 |
| Fig. 1 (click to enlarge) |
In this series I have been describing some of the processes involved in the development of software which attempts to project future sea level rise (SLR).
So far a 3yr, 5yr, 7yr, and 10yr doubling dataset has been discussed (The Evolution of Models - 10).
The 2yr doubling dataset was discussed in another series (The Question Is: How Much Acceleration Is Involved In SLR - 5?).
The resulting 5yr, 7yr, and 10yr doubling datasets match Hansen's calculations for a 1m SLR exactly to the year:
The increasing Greenland mass loss ... can be fit just as well by exponentially increasing annual mass loss, a behavior that Hansen (2005, 2007) argues could occur because of multiple amplifying feedbacks as an ice sheet begins to disintegrate. A 10-year doubling time would lead to 1 meter sea level rise by 2067 ... 2045 ... for 5-year doubling time and 2055 ... for a 7-year doubling time.(Update of Greenland Ice Sheet Mass Loss: Exponential?). Since Hansen did not offer either a 3yr or 2yr doubling concept, I added those two.
Then, I generated graphs (all five graphs are shown in the following two posts: The Evolution of Models - 10 and The Question Is: How Much Acceleration Is Involved In SLR - 5?.
I did those individual graphs using the same doubling-ratio logic that Jim Hansen and Makiko Sato had used for their 5,7, and 10yr doubling (a.k.a. acceleration) projections.
Note that Hansen's use of the descriptive words "doubling" and "exponential" is just another way of describing "acceleration."
II. Why I Took A Different Design Approach
A recent paper in the journal Cryosphere pointed out the complex system problems that
 |
| Fig. 2 (@ page 181) |
In spite of significant improvements in the simulated GrIS topography with our discharge parameterization, for all of our simulations it was impossible to yield an error in ice thickness smaller than about 18 %.(The Cryosphere, 9, 179–196, 2015, at 191, PDF). That is talking about matching the historical records concerning the ice thickness (not the future thickness).
That approach (illustrated to a degree in Fig. 2) could be what plagued them with the 18% inaccuracy.
It tells me that they took a minutiae design approach instead of a higher level "zone approach."
Using that degree of minutia on a vast ice sheet would require that a software development team calculate many dynamics for each square, each unit.
How much water will be produced by melt in each square, how much ice remains in each square, what happens if some of the square calves, what if the ice in square 9,834,492 moves to the another square, etc. etc.
Those complexities generally tend to cause a wrestling match between the solution domain and the problem domain.
That is most often due to the software complexities growing enough to compound the real-world problem complexities.
Such a wrestling match often increases periodically until it reaches the point of taking the main focus away from the physical reality, the problem domain (future SLR), which the development team is trying to focus on.
As they encounter technical software development issues, unless they are careful, the solution domain may morph to join the problem domain.
III. Basic Design Approach
A. Core Data
Knowing that about software projects, from years of experience in software architecture, I took a very basic approach, so as to avoid the runaway increase of complexity.
I did so in order to maximize the development of the core dynamics of a solution, as well as to minimize technical software complexities.
The graphic at the top of this post (Fig. 1) has the fundamental data needed in order to build core logic for an SLR projection program.
I concluded that a clean design required focusing first on the potential amount of SLR at the two key locations, Antarctica (212.58' + 26.44' + 1.51' = 240.53 ft. of SLR potential) and Greenland (21.49 ft. of SLR potential).
Avoiding "majoring in the minors" also meant spending less focus on the minor non-Polar ice which has only 1.48 ft. of SLR potential anyway.
B. Zones
Having done that, I then sectioned off the three locations into 4 "melt zones" (coastal, inland 1, inland 2, and no-melt).
Those zones are based on degrees of being most likely to melt/calve or less likely to melt/calve, in a given year (e.g. what percent of the zone will make its way to the sea in a given year).
The coastal is most likely to make it to the sea first, proceeding through less likely inland zones, then finally to the deep interior (which is not likely to melt much for perhaps a century or more).
This technique skips all ice-to-water calculations, how much ice will melt and how much will calve instead, etc. etc. (to instead deal only with SLR directly).
The simple meaning and exercise then becomes what percentage of a zone becomes SLR rather than how it becomes SLR (e.g. melt, calving, evaporation).
It allows a focus on temperature acceleration (atmosphere & ocean temperature) as the means for developing a premise for SLR acceleration (NOAA" No Warming Slowdown, cf. NOAA Study Confirms Global Warming Speed-Up Is Imminent).
C. Percentage of SLR Per Annum
The logic that was then coded became all about what percentage of each zone's potential SLR is going to materialize into actual SLR in a given year.
The final computation is simple arithmetic: all the zone data in each geographical location is added together to derive a total SLR quantity for that year.
Then a graph is generated from the numbers derived (e.g. The Evolution of Models - 10).
IV. A Grasp of Acceleration Is The Key For Realistic Projection
No matter which design technique is used, minutiae or direct SLR, one has to develop a useful and realistic acceleration rate (or several as Hansen did).
Heretofore, models have tended to use linear (same rate of SLR forever) calculations when projecting SLR out into the future.
IOW, BAU is used in SLR models even though that is not the case for the modelling of CO2, temperature, and other climate change impacts (a concept of acceleration of the rate of increase is customary for those models).
V. Conclusion
Unfortunately, until the models take a clean approach to SLR projection, using acceleration as the core driving logic, we can expect underestimates of SLR.
That is because real-world SLR is not based on linear dynamics, it is based on acceleration dynamics.
The next post in this series is here, the previous post in this series is here.
The Yale Forum ...
Asimov on computer models: The Last Question
ReplyDelete