Floats going to zero

[scrallen]

Hi

I am running a simulation with Lagrangian floats. I have a single grid with 16x16 MPI decomposition.

The model is compiled with:
#define FLOATS
#define FLOAT_VWALK

The grid has radnug BCs on all sides (except for one, which is entirely land and is closed).

In my initial testing, I'm putting a single float at every (water) grid point, releasing them at the surface (Fz0 = -0.01), just after initialisation (Ft0 = -0.1). I'm only running it for a short time.

I notice that for some of the floats, their values turn to zero (See below). Can anyone explain why this happens? Is there some mistake I am making with the configuration of the floats? I suspect this is happening when floats leave the model grid... but is this expected behaviour?

Thanks and kind regards,
Stewart

Code: Select all

      Lon       Lat     Depth      Temp      Salt
-------------------------------------------------
      nan       nan       nan       nan       nan
      nan       nan       nan       nan       nan
 151.2060  -11.0443  -62.4504   26.4907   35.1637
 151.2030  -11.0361  -54.6605   26.6351   35.1716
 151.2000  -11.0261  -51.4100   26.6143   35.1732
 151.1975  -11.0149  -49.2034   26.7073   35.1534
 151.1965  -11.0042  -49.2259   26.7441   35.1354
 151.1969  -10.9959  -50.3247   26.7245   35.1272
 151.1978  -10.9908  -52.5509   26.6039   35.1543
 151.1980  -10.9890  -54.4823   26.5175   35.1638
 151.1967  -10.9899  -56.6882   26.3948   35.1719
 151.1943  -10.9915  -59.3447   26.2740   35.1870
 151.1917  -10.9912  -61.3027   26.2406   35.1940
 151.1890  -10.9873  -63.2200   26.1723   35.2010
 151.1871  -10.9792  -63.7728   26.1317   35.2003
   0.0000    0.0000    0.0000    0.0000    0.0000
   0.0000    0.0000    0.0000    0.0000    0.0000
   0.0000    0.0000    0.0000    0.0000    0.0000
   0.0000    0.0000    0.0000    0.0000    0.0000
   0.0000    0.0000    0.0000    0.0000    0.0000

[kate]

Yes, I believe that float has left the domain. Any MPI process that "owns" a float will update it while the rest will set it to zero. A sum-reduction is done so that only the owner process will be contributing to updating the float. If the float isn't on any tile, all processes will set it to zero.

[scrallen]

Ah, It sounds like this behaviour is expected.

However, I would add that leaving the model domain through a boundary is a bit different from moving across MPI tiles. I was expecting that any float that leaves the domain would be deemed lost and set to NaN or similar.

Also, luckily my domain does not include the point on the equator, south of Greenwich... or any freezing waters!

Thanks,
Stewart

[kate]

Feel free to change it! I only know about the sum-reduction because strange things can happen in the Arctic when two locations match as being at a certain lat,lon location: where you think it should be and over north of Greenwich where longitude goes from 360 back to 0.