Intermediary Processing

Before Blender can run, the raw LIS model outputs and MODIS SCF data must be processed into a set of smoothed, quality-controlled input files. This step is handled by the dechowjack/csd_dev repository, which contains MATLAB scripts and supporting Bash utilities.

Working Directory

All csd_dev scripts must be run from:

/discover/nobackup/projects/coressd/csd_dev

A symlink within this directory points to auxiliary data:

aux_data -> /discover/nobackup/projects/coressd/csd_aux_data/aux_data

Auxiliary Data

The aux_data directory contains:

File / Directory	Description
`HUC8_CA/`, `HUC8_US/`	HUC8 watershed shapefiles for Canada and the US
`MeanSCF/NA_MeanSCF_WYxxxx.tif`	Annual mean snow cover fraction by water year
`MODDEM1KM_fixed.tif`	MODIS 1 km DEM
`MODIS_SINUSOIDAL_fixed.prj`	MODIS sinusoidal projection definition
`scf_mask.mat`	Ocean / water body mask

Workflow

Step 1 — Run the MATLAB pipeline

Launch MATLAB from the csd_dev working directory and execute the main orchestration script:

run('csd_main.m')

When prompted, enter the 4-digit water year (e.g., 2016).

The script sequentially calls:

Sub-script	Runtime	Output
`csd_gen_mean_scf_files.m`	~10–15 min	`MeanSCF/NA_MeanSCF_WYxxxx.tif`
`csd_gen_psval_nc.m`	~1 min	`PrecipScalarFiles/WYxxxx/precip_scalar.nc`
`csd_smooth_mass_vars.m` (snowfall)	~5 min	`SmoothedInputs/WYxxxx/Snowf_tavg.nc`
`csd_smooth_mass_vars.m` (SWE)	~15 min	`SmoothedInputs/WYxxxx/SWE_tavg.nc`

Total runtime: approximately 30–40 minutes.

Gaussian Smoother (`csd_smooth_mass_vars.m`)

The csd_smooth_mass_vars.m function applies a Gaussian spatial smoother to the snowfall and SWE fields before they are passed to Blender. The smoother reduces high-frequency spatial noise in the LIS mass inputs while preserving missing-value masks. It uses the custom imgaussfilt_nan helper, which mirrors the behavior of MATLAB imgaussfilt but handles NaN pixels safely.

Step 2 — Fix NetCDF metadata attributes (`fix_smoothed_attrs.sh`)

After exiting MATLAB, run the Bash metadata-fix script to correct NetCDF attributes on the smoothed output files. The script loads NCO on Discover and applies the year-specific attribute fixes with ncatted:

bash fix_smoothed_attrs.sh 2016

The expected smoothed-input science-variable metadata are:

File	Variable	Units	Standard name	Long name
`SWE_tavg.nc`	`SWE_tavg`	`kg m-2`	`liquid_water_content_of_surface_snow`	`snow water equivalent`
`Snowf_tavg.nc`	`Snowf_tavg`	`kg m-2 s-1`	`snowfall_rate`	`snowfall rate`

This step is required because the MATLAB smoothing step can leave incomplete or incorrect NetCDF attributes. Correcting the smoothed-input metadata before Blender runs makes the generated input files easier to inspect and prevents copied template attributes from propagating into later products.

Step 3 — Verify output files

Confirm that the following files exist and are non-empty before proceeding to the Blender run:

coressd/csd_aux_data/aux_data/MeanSCF/NA_MeanSCF_WYxxxx.tif
coressd/Blender/SmoothedInputs/WYxxxx/Snowf_tavg.nc
coressd/Blender/SmoothedInputs/WYxxxx/SWE_tavg.nc
coressd/PrecipScalarFiles/WYxxxx/precip_scalar.nc

Basic metadata checks:

ncdump -h coressd/Blender/SmoothedInputs/WYxxxx/Snowf_tavg.nc
ncdump -h coressd/Blender/SmoothedInputs/WYxxxx/SWE_tavg.nc

Confirm that the science variables use the units and names listed above.

Repository Structure

csd_dev/
├── csd_main.m                  # Orchestrates the full workflow
├── csd_gen_mean_scf_files.m    # Generates climatological mean SCF GeoTIFFs
├── csd_gen_psval_nc.m          # Generates precipitation scalar NetCDF
├── csd_smooth_mass_vars.m      # Gaussian smoother for Snowf_tavg and SWE_tavg
├── fix_smoothed_attrs.sh       # Bash script to fix NetCDF metadata post-MATLAB
├── aux_data -> ...             # Symlink to auxiliary data on Discover
└── functions/                  # Additional preprocessing and helper scripts