Friday, July 26, 2013

New paper finds the same climate model produces different results when run on different computers

As if climate models didn't already have enough problems, a paper published today in the Monthly Weather Review finds that the same climate model run on different computer hardware and operating systems produces different results, "primarily due to the treatment of rounding errors by the different software systems" and that these errors propagate over time. According to the authors, "The [hardware & software] system dependency, which is the standard deviation of the 500-hPa geopotential height [areas of high & low pressure] averaged over the globe, increases with time." The authors find, "the ensemble spread due to the differences in software system is comparable to the ensemble spread due to the differences in initial conditions that is used for the traditional ensemble forecasting." The initial conditions of climate models have already been shown by many papers to produce significantly different projections of climate

Could climate catastrophe be due to a rounding error?


An Evaluation of the Software System Dependency of a Global Atmospheric Model

Song-You Hong,1 Myung-Seo Koo,1 Jihyeon Jang,1 Jung-Eun Esther Kim,2 Hoon Park,1,3 Min-Su Joh,4 Ji-Hoon Kang,4 and Tae-Jin Oh5
1 Department of Atmospheric Sciences, Yonsei University, Seoul, Korea
2 National Oceanic and Atmospheric Administration (NOAA)/Earth System Research Laboratory (ESRL), Boulder, Colorado, USA
3 Numerical Weather Prediction center, Korea Meteorological Administration, Seoul, Korea
4 Supercomputer Center, Korea Institute of Science and Technology Information, Daejeon, Korea
5 Korea Institute of Atmospheric Prediction Systems, Seoul, Korea


Abstract
This study presents the dependency of the simulation results from a global atmospheric numerical model on machines with different hardware and software systems. The global model program (GMP) of the Global/Regional Integrated Model system (GRIMs) is tested on 10 different computer systems having different central processing unit (CPU) architectures or compilers. There exist differences in the results for different compilers, parallel libraries, and optimization levels, primarily due to the treatment of rounding errors by the different software systems. The system dependency, which is the standard deviation of the 500-hPa geopotential height averaged over the globe, increases with time. However, its fractional tendency, which is the change of the standard deviation relative to the value itself, remains nearly zero with time. In a seasonal prediction framework, the ensemble spread due to the differences in software system is comparable to the ensemble spread due to the differences in initial conditions that is used for the traditional ensemble forecasting.

9 comments:

  1. In any numerical (finite math) simulation run long enough, the rounding errors become the result. This even if the equations translated into computer code are technically correct. This was in the texts on all the books on the subject over 50 years ago. It should have been obvious. Yet the kids who write such crappy code don't know enough of history or math to balance their checkbooks.

    To answer your question: Yes! The much publicized climate catastrophe IS due to a rounding error.

    Unfortunately, there still is the belief that if a computer says so, what it says is true. The combined wealth that the world could have generate in a decade has been squandered on the monumental fraud of climate catastrophe because of that belief.

    ReplyDelete
  2. The only reason why these kinds of problems haven't been seen more frequently is that very few climate scientists are willing to share their data or software. If anyone had ever tried to reproduce any of the published results of climate modeling, they would have immediately realized how sensitive these models' results are to numerical errors or to tiny changes in their inputs.

    ReplyDelete
  3. 1) There is an IEEE floating point standard designed to prevent this very problem. For critical numeric variables the maximum precision should have been defined.
    2) Compiler differences between machines can be maddening. Unfortunately there can also be differences on the same machine and with the same compiler which depended on the order of the compiler flags. Once I noticed this issue I met with the other programmers on the team and we agreed to a standard template for the compiler switches for the compiler, the library creation, and the executable image linker. Differences in compiled code due to flag differences.

    ReplyDelete
  4. 3) Clear different compilers on different platforms can cause differences between the behaviour even when compiled from the same code base. There is however a multi platform compiler, linker, and software image construction environment available from the free software foundation (FSF). The Gnu C Compiler is available on a huge variety of machines and in fact forms the foundation and basis for most Linux systems and builds. It has also been ported to the MS-Windows world.

    ReplyDelete
    Replies
    1. Clearly, the climate modelers are way out of their league on software design & testing, using ancient and very-poorly documented Fortran code that nobody really understands or verifies. Shameful the billions wasted on these computer games.

      Delete
  5. http://wattsupwiththat.com/2013/07/27/another-uncertainty-for-climate-models-different-results-on-different-computers-using-the-same-code/

    ReplyDelete
  6. http://joannenova.com.au/2013/07/oops-same-climate-models-produce-different-results-on-different-computers/

    ReplyDelete
  7. http://www.climatemonitor.it/?p=33827

    ReplyDelete
  8. http://www.eike-klima-energie.eu/climategate-anzeige/eine-weitere-unsicherheit-fuer-klimamodelle-gleiche-modelle-liefern-unterschiedliche-ergebnisse-wenn-auf-unterschiedlichen-computern-gerechnet/

    ReplyDelete