David E. Atkinson
Geological Survey of Canada (Atlantic)
Bedford Institute of Oceanography
Abstract The NCEP/NCAR Reanalysis project is providing an important source
of climate data for a wide variety of studies ranging over meteorological,
climatological, and climate-dependent themes. The project combines
observational data with an operational model to generate a variety of
climatic parameters at multiple levels over lat/long or standard gaussian
grids. In data sparse regions use of the model as an integrative basis
forms an effective means to interpolate between and extrapolate into
regions with little or no observational data. The contribution of
observational data to the final result is thus, by definition, highly
variable. Many applications for these data are at regional or local
spatial scales, often below the effective resolution of the Reanalysis
grid. The question of the accuracy with which the Reanalysis data
reproduce observational records is thus of interest. This question is
explored in this talk, focussing specifically on the wind field.
A direct comparison using vector and Pearson correlation statistics was
conducted between in situ observational data and data from the nearest
grid point in the 6-hly (10 m) wind field of the NCEP/NCAR Reanalysis set.
This work was performed at a circum-polar scale, using 183 weather
stations from throughout the circum-polar region: 50 stations from Russia,
Norway, Greenland and Alaska, and 74 Canadian coastal and 59 Canadian
interior sites from the Yukon, Nunavut and NWT. Other sources were also
employed, including hourly data from Russian and American ice islands from
the Arctic Ocean as well as southern Beaufort Sea oil platform data from
~1975 - ~1985. The comparisons were conducted over the 1950 - 2000 period,
for various seasonal breakdowns (calendar seasons and open water/freeze
up, as well as annual). Time series assessments, in which the correlation
results for the entire circum-Arctic are averaged by year, were also
performed, including comparison with indices of the Arctic Oscillation.
Results indicated good correlation for wind direction, improving at higher
speeds, but systematic underestimation of wind speed, especially during
high-magnitude events. Given that such events often engender some of the
more dramatic responses in the natural environment, it was of specific
interest to ensure they are adequately portrayed in the Reanalysis record.
For this, correction algorithms were developed, trained against
observational data, and applied. Results of the correction work were
reasonably good, in that many events were identified and corrected, but
some were missed because they have no discernible signature in the
Reanalysis data. Time series results suggested a sensitivity of the model
to decadal-scale variability associated with the Arctic Oscillation.