SeaDAS new l2gen very slow

[melinfr]

Dear Alex,
thanks a lot for your feedback, this is really appreciated.
I am also using l2gen on the command line (from a shell, I tried csh and bash).
The test image I am using is actually ~ twice smaller (7848 lines) because it is cut at 10 deg. North.
With l2gen associated with V2023 (9.6.0-V2023.3), it took 1066 seconds (so similar to your calculation) but 8456'' for V2025 (9.11.0-V2025.1).
At that stage, the only test I still envision is to try on other operating systems. Or if you have other ideas?
Thanks again,
Best,
F.

[OB ScienceSW - alex]

You could try a different OS.
Also, could you compare the output files produced with the V2023 and V2025 tags? An L2 file has a group named processing controls, which contains all the input parameters and files. These should be the same in both tags.

[roger_scott]

For those interested I was also seeing a similar performance issue which after some digging I tracked down to a change in the code related to reading the no2 climatology. There's a HDF format check which was causing the file to be frequently opened and closed (multiple Hishdf calls) which is more noticeable with slower or networked storage. The solution was to patch the code so the format check is only done once as the file is unlikely to change for the duration of processing (see patch below). It would be good if this or something similar could be merged into the distribution.

Regards,
Roger.
-----------------------------------------
--- setanc.c.orig 2025-09-10 11:23:12.712060000 +1000
+++ setanc.c 2025-09-10 11:23:44.187100000 +1000
@@ -246,4 +246,5 @@
void no2conc(char *no2file, float lon, float lat, int32_t doy, float *no2_tropo, float *no2_strat) {
static int firstCall = 1;
+ static int isHDFfile = -1;
static int nx = NXNO2;
static int ny = NYNO2;
@@ -265,6 +266,9 @@
*no2_strat = 0.0;

- if (!Hishdf(no2file)) {
+ if (firstCall) isHDFfile = Hishdf(no2file);
+
+ if (!isHDFfile) {
no2concGeosCf(no2file, lon, lat, no2_tropo, no2_strat);
+ firstCall = 0;
return;
}

[OB.DAAC - SeanBailey]

Roger,

Yep, that was discovered recently by Alex as well and will be corrected in an upcoming release. The wee bit o'code whoopsie that caused the woe was introduced in August 2024 (in the source code),which jives with Fred's report that the slowness began with V2025.1.

Regards,
Sean

[melinfr]

Dear Sean and Roger,
thanks a lot for the feedback! This could really mean a lot for us.
If I understand well, it means updating setanc.c but there would be a need for quite a compiling effort, wouldn't there ? Or could we expect a software update in the short term?
Thanks again!
Best, F.

[OB.DAAC - SeanBailey]

Yes, a test tag with this fix is being prepped and should be available sometime next week. It will take a little longer to get into a release tag, but as SeaDAS is planning another release in October, it shouldn't be too long of a wait

Regards,
Sean

[melinfr]

Dear Sean,
this would be wonderful, thanks a lot! I'll watch for the new test version and release.
Best,
F.