|
|
|
|
|
I3RC Challenges
I3RC has been faced with a
number of challenges. It is well underway in meeting these, but the
process is not complete. The challenges are:
- Diversity of methods.
Participating methods that solve the exact 3D transfer equation include
3D discrete ordinates and spherical harmonic discrete ordinates (SHDOM).
Many of the participating methods are based on Monte Carlo.
These include several versions of Monte Carlo - forward andbackward.
Most Monte Carlo codes share similar techniques, such as
maximal cross section, forefficiency. Monte Carlo approaches
solve the exact transfer equation, and have relativelywell-understood
errors, with each photon constituting an independent Bernoulli trial.
Thus Monte Carlo is useful in evaluating errors in other methods.
3D methods that begin by approximating thetransfer equation,
such as diffusion and discrete angle methods, also participate in I3RC,
and canoften gain speed advantages over the exact methods, sometimes at the
expense of significantly largererrors. So I3RC is well along in meeting
the challenge of diversity.
- Applicability. For
I3RC to benefit both remote sensing and climate modeling, it is necessary
forI3RC computations to include a wide variety of radiances, fluxes, and
heating rates. Outputs quickly multiply. Even the restricted
set of fields and outputs of Phase 1 led to a very large number ofcomparisons
to examine, reflected in the multiplicity of plots posted to the I3RC
web. For these tobe useful and accessible required a simple
and flexible Web interface, which will continue to be improved,
and will be important in fulfilling I3RC's educational objective.
- Portability. Many
of the codes have now been ported to a common Linux computer.
The open source initiative within I3RC will require that all submitted
code be tested with the Portland Fortrancompiler available there.
Some routines will be selected for porting to a multiprocessing environment.
Maintaining portability will be an important ongoing effort,
without which shared code would have limited usefulness.
- Scalability.
Input cloud fields for I3RC must have a spatial resolution capable
of re-solvingtypical photon mean-free-path on the order of 100 meters, in
order to represent 3D radiation effects, yet cover a sufficiently
large domain to fairly represent cloud variability unresolved by global
climate models, currently exceeding 25 kilometers horizontally.
These two goals are notsimultaneously achievable with current commonly available
computing resources. The I3RC baselinecases handle this
problem by choosing relatively small domains within which 3D effects
arewell-resolved, and assuming that plane-parallel biases in domain-averaged
quantities can bescaled up to the larger scales needed by models.
This relies on empirical and cloud-resolvingmodeling studies of the
scaling properties of clouds, that are still ongoing.
Empirical cloudstudies themselves rely on 3D radiation codes.
- Speed.
Cloud resolving models currently use plane-parallel theory to compute
radiative heatingrates within atmospheric columns that may be only tens of
meters in diameter, thus neglecting nethorizontal fluxes that are often
a large as the net vertical fluxes used to drive the models.
Domain-average horizontal fluxes are small, but they may still have
an effect on local convection,and may even produce large-scale drift.
To accurately simulate these effects requires a 3D radiativecode capable
of running fast enough to keep up with the rapid dynamical development
of convection. That is indeed a challenging problem for 3D radiation,
and emphasizes the need for any 3Dapproximation that can gain a significant
advantage in speed.
The primary result of I3RC's
first phase has been a demonstration of the high degree of accuracy that
exact 3D radiation codes are able to achieve, as well as the excellent accuracy
that even approximate methods can achieve for certain radiative quantities.
The many comparison plots available on this site document this accuracy,
and the individual efforts that went into achieving it.
| |
|
