Abstract:
Past studies with a seasonal forecast system have confirmed that the
accurate initialization of land moisture reservoirs increases the
skill of subseasonal precipitation and air temperature forecasts. Soil
moisture initialization appears to affect forecasts mostly in the
transition zones between humid and dry regions. In most other areas of
the globe, the evaporation rates are either too small or too
unconnected with variations in soil moisture to have a coordinated
effect on rainfall and temperature variations in the presence of
background atmospheric variability. Such forecast systems, however,
have strong and identifiable biases in the spatial structures of
climate statistics. For example, rainfall structures in the
NASA/Global Modeling and Assimilation Office (GMAO) seasonal forecast
system are spatially too compact; the horizontal length scale of
spatial rainfall correlation is much smaller than that derived from
observational datasets. By accounting for this bias, we may be able to
extend the usefulness of the forecasts from skillful regions to
regions where the model appears unaffected by land initialization. In
this talk, we illustrate the differences between modeled and observed
spatial correlation structures of rainfall and air
temperature. Through both ad-hoc and optimized approaches applied to
an existing forecast experiment, we demonstrate how these biases can
be overcome, leading to a significant improvement in overall forecast
quality.