Runtime error with ECMWF forecast data and WPSV3.8

[tarunjsph48]

Hi,

I have been using ECMWF forecast data (GRIB 1 format) to run WRF (Version 3.8). Of late I have compiled a collection of errors (ungrib_expanded_errors.txt gives the detailed backtrace of each error, condensed_overview_runtime_errors_ungrib.ods gives brief summary of the errors faced for each variable provided in the Vtable) that I have encountered while executing ungrib.exe on ECMWF forecast files. I have noticed that some errors are unique to the variable specified in the Vtable at the time of execution of ungrib.exe. The variables that I'm interested in are provided in the Vtable (please see the attachment named Vtable). These errors were collated when ungrib.exe was executed each time for a Vtable that consisted of only one variable of interest at a time.

I've attempted to troubleshoot this scenario using the following ways:
Attempt 1
I've checked the content of GRIB files to ensure that the variable of interest mentioned in Vtable is available in the file using g1print.exe (please have a look at the attachment named g1print.txt). All the files appear to contain all the variables mentioned in the Vtable.

Attempt 2
I've been using WPS 3.8 for ungribbing the ECMWF forecast files. I've even upgraded the WPS suite to version 3.9.0.1 to attempt the same. However, it's throwing the same errors as ungrib.exe of WPS 3.8 suite.

Attempt 3
For errors that involve segmentation fault (core dumped), I've tried using 'ulimit -s unlimited' command prior to the execution of ungrib, yet the same set of errors still persist.

How can we go about resolving the run time errors mentioned in the ungrib_expanded_errors.txt file?

Kind regards,
Tarun Joseph.

 

[Ming Chen]

What is the grid structure of the ECMWF forecast data? We are aware that some data on special grids like thinned gaussian grid cannot be processed by WPS, even if the Vtable is correct.

 

[tarunjsph48]

I've investigated into the grid structure of the ECMWF forecast files. All the variables of interest were found to follow regular_ll grid structure (https://apps.ecmwf.int/codes/grib/format/edition-independent/1/19/). I was able to run the ungrib.exe on ECMWF forecast files with regular_ll grid structure before. However, I've noticed that ungrib.exe run began to break often for ECMWF forecast files around the time when a change in the file structure was implemented (https://confluence.ecmwf.int/display/UDOC/New+Product+Generation+software#NewProductGenerationsoftware-Differencesonthedatafiles). Could it be because the ECMWF forecast files in the new format are in some way incompatible with ungrib.exe of WPS version 3.8?

 

[Ming Chen]

That is quite possible. We often see ungrib no longer works once the variables/grid structure of the input data is changed. Please explore the new data and make sure Vtable and source codes of ungrib can be adjusted accordingly.