This document summarizes an exersize to address some questions regarding the AMIP II model output. Sampling, interpolation and data compression tests have been made with the LMD, LLNL and NCAR models. Other groups that have performed relevant calculations are encouraged to contact the AMIP Project Office (amip@pcmdi.llnl.gov) to arrange having their findings posted at this site. Model output resulting from these experiments will be made available to the AMIP diagnostic community so that they may study the importance of the issues raised here to their own analysis. The problem: Unquantified or not fully understood implications of some recommendations associated with the AMIP II standard model output.
Questions addressed
Temporal sampling:
To what extent are monthly averages biased by using 6-hour snapshots instead of accumulating every time step?
Tests: Calculate the monthly average both ways for those fields in Table 1 with footnote 2 (i.e., based on 4/day instantaneous values), and ground temperature, surface pressure, u/v(10m) and q(2m) from Table 2. For consistency, all 3-d calculations should be accumulated to the 17 p-surfaces and then time averaged.
What is the effect of computing monthly average mean products (or covariances) based on 6-hour instantaneous values rather than on every model time step? Additionally, what is the effect of using 6-hr averages rather than 6-hour snapshots, and is the estimate much different from sampling daily?
Tests: Calculate the four mean products in Table 1b five different ways, based on i) 6-hr instantaneous, ii) 6-hr averages, iii) daily instantaneous iv) daily averages v) every time step. For consistency, fields should be interpolated to p surfaces and then time averaged.
Vertical interpolaton:
What are the differences between computing monthly averages in either i) model coordinates, ii) in model coordinates weighted by time varying mass (as described in Trenberth, NCAR/TN-396+STR), or iii) on the 17 WMO pressure surfaces?
Tests performed: Each of the three calculations are made for all terms in Table 1. The sampling is consistent with the AMIP II recommendations which are variable dependent.
Are vertical profiles of fields with sharp gradients such as cloud-related fields significantly effected by the interpolation process?
Tests performed: Monthly mean data was available on the model sigma/pressure hybrid vertical coordinate. Using the surface pressure at each gridpoint, the corresponding pressure level at each gridpoint was determined. These data were then linearly interpolated to the 17 AMIP pressure levels. Although more sophisticated interpolation techniques might be used, it was felt at this stage to keep this aspect of the study very simple. Additionally, the model pressure data (ie, the equivalent pressure of the monthly mean eta coordinate levels) and the AMIP interpolated data were linearly interpolated to a very fine vertical grid of 5 mb spacing. This was done to allow differences to be computed.
Compression issues
What are the potential implications of "aggressive" compression on 6-hourly model output. For example, what effect does this have on calculations that involve divergence?.
Tests: Save at 64bit the fields in Table 3, and as much of the Table 6 fields as possible. These data will be truncated to varying degrees with LATS (either at PCMDI or by each participating modeler). These data will be used in a variety of diagnostic calculations expected to be sensitive to the truncation.
UCRL-MI-135952