JRA download isn't getting the best resolution data

[nicholas512]

Grid

JRA currently downloads a 1.25 degree grid

288:145:90N:0E:90S:1.25W:1.25:1.25

However, ~0.5 degree grid is available (model resolution)

e.g. from interactive subset online

- Output format: netCDF
- Start date: 2020-01-01 00:00
- End date: 2022-06-01 00:00
- Parameter(s):
    Total precipitation
    Downward longwave radiation flux
    Downward solar radiation flux
    Pressure
- Vertical level(s):
    Ground or water surface
- Grid(s):
    0.563-deg x ~0.562-deg from 0E to 359.438E and 89.57N to 89.57S (640 x 320 Longitude/Gaussian 
Latitude)
- Spatial subsetting (bounding box):
    Latitudes (top/bottom): 70 / 60
    Longitudes (left/right): -140 / -90

1.25 degree grid (what we get) 0.56 degree grid (what we could get)

Implementation

it shouldn't be much trouble if we can just request the finer grid

3- and 6-hourly forecast

Currently, only 0-3 hourly forecast is downloaded 'product': '3-hour Average (initial+0 to initial+3)', but its also possible to get initial+3 to initial+6, which would be a finer temporal resolution for certain variables. Comparing what we get with what we could get:

Implementation

we might be able to get both products (3 and 6 hr) by omitting the 'product' specification in the requst (online interface defaults to "all available data')

'dataset': 'ds628.0',
                'date': self.makeDate(),
                'param': '/'.join(self.param),
                'level': 'Ground or water surface:0',
                'oformat': 'netCDF',
                'nlat': str(self.area['north']),
                'slat': str(self.area['south']),
                'wlon': str(self.area['west']),
                'elon': str(self.area['east']),
                'product': '3-hour Average (initial+0 to initial+3)',
                'compression': 'NN',
                'gridproj': 'latLon',
                'griddef': '288:145:90N:0E:90S:1.25W:1.25:1.25'}

Then its a question of:

1. weaving the T+3 and T+6 arrays together (they come as separate arrays) - do we do this during the download step or interpolate step? We would need a new time array as well (it comes as the forecast time)
2. handling time-step mismatches in the scaling step (i.e. repeating 6-hourly values so that we can do operations on the arrays)

[BinCAO-PF]

Yes, JRA-55 has a higher resolution for model level comparing to the reanalysis we used. I discussed with Stephan when I was working on GlobSim. A significant drawback of the model level is it does not (or very little) assimilate observations, and hence has worse performance than the reanalysis although has a higher resolution. I did not realize the "3- and 6-hourly forecast” issue. In fact, I simply followed the original globsim document, and mostly focusing on data downloads.