Issues with GRACE data Preprocessing in LDT and LIS DA Output for LISF Noah-MP DA

[ThangjamSatrajitSingh]

Hi,
I'm running into some LDT's GRACE preprocessing issues that I'd like to discuss and get input on.

Issue 1: LDT Preprocessing of CSR GRACE (No Scaling)
I used the GRACE CSR product as DA input. As the data itself mentioned that no correction is needed, i have kept scaling as none. Below is the config file used:
ldt.config.noahmp401.da-grace.txt

when we compared the ldt generated grace binary files with GRACE CSR + OL_baseline(2004-2009), we got a mismatch:

• Early timesteps have very low magnitude
• Only the final timesteps align with expected TWS values.

Similar test is done with scaling by taking JPL data by keeping everything same. the config file and plot are attached below:
ldt.config.noahmp401.da-grace_JPL.txt

the plot shows that using scaling fixes the problem.

Issue 2: LDT Preprocessing of JPL GRACE (With Scaling)
LIS DA run was conducted using the newly generated binary files. But the LIS runs gave error of integer overflow. So when we checked the values of the processed grace binary files(SCALED JPL), Inspecting the binaries revealed extreme outliers of min= -5824723.00 mm to max=15482984.00 mm(for a sample day).

As a workaround, new binary files were generated without using the ldt but following similar steps (interpolating data to match LIS dimension, scaling, adding baseline calculated from OL, filling missing months using data from the nearest available month that has multiple days of observations.) The custom binaries had reasonable values (e.g., min = 1778.37 mm, max = 7408.64 mm for a sample day) and allowed LIS to run without the integer overflow errors.

Issue 3: Questionable DA Outputs
Despite using the custom binaries, the DA outputs are unphysical when compared to open-loop (OL) and observed data at DSOBS (from LIS data outputs). I have attached the config, DA_input files used and the plot below:
lis.config.noahmp401.da-grace.txt
DA_INPUT.zip

-Absolute values are used instead of anomalies as the comparison are between LIS generated OL, DA and DSOBS outputs.

• DA TWS values diverge from the expected range between OL and DSOBS.

So, i have some queries that i would like to discuss from this:

• Why does LDT preprocessing of CSR GRACE data (no scaling) produce flattened values? Is this a bug in using pre-scaled data or a config issue?
• For JPL data, why do LDT generated grace binaries contain outliers causing integer overflow?
• Can any pointers be provided as to why the DA result is not right? Is it caused by Observation attributes varmax being too low?

Best Regards,
Thangjam Satrajit Singh
Phd Scholar
Hydrosense Group, IIT Delhi