Maslanik, University of Colorado, Boulder. Updated from Maslanik et al., 2007.
What does this all mean for this year’s minimum? Well, much will depend on the weather for the rest of the summer. As NSIDC states in its most recent post, we’ve expected we may see the rapid decline begin to slow because the melt will soon run into older, thicker ice, which will slow the loss of ice. Steve has said essentially the same thing and indeed we’ve the rate of loss slow over the past few days. Of course, there still a lot of time left in the melt season, and pace of melt continue to be relatively slow or it may speed up again, so we’ll see what happens. Regardless of what happens this summer though, the most important fact is that, despite some areas of the Arctic being a bit thicker this year, the long-term thinning and declining summer ice extent trend continues.
One final note about both PIPS and PIOMAS: Steve has claimed that “everyone agrees that PIPS 2.0 is the best data source of historical ice thickness”. Well, no scientist would even agree that what PIPS 2.0 produces are data! Being a data person myself, this is a bit of a pet peeve, but it’s important to make the distinction that model outputs are not data. Models are tremendously useful for obtaining information where data doesn’t exist (i.e., data sparse regions, historical periods without data), for projecting future changes, and for understanding how physical processes interact with each other (e.g., changes in climate due to changes in forcings).
However, model results are simulations, not observed data. And if there is good data available, I trust data over model estimates. And there is good historical data on ice thickness from submarine and satellite records (Kwok and Rothrock, 2009) and from proxy thickness estimates from ice age data (e.g., Maslanik et al., 2007). These data clearly show a long-term thinning trend. And while 2010 has relatively less of the thinner, first-year ice than the last couple of years, the ice cover is still much thinner than it was in earlier years. And it is clear that the models don’t entirely capture the spatial distribution of thickness correctly. As an example, compare the first-year ice in the ice age figure above with the PIPS 2.0 estimate from the same time period (below). In May, PIPS showed most of the central Arctic covered by ~3+ m ice, all the way to the Siberian coast. This is simply not realistic because the ice age data indicate first-year ice on much of the Siberian side of the Arctic (see images above), which would average at most 2 m. Thus Steve’s comparison of May 2010 and May 2008 with PIPS data is not valid because the model results are capturing observed spatial patterns of thickness.
Kwok , R. and D.A. Rothrock, 2009. Decline in Arctic sea ice thickness from submarine and ICESat records: 1958–2008, Geophys. Res. Lett., 36, L15501, doi:10.1029/2009GL039035.
Maslanik, J.A., C. Fowler, J. Stroeve, S. Drobot, J. Zwally, D. Yi, and W. Emery, 2007. A younger, thinner Arctic ice cover: Increased potential for extensive sea-ice loss, Geophys. Res. Lett., 34, L24501, doi:10.1029/2007GL032043.
Key PIPS 2.0 Model references:
Cox, M., 1984. A primitive equation, 3-dimensional model of the ocean, Geophysical Fluid Dynamics Laboratory Ocean Group Technical Report, Princeton, NJ, 1141 pp.
Hibler, W.D. III, 1979. A dynamic thermodynamic sea ice model, J. Phys. Oceanogr., 9(4), 815-846.
Hibler, W.D. III, 1980. Modeling a variable thickness sea ice cover, Mon. Weather Rev., 108(12), 1943-1973.
Thorndike, A.S., D.A. Rothrock, G.A. Maykut, and R. Colony, 1975. The thickness distribution of sea ice, J. Geophys. Res., 80(33), 4501-4513.
Key PIOMAS Model references:
Dukowicz, J.K., and R.D. Smith, 1994. Implicit free-surface method for the Bryan-Cox-Semtner ocean model, J. Geophys. Res., 99, 7791-8014.
Winton, M., 2000. A reformulated three-layer sea ice model, J. Atmos. Oceanic Technol., 17, 525-531.
Zhang J., and W.D. Hibler III, 1997. On an efficient numerical method for modeling sea ice dynamics, J. Geophys. Res., 102, 8691-8702.
Zhang, J., and D.A. Rothrock, 2001. A thickness and enthalpy distribution sea-ice model, J. Phys. Oceanogr., 31, 2986-3001.
Zhang, J., and D.A. Rothrock, 2003. Modeling global sea ice with a thickness and enthalpy distribution model in generalized curvilinear coordinates, Mon. Weather Rev., 131, 845-861.