I have tried to retrieve wind using Pydda and gridded the data at 1km along x, y and 0.5km along z axis… I have averaged the at each height level and tried to compared with GPS-sonde observation and VAD method… I have done intisalization with constant filed, Radiosonde and ERA methods.
But The magnitude of observed LLJ (U values) is underestimated? What could be the reason?
The DDA wind closest is to VAD-ART at lower levels but again it overestimates the radiosonde at ART for higher levels…
Is it correct to compare like this?
What should be value of dBZ is acceptable to have a reliable pydda wind estimation?
I would first like to see your 2D wind field at 2 km to see what could be happening. Given that you have convection around, it could be that the winds sampled by the radar are being modified by the convection and that is bearing out in the average wind field. How close in time is your sounding to the storm?
Also, at higher altitudes (> 8 km), spacing between the radar scans will be higher causing more data gaps in your PyDDA retrieval which can in turn drastically increase the uncertainty in your retrieved wind fields.
Thank you very much for your reply. Here I am addressing your queries…
Radiosonde flight time is 05:30 to 06:55 UTC, whereas radar observation is 10:22 UTC on 2nd August 2023. But LLJ is a monsoon characteristic. Both radars were operated with a similar scan strategy with elevation angles of 0.1,0.7,1.3,2.3,4.0,8.5,11,15 and 20 in degree. I am attaching 2D winds at lower altitudes for your kind perusal.
I am also attaching the 2D winds at higher altitudes so you can verify why uncertainty in winds is higher wrt to VAD.
I have the following parameters for wind estimation
Co=1, Cm=256.0,Cx=0, Cy=0, Cz=0, Cb=0, frz=5000.0, filter_window=5,
mask_outside_opt=True, upper_bc=1,wind_tol=0.2, engine=“tensorflow”
I would consider setting Cx and Cy = 1 to test out if smoothing can help with your issues in the upper level winds.
I also see the winds turning more southerly near the storm edges as you go to higher altitudes. I wonder if this is due to the loss of wind data outside of the storm causing the mass continuity constraint to correct for this by turning the winds with height. What happens when you only use winds in regions > 5, 10, 20 dBZ? There we would expect a fuller profile of winds for the variational retrieval to work with. It looks like the winds in > 5 DBZ regions look to be more continuous. Smoothing may also hash out some of these issues as well.
Thank you very much for your reply. I have set Cx=1, and Cy = 1; but not a major improvement has been observed… the figure is a
smoothed profile with no dBZ condition.
The profile above dBZ conditions…
with dBZ conditions it looks improved but U around 4.5 km is near zero…
Can mass continuity constraints be adjusted?
I would try adjusting mass continuity constraints. Since you don’t have widespread coverage in your dual doppler lobes the variational technique may not have enough data to fully resolve the dynamics of the storm. I would also try this for a case where you have more precipitation coverage in the lobes and see how your winds compare.