Segmentation Faults Running WRF at High Latitudes

[dstabile]

Hi all,

I'm using WRF to downscale global climate data. I've been running into the following error (from rsl.error.0000) when I run WRF at high latitudes with the polar projection:

Caught signal 11 (Segmentation fault: address not mapped to object at address 0xffffffe0077b2de0)
==== backtrace (tid: 891244) ====
0 /lib/x86_64-linux-gnu/libucs.so.0(ucs_handle_error+0x2e4) [0x14d73b074fc4]
1 /lib/x86_64-linux-gnu/libucs.so.0(+0x24fec) [0x14d73b078fec]
2 /lib/x86_64-linux-gnu/libucs.so.0(+0x251aa) [0x14d73b0791aa]
3 ./wrf.exe() [0x239d4e5]
4 ./wrf.exe() [0x2360a34]
5 ./wrf.exe() [0x235874f]
6 ./wrf.exe() [0x234a3bc]
7 ./wrf.exe() [0x1d207f3]
8 ./wrf.exe() [0x1e56aff]
9 ./wrf.exe() [0x16de55c]
10 ./wrf.exe() [0x14fc568]
11 ./wrf.exe() [0x5b01c3]
12 ./wrf.exe() [0x40da21]
13 ./wrf.exe() [0x40d9df]
14 ./wrf.exe() [0x40d97d]
15 /lib/x86_64-linux-gnu/libc.so.6(+0x29d90) [0x14d8a2a00d90]
16 /lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0x80) [0x14d8a2a00e40]
17 ./wrf.exe() [0x40d895]
=================================
forrtl: severe (174): SIGSEGV, segmentation fault occurred

I don't see this error when I use similar namelist.input and namelist.wps files for lower latitude simulations (which use the mercator and lambert projections). At this point I think I'm using the polar projection incorrectly and could use help debugging what is going on. Attached are my namelist.input, namelist.wps and rs1.error.0000 files.

FYI I shortened the start to end interval as I tried to debug what's going on so some of my write intervals (like auxshist3_interval) may not make sense since they were set for longer runs.

Thanks in advance!

 

[Ming Chen]

Hi,
Your namelist.wps looks fine to me. However, I have a few concerns on the namelist.input:
(1) parent_time_step_ratio = 1, 3, 3 should be parent_time_step_ratio = 1, 5, 5
(2) time_step can be larger, for example, 120 or 150
(3) radt should be smaller. You may change it to 30
(4) change gwd_opt = 1, 1, 1, to gwd_opt = 1, 0, 0,
This case failed after running about 1-hour. I am suspicious that some physics/dynamics could be wrong. Please modify your namelist.input and try again.

 

[dstabile]

Hi @Ming Chen,

Thanks for the feedback.

I've since decided to reduce the number of domains to 3 starting and ending with 25km and 1km resolutions. I've incorporated your suggestions although I made radt = 15 to match the changes I made to the domain configuration. I've played around with many physics and dynamics configurations but I still run into the same segmentation fault. I think I made things worse because wrf.exe crashes within minutes now.

Currently, I am using physics that I got from this paper and dynamics from this source. Would you be able to look at my attached namelists and let me know if I'm doing anything incorrect?

Thanks!

Note: I realized I uploaded the wrong files. They've now been updated.

 

[Ming Chen]

There is no other error information except the segmentation fault in your rsl file. Seg fault can be caused by various factors and we would like to know exactly when and where the model crashed for what reason. So, would you please compile WRF in debug mode, then rerun this case to obtain some helpful information?

Another issue is that, we seldom run WRF over polar region and I am not sure whether any special tuning is required. I know that a group in OSU develops a WRF system specifically for polar region. Please take a look at the polar WRF at:

https://polarmet.osu.edu/PWRF/

Probably you can try PWRF and hope it works for you.