How we know what the weather was like
What is this Site?
This is a
website for people new to the concept of reanalysis. The purpose
of this website is to introduce reanalysis, outline the steps, and lay
out some details from one of those steps: data assimilation.
There
are
many
methods
of
data
assimilation,
and
they
keep
changing as
computer algorithms evlolve, but only one method will be explained on
this site, the 3D-var method called Spectral Statistical Interpolation
(SSI) implemented by NCEP in the early 90's. This is the
algorithm used for the NCEP/NCAR 50-year reanalysis.
What is reanalysis?
Reanalysis is
climate data from the past; it is a
series of data from all over the world based on observations and
weather forecasts. In order to analyze
what the global climate is doing over time, data from all
over the world is needed, for as much time as possible. However,
there
aren't enough observations at any given time
to
know everything about the weather all over the world. So a
weather forecast is run, but instead of predicting the future, this
forecast predicts the past. Then both the forecast and the
observations are carefully combined to create a new set of data:
reanalysis.
Why is reanalysis used?
There are several reasons that reanalysis data is
more useful or convenient than observational data. These include:
http://www.cpc.ncep.noaa.gov/products/wesley/cdrom/bams96/FIGURE1.GIF
- Changes in technology: When a new observation system is brought online (e.g. a new sattelite), or an old one is taken offline (e.g. a site stops sending up daily weather baloons) the data suddenly jumps. The graph to the left is an example of an average temperature over the tropics before and after the use of sattellite data: before sattellites, the average temperature reflected a few observations rather than the average of the entire area, whereas with sattellites, the average temperature reflected the entire area.
- Inconsistent data: Many observational data sets have missing data, whether from equipment malfunction, funding issues or historical events. These gaps in the data prove to be major obstacles to the analysis of the data.
- Unevenly spaced observations: Data is not collected evenly all over the world. For example, often there are many stations collecting data in the vicinity of a large city, but relatively few in rural areas. Furthermore, some countries have many observation stations whereas some have none. This makes analyisis of the data over time less accurate for many places.
How does reanalysis work?
There
are several steps not mensioned, but this is more or less what happens:
- Forecast data and Observations are acquired
- The data assimilation algorithm is run: this changes the forecast data to match the observations as closely as possible.
- This new data is recorded, then entered into the forecast model
- A new forecast is created for the next time, and corresponding observations are acquired.
This process is repeated until the entire time
period is reanalyzed, and a new data set is created. This data
set is much more complete than the observations, but is not quite the
same as what was "really" going on.
Design by JeremyD