Data AssimilationAdded : Tuesday at 13:05 Yesterday, we talked about how satellite data and the increased use of it is vital to numerical weather models. In fact, it's usually when modelers identify a weakness in a weather model that a new sensor is added to a satellite and sent to orbit to help out. This has happened for things like soil moisture, ice coverage and sea temperature to name but a few.
Data Assimilation though is the most important part of any numerical weather model and modelers know this too. In the USA, the main GFS ► model uses what's called GDAS (Global Data Assimilation System) as starting conditions for its model.
Now, you might think, let's just get all the observations for 12:00UTC then put them in the model and run it and see what we get? But, if you did that, then the models would go haywire after a couple of days due to things like data errors, missing data, data sent in at the wrong or different times.
Data Assimilation is all about trying to make the starting conditions as near to perfect as possible, after all, the better the initial conditions and closest to the current atmosphere, the better the model will handle the changes in the future.
But this isn't as straight forward as it might seem. Observations are taken from a plethora of sources including land observations, satellites, radar, ships and also aircraft which take 3d observations as they take off and land, but not all flights take off at 12:00UTC!
What happens, is that the observations are all sent in with a timestamp for every observation and the exact latitude, longitude and elevation. Data Assimilation will then compare these to the forecast which the model suggested previously to see not only if there are any glaring errors, but also to see how accurate its own forecasting model is for that exact timestamp.
This timestamp is important. Most models these days contain 4d-VAR which is the ability to add time to the observation too. So, if an observation comes in at 12:15UTC then it is added in at the 15 minute timestep of the model. 4d-VAR was a major milestone in numerical weather forecasting and the ECMWF ► introduced it in 1997. It allows polar orbiting satellites to send in data observations at any time which can then be slotted into the Data Assimilation for the next model run regardless of when the observation was taken.
But, the model wants starting conditions for every grid point in the world. It does this by applying a Gaussian Filter to smooth out current observations and then applying a Kalman Filter in order to mask out "noise" from data errors, missing observations, equipment problems, cats weeing in rain gauges etc.
So, then we have the starting conditions right? Wrong... What happens if the model forecast a band of rain across say Southern England in 6 hours, but the frontal system actually fizzled out? Data Assimilation will then compare the observations to the previous forecast to see any differences.
It's this area which is where Data Assimilation comes into its own. Some agencies use humans to correct the starting data, others use Machine Learning or Artificial Intelligence in order to get the balance of correct observations and correct forecast matching.
This process can take a while for human input, but it's the Data Assimilation which makes or breaks how accurate a model is. The ECMWF ► takes this stage incredibly seriously and their Data Assimilation is world class which results in a more accurate model, but the downside is that the model is run twice a day and takes a little longer to come through.
So, a bit of background information for you and why taking numerical output at face value is never a clever thing to do.
At Metcheck, we have our own Data Assimilation which compares observations around the world every 30 minutes with various weather models to decide the most accurate one and then weights this accordingly in the forecasts.
The end result is pretty pictures and colour coded data, but the work which goes underneath the bonnet is truly colossal and we hope this discussion has given you a small taste of what happens behind the scenes.
METEOROLOGIST : MARSH |