I want to compare a point observation (vertically point radar) at a fixed site location with a model output data (100 km spatial resolution) closest to the observational site. I need to bring both the data to a comparable scale so that my comparison would be meaningful. Since the point observations at the fixed site gives temporal output, I am converting the temporal scale to spatial equivalent by multiplying the horizontal winds until the temporal averaging considered matches the model resolution. In doing so, I can compare apple-apple observations and do quantitative comparison.
But, I am not sure if:
1. the assumption of Taylor hypothesis is reasonable?
2. Are probabilistic approaches better than quantitative comparisons (which needs averaging)?
3. What kind of approach for comparison would be the best way?
4. Are there any previous/ongoing work which address this issue?
Any suggestions/comments would be greatly appreciated.
Dear Chandra,
The model verification depends on which kind of variables do you use (continuous or discrete). As you mentioned you use radar observed elements which provides continuous variables. According to my best knowledge in continuous variables case the most recommended verification metrics are:
mean error (ME)
mean squared error (MSE)
root mean squared error (RMSE)
RMSE skill score
sample standard deviations of observations and forecasts
Detailed description about these methods you can find in Wilks, 2006: Statistical methods in the Atmospheric Sciences, Elsevier, 649 or Jollife and Stephenson, 2003: Forecast Verification A Practitioner`s Guide in Atmospheric Science. Wiley, 240
I hope I could help you!
Dear Arunchandra,
"Since the point observations at the fixed site gives temporal output, I am converting the temporal scale to spatial equivalent by multiplying the horizontal winds until the temporal averaging considered matches the model resolution. "
Be careful with this approach, because Taylor's hypothesis might come not true depending on, at least, the nature of the observable, the governing meteorological processes, the complexity of the sorrounding topography and wind speed fluctuations (since the proportionalities between 100km and your monitoring/sampling times might also strongly fluctuate).
Furthermore, if the model does not resolve meteorological processes under 100km horizontal resolution, regardless the methodology, you will have strong limitations when comparing model results with "artificially upscaled" observables (that are driven by lower,
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere