Node detections over land
Opened this issue · 9 comments
I noticed that detecting a storm track over land does not work well. Compairing IBTrAC data and tracks generated with TempestExtremes during the same time periods looks good over oceans but it looks very poor over land. Is this something that you also have noticed? Or gotten a report from? I am only using surface pressure and prescipitable water. I wish I have the geopotential field like you have used in your paper, but that is not available for my situtation. Would you recommend some other ways to do it? Thanks.
TE should be able to track strong TCs immediately inland (e.g., if the storm moves into the southeastern US) -- if the tracks seem to be terminating right at the coastline (or just inland) it generally is because one is using true surface pressure instead of sea level pressure. If you are using surface pressure, TE will struggle over land because the elevation will strongly imprint on the 2D field.
Tracks over land are difficult algorithmically. Results become sensitive to the PSL reduction algorithm (not a problem over the ocean). One could use vorticity, although that becomes noisy as well in the presence of orography.
Once a TC moves far enough inland, it weakens. Storms decay over land due to friction and greatly reduced enthalpy fluxes. Since most models/reanalyses already have weak biases, this makes this even more difficult -- storms that are very borderline in an observational dataset will likely fall below the "weak tail" cutoff in TE.
IBTrACS is hand-curated, so it is easier for forecasters to subjectively place storms in the presence of local minima in station obs pressure fields and also track them on satellite images. Forecasters will also track storms in IBTrACS post-tropical (e.g., post-tropical depression, remnants, etc.) -- for a true apples-to-apples comparison with TE, one should remove those points.
Precipitable water decays greatly inland as there is no moisture source to the atmosphere. I would strongly encourage using an upper-level temperature or geopotential field to enforce a warm core. I would be surprised if the only data you have available are PS and TPW -- perhaps other data can be acquired?
Another potential solution is to weaken the criteria (e.g., reduce the magnitude of the closed pressure contour) but generally that rapidly increases the false alarm rate of the tracked storms.
Thank you so much for the detailed advice.
I am workinig with model data, and unfortunately it does not have sea-level pressure output. Another issue is that it does not have geopotential outout as it is an anelastic model. But it has temperature output in 3D. So, I guess I could impose warm core condition with upper-level temperature. Would you recommend a rough criteria for upper-level temperature constraint? Say, some temperature tolerance at 500 hPa or 300 hPa? It seems that there are many different implementations doing this.
Yes, this paper found that T400 is the best upper level temperature check to TCs.
However, if you do not have T400, you can use TE to do _AVG(T300,T500) which is a decent approximation.
For example:
If you have a line:
--closedcontourcmd PSL,200.0,5.5,0;_DIFF(Z300,Z500),-6.0,6.5,1.0
the simplest thing to do is try T400
--closedcontourcmd PSL,200.0,5.5,0;T400,-0.4,6.5,1.0
T400 is a non-standard variable and may not be available, so you could also average T200/T250/T300 and T500... Ex:
--closedcontourcmd PSL,200.0,5.5,0;_AVG(T250,T500),-0.4,6.5,1.0
Thank you for you advice. For the value of -0.4 K, is it just a value in between -2.0 K to -0.1 K of wcFOmag (T) from the paper? Thanks.
That was a value that scored highly when balancing hit rate and false alarm rate for reanalyses against IBTrACS. We just never did a formal optimization for it in the paper.
Thank you. I will try it on my dataset. Thanks.
Yeah.... I'v tried many times iwth surface pressure and temperature at 400 hPa level, I am still not getting any good tracks over land. I think it is mostly caused by using the surface pressure like you mentioned earlier.
Could you crudely convert surface pressure to sea level pressure? For instance using the approximation PSL = PS * exp(dz/H) with dz = surface height (in meters), and H = RdT/g with Rd=287, g=9.8 and T=surface temperature in K? ~ Paul
…
On Jan 28, 2024, at 7:14 AM, wreckdump @.**> wrote: Yeah.... I'v tried many times iwth surface pressure and temperature at 400 hPa level, I am still not getting any good tracks over land. I think it is mostly caused by using the surface pressure like you mentioned earlier. — Reply to this email directly, view it on GitHub <#67 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABIVPZC6OLWMZCMMOHXBQK3YQZTOVAVCNFSM6AAAAABCLSOTDGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMJTGYZDSNJUGQ. You are receiving this because you are subscribed to this thread.
Yeah. I'll try this.