CIRC to fill critical need for continual evaluation of radiation codes

The computer codes that calculate the energy budget of solar and thermal radiation in Global Climate Models (GCMs), our most advanced tools for predicting climate change, have to be computationally efficient in order to not impose undue computational burden to climate simulations. By using approximations to gain execution speed, these codes sacrifice accuracy compared the more precise, but also much slower, alternatives. International efforts to evaluate the approximate schemes have taken place in the past, but they have suffered from the drawback that the accurate standards were not validated themselves for performance.

Figure 1:  Percentage errors commited by solar radiative transfer (RT) codes (identified only by their index no from 1 to 13) when calculating atmospheric absorption for CIRC Phase I cases. The “A” and “B” indicate simpler version of the main cases 1-7. Blue indicates underestimates.

Figure 2:  Percentage absolute error commited by the solar codes of Fig. 1 when considering collectively reflected, transmitted and absorbed radiation fluxes for all CIRC Phase I cases.

In a recent manuscript1 the main results of the first phase of an effort called “Continual Intercomparison of Radiation Codes” (CIRC) where the cases chosen to evaluate the approximate models are based on observations and where we have ensured that the accurate models perform well when compared to solar and thermal radiation measurements. The effort is endorsed by international organizations such as GEWEX and the International Radiation Commission and has a dedicated website where interested scientists can freely download data and obtain more information about the effort’s modus operandi and objectives.

The figures above show errors commited by approximate codes calculating solar radiative fluxes when compared to exact reference calcuations. These errors are significantly higher than counterpart errors for infrared radiative fluxes reflecting the greater challenges of shortwave (SW) radiative transfer where aerosol and surface albedo effects have to be carefully accounted for. Fig. 1 shows a map of percentage errors for each CIRC case of atmospheric absorption. Errors tend to be negative (underestimates by the participating codes) across the board, a result that is consistent with several previous studies. A summary of model performance is shown in Fig. 2 where three flux components are considered collectively for all 23 CIRC cases (main cases and subcases) to calculate an ensemble relative error using absolute deviations from the reference fluxes. Only 5 models out of 13 exhibit errors around 1%, while 3 record errors ~5% or above.


Data Sources:

To build the CIRC Phase I cases, a wide range of observations on the state of the atmosphere was compiled from the Atmospheric Radiation Measurement facilities of DOE’s Atmospheric System Research program. In addition, in-situ and satellite measurements of radiative fluxes were used. For the intercomparison itself reference radiative fluxes generated by AER’s line-by-line codes and submissions from participants operating approximate radiative transfer codes were analyzed.