File Structure:
We have designed our data product in a way that each model output has its own ncdf file with the following structure:
Dimensions:
- site = 493;
- Year = 33;
- ensemble = 20;
- time = 119726;
Variables:
double site(site=493);
float VARIABLE (ensemble=20, site=493, time=119726);
float weights(Year=33, ensemble=20, site=493);
float weights_rel(Year=33, ensemble=20, site=493);
int Year(Year=33);
double ensemble(ensemble=20);
double time(time=119726);
`weights_rel` is the variable that carries relative weights of each ensemble member over time and space/site. R scripts are developed (can be found in the file manager and under R_scripts) that helps to apply the weights to our raw model outputs for fluxes.
We have two different time dependant variables (time/Year) mainly because weights and relative weights vary on the annual time scale, whereas raw model outputs vary on 3 hourly time step.
The difference between weights and relative weights also is just simply the fact that the weights are the raw log-likelihood whereas relative weights will add up to 1 for each site*year.