This subtask in the Intercomparison of 3D Radiation Codes (I3RC) project was defined in broad terms during a break-out session at the 1st International I3RC Workshop in Tucson (Az), Nov. 17-19, 1999. This document (created 12/22/99, updated 9/21/00, 11/30/00) reflects the assessment of 3D approximation methods in atmospheric radiation transport from that session, and some developments since then, including discussions at the 2nd I3RC Workshop (also in Tucson, Nov. 15-17, 2000).

Two distinct categories of approximate/fast 3D radiative transfer models can be defined:

"approximate-but-deterministic" models, directly comparable to the bulk of (exact) computational techniques represented in I3RC;

"approximate-since-probabilistic" models, which do not use all the standard I3RC input nor predict much beyond the domain-averages of a few I3RC output fields.

The first class of models has a natural place in I3RC because they produce radiation fields for given extinction fields and thus can be directly compared to "exact" methods based on the radiative transfer equation. The difference is that they numerically solve simpler sets of equations and therefore are expected to be orders of magnitude faster than Monte Carlo and even explicit methods of solving the full-blown radiative transfer equation (RTE) such as SHDOM. A well-known example is 3D diffusion theory which can be derived from the RTE in a variety of ways, and these derivations give us insight into where the approximation should and should not work. Now there are two levels of accuracy to ascertain:

Here is an evolving list of 15 possible candidates, in no particular order:
 

IPA: Independent Pixel Approximation
ED3D: Eddington-Delta in 3D
SHDOM: Spherical-Harmonic Discrete-Ordinates Method
DA: Discrete-Angle radiative transfer (typically using 6 beams, along orthogonal axes)

* means at least one person is involved in I3RC and was present at 1st and/or 2nd subtask meetings in Tucson (Phase-1,2 Workshops).

Other possible entries, both using a "DA, 1st-order, relaxation" approach: D. Schertzer (Paris VI) and C. Naud (Imperial College); H. Isaka and R. Borde (LaMP).

Possible applications:

The prime application for this type of model is a situation such as dynamical cloud modeling (LES- or CRM-style) where computer time is a concern for every proposed enhancement. Another, longer term, application would be computer-aided cloud optical tomography where cloud shape and structure would be varied to fit a number of observations, i.e., fully 3D cloud remote sensing.

Intercomparison protocol:

No fundamental difference here with the ongoing "exact" I3RC, part from the ability to follow the prescribed schedule. So, SERT modelers are encouraged to use the same cases, starting with the simpler ones in "Phase 1" (see I3RC home-page) if that is helpful. If possible, the same outputs should be produced again staring with the simpler ones (fluxes rather than radiances) if that helps. However, ramping up to cases from Phase 2 (at Priority 1, then 2) is strongly encouraged since they were designed purposefully to be very close to the level of detail required in the applications.

Contact-person for SERT codes: Anthony Davis <adavis@lanl.gov>
 
 

The second class of models makes no attempt whatsoever at predicting the specifics of radiation fields for some extinction field, only domain averages are estimated, possibly even only "ensemble" averages of the domain average. Here the extinction field is only used in some kind of statistical preprocessing to numerically determine the one or more variability parameters that these models all possess, since they all have a probabilistic flavor. Then the approximate radiative transfer calculation per se is performed, hopefully in hardly more time than a typical 2-stream computation. A good example is now classic Markovian-stochastic radiative transfer used extensively to model broken cloudiness effects and where the key structural parameters are two correlation lengths (mapped to the cloud fraction and characteristic cloud size). It is suggested to call these models radiative "mean-field theories" or MFTs, following a well-established tradition in statistical physics.

Here is an evolving list of 10 candidates, in no particular order:
 

ICA: Independent Column Approximation

* means at least one person is involved in I3RC and was present at 1st and/or 2nd subtask meetings in Tucson (Phase-1,2 Workshops).
? means has shown some interest in I3RC, but not present at either subtask meeting.

Other possible entries: S. Lovejoy (McGill) and B. Watson (Lawrence Col.) with a multiscaling approach; Vitaly Galinsky (Scripps Inst. Oc.) with a semi-analytical diffusion approach.

Possible applications:

Most of these models can be viewed as radiative parameterizations for GCMs in the early design phases: they target the large-scale properties of interest in climate studies but, for instance, they are not yet broadband capable. Since MFT models are generically based on some conceptual idea of how 3D cloud variability affects the overall patterns of radiation transport another potentially important application for them is simply as "diagnostic" tools to interpret data (appropriately averaged to some relevant scale). For instance, these simple models can help to sort real-world situations into more and less variable cases ?there is always that variability of the variability. From there, the models can help to make predictions on other observable quantities and/or to plan new measurements.

Intercomparison protocol:

Three basic problems arise here: (1) How much data is needed to define model parameters? (2) How representative is an ensemble-mean with respect to a single realization? (3) What radiative properties do we target?
 

1. This "sampling" problem means that this I3RC subgroup needs more raw data to experiment with, to define statistically comparable situations, to map natural variability to model parameters.
2. This "ergodicity" problem means that this I3RC subgroup could be putting a large computational request into the "exact" group to run enough cases to obtain at least ensemble-mean and ensemble-variance of the domain-average fields of interest.
3. This "comparison" problem will likely need some iteration, as was the case for the "exact" I3RC study; I would personally favor a minimal set of GCM-type requirements: albedo, transmission, heating-rate profile in cloud layer.

Clearly, this scenario is very different from I3RC and much closer to the on-going ICRCCM-III exercise coordinated by Howard Barker, Phil Partain, and Graeme Stephens, all involved in or closely following I3RC. ICRCCM-III (Intercomparison of Radiation Codes in Climate Models - Phase III) is designed for mature GCM short-wave parameterization schemes. This means, in particular, that they are broadband-enabled.  Even though ICRCCM-III is quite advanced in its agenda, it seems desirable that a dialog be initiated with this group (i) to see how their input information can be made available to MFT modelers in I3RC and (ii) to follow as closely as possible their intercomparison methodology for this sub-subtask.

Acting contact-person for MFTs: Anthony Davis <adavis@lanl.gov>