Reanalysis Ideas — A Strawman For Homogenization?
Some Ideas For Future Reanalysis Efforts
A Strawman For Homogenization?
- As part of the future Reanalysis effort, design a specialized system to produce reference series, the sole purpose of which would be to aid in homogenization of radiosonde temperature data. The homogenized temperature series would:
- Be of value in their own right for characterizing climate variability
- Would serve as input to the full Reanalysis system.
- Inputs to the system which produces the reference series would be severely limited, and if possible, no radiosonde temperatures would be input. It would be desirable to have radiosonde winds serve as a backbone for the system, supplemented by surface data (homogenized temperature data from the Global Historical Climatology Project? and Sea-Level Pressure data?). The appeal of winds data is:
- Wind measurement is simpler with less potential for inhomogeneities than for temperature.
- Inhomogeneities in wind and temperature would not necessarily occur together.
- Given thermal wind constraints, any inhomogeneities in wind may not translate strongly into resulting reference temperature.
- Only monthly mean outputs are needed, thus, considerable costs could be saved if the system were designed to ingest monthly means. [In most cases the time of an artificial jump can not be pinpointed more precisely than the nearest month or several months to a year].
- The reference series need not reproduce all of the details of the temporal variability.
- During the developmental phase, the already homogenized LKS data could provide benchmarks against which to judge the progress. Some validation could be provided by comparing both the reference series as well as the final homogenized temperatures with independent MSU temperature data. Although there are concerns regarding the homogeneity of the MSU data, two different versions are now available (Christy/Spencer and Wentz et al.).
- A flowchart can be used to conceptualize the homogenization scheme. The input data stream would be partitioned into “Bad” data, in this case radiosonde temperatures containing inhomogeneities, and “Good” data, which are presumed less susceptible to homogeneity problems and/or independent of the “Bad” data. The “Good” data are transformed by some sort of analysis system which produces reference series (Tr). Homogenization is greatly facilitated by operating on the difference series (To-Tr) which is presumably dominated by the time-varying artificial systematic error of the original “Bad” data (To). After homogenization, the temperature time series, which now are of considerable value in their own right, are fed back into the reanalysis system to produce the “Final” Reanalysis products.