Benchmark Calculations

Each of the six CRM fields will have a series of benchmark calculations. Benchmarks will be performed assuming different configurations of unresolved clouds, but in each case profiles of condensed water mass and cloud fraction will be conserved. Thus, the benchmark experiments are:
  • (A) Full 3D Monte Carlo with CRM horizontal grid-spacings -> these cases will constitute the "TRUTH".
  • (B) Independent Column Approximation (ICA): make horizontal grid-spacing huge thus suppressing net horizontal transport yet preserving all the horizontal cloud fluctuations and overlapping patterns.
  • (C) Perfect overlap: same as ICA but clouds are homogenized on a per layer basis thus preserving the exact overlap of cloud (this is a direct test for 1D plane-parallel codes that attempt to address overlap)
  • (D) Maximum/random overlap: like 'perfect overlap' but clouds in contiguous layers are maximally overlapped (excess portions are positioned at random) but if and when a cloudless layer exists, one 'block' of clouds is randomly overlapped with the other 'block(s)' of clouds (direct test for codes that attempt to address this standard overlap assumption)
  • (E) Random overlap: like 'perfect overlap' but clouds are positioned at random across each layer.

Thus for example, if your code treats clouds as though they were horizontally homogeneous but attempts to represent their vertical overlapping structure as closely as possible, you would compare your results directly to benchmark (C). While your model might perform well (i.e., it does what it is intended to do), it would still lack the effects of horizontal variability which are represented by the difference between (B) and (C) (differences between (A) and (D) would show your ultimate error).

Benchmarks (D) and (E) involve random number generation to position cloudy cells. For solar fluxes, however, experience suggests that a single realization represents the population very well.