ASSIMILATION OF SURFACE WEATHER OBSERVATIONS IN COMPLEX TERRAIN
Xingxiu Deng
Present computing power allows fine-resolution numerical weather
prediction models to resolve mesoscale flows within individual valleys.
Such resolution is critical for mountainous British Columbia, because
the valleys contain most of the population centers, industries, and
transportation routes. Accurate high-resolution forecasts depend on
accurate initial fields from which to start. To this end, dense local
surface weather observations should be utilized to supplement the
existing coarse-resolution analysis, while keeping computational costs
of data assimilation reasonable for local mesoscale modeling.
A technique has been developed to allow the creation of a new
anisotropic background-error correlation model for complex terrain,
which horizontally spreads surface weather observations along circuitous
valleys. The technique, called the mother-daughter approach, is based on
first-order boundary-layer characteristics in mountainous terrain. The
approach is further refined to account for land-sea anisotropy, and to
treat mountain-top observations differently from valley observations.
The mother-daughter approaches are tested and compared with two existing
methods, using virtual and real observations over different domains in
mountainous British Columbia. It is found that the mother-daughter
approaches outperform the other methods. The coastline refinement adds
value to the original mother-daughter approach in maintaining thermal
contrast across coastlines.
The resulting improved analysis from combining the detailed surface
analysis with pseudo upper-air data from the Eta model analysis is used
to initialize a high-resolution MC2 model. Numerical experiments are
performed to assess the impacts of assimilating surface observations in
complex terrain on subsequent forecasts of near-surface parameters.
Experimental results show improvement on subsequent near-surface
forecasts of the variables (e.g., temperature and humidity) that are
directly assimilated into the model. These findings are confirmed by
operational runs, and only minor differences are found.