**  Mon Feb  2 12:47:14 2015   Andy Jacobson <andy.jacobson@noaa.gov> **

  WARNING

  In its optimization process, CarbonTracker estimates a scaling
  factor for each ecoregion and week.  This scaling factor multiplies
  an "initial guess" (or prior) flux from each grid box in the
  ecoregion/week domain.  The prior values for these scaling factors
  are close to 1.0, with propagation of information from previous
  weeks in the estimation.  In under-constrained estimation problems,
  where there are not enough data to uniquely identify a solution, it
  is possible to develop "dipoles" where positive flux corrections in
  one time and place are offset by negative flux corrections in
  another.

  Statistically, some small number of the scaling factors can become
  negative, because they are modeled as Gaussian random variables and
  no part of the real number line is excluded for such a distribution.
  This is almost always indicative of a flux dipole, which suggests
  that the system is attempting to resolve scales more finely than the
  data can support.  Solution to this problem include postaggregating
  fluxes to average out these dipoles, or coarsening the scale at
  which the inversion is performed. Negative scaling factors result in
  a reversed diurnal cycle, with unrealistic features like nighttime
  photosynthetic uptake of CO2.

  We present high-resolution (1x1 degree, 3-hourly) fluxes here.  The
  effects of dipoles are evident in some of these fluxes, and
  unrealistic fluxes are present in a small number of cases.  Please
  contact andy.jacobson@noaa.gov with questions about this issue.