Skip to content

Reanalysis Ideas — A Path To A Solution?


Some Ideas For Future Reanalysis Efforts

John Lanzante


A Path To A Solution?

  • Based on the experiences of all data homogenization teams, the single
    most desirable tool for enhancing future efforts would be the availability
    of suitable independent reference time series (Tr). A reference series should
    capture the basic features of the real low-frequency climate variability,
    but not necessarily all of the details. It may be feasible, although not
    trivial, to produce an automated homogenization procedure, through the use
    of the difference series (To-Tr), with the techniques developed by LKS.
    The assumption is that both To and Tr share the same essential climate signal
    so that their difference is white noise plus the systematic, time-varying
    bias component of the raw observations (To).

  • Time series from the current NCEP/NCAR Reanalysis are not well suited
    to serve as reference series because of a number of uncontrolled factors:

    1. The raw input data contain artificial inhomogeneities, some of which
      are the the same as the data for which they would serve as a reference.

    2. Data inhomogeneities are introduced by an ever changing mix of data
      types, for example, the introduction of satellite data in the 1970s.

    3. The ever changing spatial distribution of data (station moves, closing
      and opening of new stations) introduces artificial inhomogeneities.

  • A unique problem for radiosonde data, as compared to surface data, is
    that instrument changes are often made on a country-wide basis. Thus,
    large contiguous areas may share the same malady. For example, the
    former Soviet Union, Australia, and the U.S./Canada/Caribbean (which
    have often shared instrumentation) dominate large regions of the world.
    In Africa, although instrumentation varies more between countries, LKS
    found that overall data quality is often very poor. In data sparse regions,
    such as for remote islands, there are no nearby stations. As a result of
    these conditions, any scheme designed to homogenize radiosonde
    temperatures can not rely in any large way on radiosonde temperatures
    from other nearby stations as a reference because too often nearby
    stations may:

    1. Share the same malady
    2. Have sub-standard data
    3. Not exist

Back to the main page