A quick skim of the regional chapters in AR5 shows predicted changes to average temperature and precipitation. If I understand (in rough terms) the modelling methodologies used, a variety of models (an ensemble) are run many times each, each with different initial conditions. The output is $\mathbf{Y}_{xyztmi}$ where $xyzt$ are coordinates and time, $m$ is a model, $i$ is one run of a specific model, and $Y$ is some modeled climate variable. The ensemble mean prediction at time $t$ for gridcell $xyz$ is just the average over $m$ and $i$.

The variance of $Y_{mi}$ however is comprised both of real variability in the climate system, and model error. This paper (9 years old now) suggests that drivers of intra-annual precipitation are really hard to model with GCMs. Can it be done effectively with RCMs? Are there dynamical arguments suggesting changes to higher moments of climate parameters (particularly in the tropics)?

Changes to distributions are commonly talked-about in the impacts literature, but I haven't come across concrete projections of changes to distributions from the climate science community. Citations welcome.

  • $\begingroup$ $m$ and $i$ could practically be rolled in to one. Often an institution will submit simulations from one model, but with slightly different structure (e.g. components switched out), or with different parameters. Also, some intitution's models are more or less copied from another institution, and then altered slightly. And initial conditions are more or less irrelevant after a few decades of simulation (spin-up) anyway (that is, little differences can cause big differences in results, swamping the original differences). $\endgroup$
    – naught101
    Jan 30, 2015 at 2:45
  • $\begingroup$ And variability in precipitation in climate models is a very well known problem. Many models predict way too many "drizzle days" - days with light rain, and too few extremes. $\endgroup$
    – naught101
    Jan 30, 2015 at 2:46
  • $\begingroup$ ^right -- I'm aware that GCMs can't do convective rainfall so well, because the relevant processes happen at super-fine spatial scale. I'm also aware that most RCMs are finer-scale versions of GCMs, driven by GCM boundary conditions. Here is an idea that I bet climate scientists have already had: embed a RCM in a GCM, and then embed a synoptic weather forecasting model inside the RCM. Would that work? If it can predict current weather, it might be what's needed to get at changes to the variability of future climates. $\endgroup$ Jan 30, 2015 at 4:51
  • $\begingroup$ or perhaps a more straightforward approach would be to simply model (statistically) the higher moments of rainfall and temperature distributions as functions of their means, and then extrapolate those results to projected future means. that might be really noisy however given uncertainty in most GCM projections, and it might be tough to quantify extrapolation error. $\endgroup$ Jan 30, 2015 at 4:53
  • 1
    $\begingroup$ Yeah, that happens quite regularly. There's a section on nested regional climate modeling in the AR4. (this was in response to the comment three previous to this :) $\endgroup$
    – naught101
    Jan 30, 2015 at 4:56


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.