FUSE Model Guide#

Overview#

FUSE (Framework for Understanding Structural Errors) is a modular hydrological modeling framework that allows systematic evaluation of model structural uncertainty. FUSE combines different model components to create alternative model structures, enabling users to test hypotheses about dominant hydrological processes.

Key Capabilities:

  • 1,248 unique model structures from component combinations

  • Flexible spatial modes (lumped, semi-distributed, distributed)

  • Elevation band discretization for snow modeling

  • Multiple PET calculation methods

  • Structure ensemble modeling

  • Optional routing integration (mizuRoute)

  • Synthetic hydrograph generation for testing

  • Modular component-based architecture

Typical Applications:

  • Model structural uncertainty quantification

  • Comparative hydrology studies

  • Snow hydrology with elevation gradients

  • Process hypothesis testing

  • Ensemble streamflow prediction

  • Benchmarking other models

Spatial Scales: Point to basin (tested up to 10,000 km²)

Temporal Resolution: Daily (primarily) to sub-daily

Model Structure and Components#

FUSE Philosophy#

Unlike traditional fixed-structure models, FUSE builds models from modular components. Each component represents a conceptual choice for a hydrological process:

Four Core Model Components:

  1. Upper soil layer architecture (2 options)

  2. Lower soil layer architecture (2 options)

  3. Base flow (2 options)

  4. Surface runoff (4 options)

This creates 2 × 2 × 2 × 4 = 32 parent structures.

Additional Decisions:

  • Evapotranspiration (3 options)

  • Interflow (2 options)

  • Time delay mechanisms (2 options)

  • Percolation (3 options)

  • …and more

Total combinations: 1,248 unique model structures

Common FUSE Structures#

Several pre-defined structures replicate well-known models:

Structure 60:  TOPMODEL-like structure
Structure 230: SACRAMENTO-like structure
Structure 342: ARNOVI-like structure
Structure 426: PRMS-like structure
Structure 900: VIC-like structure

In SYMFLUENCE, structure is specified via model ID (60-1248).

Model Components Explained#

Upper Soil Store:

  • tens1a: Single tension storage

  • tens1b: Cascading tension storages

Lower Soil Store:

  • fixedsiz_lower: Fixed-size lower zone

  • tens2pll_2: Tension pool with parallel stores

Baseflow:

  • qb_prms: Non-linear baseflow (PRMS-style)

  • qb_topmodel: Exponential baseflow (TOPMODEL-style)

Surface Runoff:

  • arno_x_vic: ARNO/VIC infiltration excess

  • prms_variant: PRMS saturation excess

  • tmdl_param: TOPMODEL saturation

  • fix_y_frac: Fixed fraction runoff

Spatial Modes in SYMFLUENCE#

FUSE in SYMFLUENCE supports three spatial modes:

Lumped Mode#

Single HRU represents entire basin:

FUSE_SPATIAL_MODE: lumped

# Simple basin-wide model
# Fastest execution
# Good for conceptual studies

Semi-Distributed Mode#

Multiple subcatchments, each with single HRU:

FUSE_SPATIAL_MODE: semi_distributed

# Subcatchments from delineation
# Routed with mizuRoute
# Balances complexity and performance

Distributed Mode#

HRUs within each subcatchment (typically elevation bands):

FUSE_SPATIAL_MODE: distributed

DISCRETIZATION:
  elevation_bands:
    n_bands: 5  # 5 elevation zones per subcatchment

# Detailed snow modeling
# Accounts for elevation gradients
# Most accurate for mountainous basins

Configuration in SYMFLUENCE#

Model Selection#

HYDROLOGICAL_MODEL: FUSE

Key Configuration Parameters#

Installation and Execution#

Parameter

Default

Description

FUSE_INSTALL_PATH

default

Path to FUSE installation

FUSE_EXE

fuse.exe

FUSE executable name

FUSE_TIMEOUT

3600

Execution timeout (seconds)

SETTINGS_FUSE_PATH

default

Working directory for FUSE

SETTINGS_FUSE_FILEMANAGER

default

FUSE file manager path

Spatial Configuration#

Parameter

Default

Description

FUSE_SPATIAL_MODE

distributed

Spatial discretization mode

FUSE_N_ELEVATION_BANDS

1

Number of elevation bands per subcatchment

FUSE_SUBCATCHMENT_DIM

longitude

Dimension for subcatchment extraction

Model Structure#

Parameter

Default

Description

FUSE_DECISION_OPTIONS

{}

Model structure ID or component decisions

FUSE_FILE_ID

null

Custom identifier for output files

Routing Integration#

Parameter

Default

Description

FUSE_ROUTING_INTEGRATION

default

Enable mizuRoute integration

ROUTING_MODEL

null

Set to ‘mizuRoute’ for routing

Calibration#

Parameter

Default

Description

SETTINGS_FUSE_PARAMS_TO_CALIBRATE

null

Comma-separated list of parameters

EXPERIMENT_OUTPUT_FUSE

default

Output directory for experiments

Model Structure Selection#

Using Structure ID#

Simplest approach - specify a structure number:

FUSE_DECISION_OPTIONS:
  smodl:
    model_id: 60  # TOPMODEL-like structure

Using Component Decisions#

Explicitly define each component:

FUSE_DECISION_OPTIONS:
  smodl:
    rferr: 'additive_e'        # Rainfall error model
    arch1: 'onestate_1'        # Upper layer architecture
    arch2: 'tens2pll_2'        # Lower layer architecture
    qsurf: 'arno_x_vic'        # Surface runoff
    qperc: 'perc_f2sat'        # Percolation
    esoil: 'sequential'        # Evaporation
    qintf: 'intflwnone'        # Interflow
    q_tdh: 'rout_gamma'        # Time delay histogram

Structure Ensemble#

Test multiple structures simultaneously:

FUSE_DECISION_OPTIONS:
  ensemble:
    - model_id: 60   # TOPMODEL
    - model_id: 230  # SACRAMENTO
    - model_id: 900  # VIC
    - model_id: 426  # PRMS

# SYMFLUENCE will run all structures and compare performance

Input File Specifications#

FUSE uses ASCII control files and NetCDF forcing data.

Forcing Data (NetCDF)#

File: <basin_name>_input_<mode>.nc

Required variables (distributed mode):

hru         : HRU/elevation band dimension
gru         : Subcatchment dimension
time        : Time coordinate

pptrate     : Precipitation rate [kg m-2 s-1]
airtemp     : Air temperature [K]
spechum     : Specific humidity [kg kg-1]
SWRadAtm    : Shortwave radiation [W m-2]
LWRadAtm    : Longwave radiation [W m-2]
windspd     : Wind speed [m s-1]
airpres     : Air pressure [Pa]

Required variables (lumped mode):

time        : Time coordinate
pptrate     : Basin-averaged precipitation [kg m-2 s-1]
pet         : Potential evapotranspiration [kg m-2 s-1]

Control Files (ASCII)#

File: fm_catch.txt (File Manager)

Contains paths to all FUSE input/output files:

CONTROL_FILE      ./input/fuse_control.txt
FORCING_PATH      ./forcing/
OUTPUT_PATH       ./output/
MBANDS_INFO       ./input/elev_bands.txt

File: fuse_control.txt (Control File)

Specifies model configuration:

! Model structure
SMODL 60

! Simulation period
SSTRT 2015 01 01 00
SFINISH 2020 12 31 23

! Timestep (seconds)
DT 86400

Elevation Bands File#

File: elev_bands.txt (for distributed mode)

! Elevation band information
! nBands  lowerElev  upperElev  midElev  z_weight  area_frac
5         500        1000       750      0.2       0.15
5         1000       1500       1250     0.2       0.25
5         1500       2000       1750     0.2       0.30
5         2000       2500       2250     0.2       0.20
5         2500       3000       2750     0.2       0.10

Output File Specifications#

FUSE outputs NetCDF files with states and fluxes.

Standard Output (NetCDF)#

File: <basin>_<structure>_<mode>_output.nc

Output variables:

time              : Time coordinate

# Fluxes [mm/day or mm/timestep]
total_q           : Total runoff
eff_ppt           : Effective precipitation
evap_1            : Evaporation from upper layer
evap_2            : Evaporation from lower layer
q_sf              : Surface runoff
q_if              : Interflow
q_perc            : Percolation
q_base            : Baseflow

# States [mm]
tens_1            : Upper tension storage
tens_2            : Lower tension storage
free_1            : Upper free storage
free_2            : Lower free storage
watr_1            : Upper water storage
watr_2            : Lower water storage

Routed Output (with mizuRoute)#

File: <basin>_routed_streamflow.nc

time              : Time coordinate
seg               : River segment ID

dlayRunoff        : IRF-routed runoff [m³/s]
instRunoff        : Instantaneous runoff [m³/s]
KWTroutedRunoff   : KWT-routed runoff [m³/s]

Model-Specific Workflows#

Basic FUSE Workflow (Lumped)#

Simplest FUSE application:

# config.yaml
DOMAIN_NAME: my_basin
HYDROLOGICAL_MODEL: FUSE

# Lumped mode
FUSE_SPATIAL_MODE: lumped
DOMAIN_DEFINITION_METHOD: polygon
CATCHMENT_SHP_PATH: ./basin_boundary.shp

# Structure selection
FUSE_DECISION_OPTIONS:
  smodl:
    model_id: 60  # TOPMODEL structure

# Forcing
FORCING_DATASET: ERA5
FORCING_START_YEAR: 2010
FORCING_END_YEAR: 2015

Run:

symfluence workflow run --config config.yaml

Distributed FUSE with Elevation Bands#

For snow-dominated basins:

# config.yaml
DOMAIN_NAME: mountain_basin
HYDROLOGICAL_MODEL: FUSE

# Distributed mode with elevation bands
FUSE_SPATIAL_MODE: distributed
FUSE_N_ELEVATION_BANDS: 7

# Delineate domain
DOMAIN_DEFINITION_METHOD: semidistributed
POUR_POINT_COORDS: [-120.5, 48.3]

# Create elevation discretization
DISCRETIZATION:
  elevation_bands:
    n_bands: 7
    method: equal_area  # Or: equal_interval

# Structure selection
FUSE_DECISION_OPTIONS:
  smodl:
    model_id: 900  # VIC-like structure (good for snow)

# Optional routing
ROUTING_MODEL: mizuRoute
FUSE_ROUTING_INTEGRATION: mizuRoute

Structure Ensemble Workflow#

Compare multiple model structures:

# config.yaml
HYDROLOGICAL_MODEL: FUSE

# Test 4 different structures
FUSE_DECISION_OPTIONS:
  ensemble:
    - model_id: 60   # TOPMODEL
    - model_id: 230  # SACRAMENTO
    - model_id: 342  # ARNOVI
    - model_id: 900  # VIC

# Calibrate each structure
OPTIMIZATION_ALGORITHM: DDS
OPTIMIZATION_MAX_ITERATIONS: 1000

# Common parameters to calibrate
SETTINGS_FUSE_PARAMS_TO_CALIBRATE: "maxwatr_1,maxwatr_2,baserte,rtfrac1,qbrate_2a"

Result: Best-performing structure identified for your basin.

Multi-Site Distributed FUSE#

For large river networks:

# config.yaml
DOMAIN_DEFINITION_METHOD: merit_basins
MERIT_BASIN_IDS: [10234, 10235, 10236, 10237]  # 4 sub-basins

FUSE_SPATIAL_MODE: distributed
FUSE_N_ELEVATION_BANDS: 5

# Enable routing to combine subcatchments
ROUTING_MODEL: mizuRoute
FUSE_ROUTING_INTEGRATION: mizuRoute

# Structure and calibration
FUSE_DECISION_OPTIONS:
  smodl:
    model_id: 426  # PRMS structure

SETTINGS_FUSE_PARAMS_TO_CALIBRATE: "maxwatr_1,maxwatr_2,baserte"

Calibration Strategies#

Parameter Bounds#

Standard ranges:

# Storage parameters
maxwatr_1:    [10, 500]       # mm
maxwatr_2:    [50, 5000]      # mm
fracten:      [0.01, 0.95]    # fraction tension storage

# Routing parameters
baserte:      [0.001, 1000]   # baseflow rate
rtfrac1:      [0.01, 0.99]    # root fraction
rtfrac2:      [0.01, 0.99]    # root fraction layer 2

# Runoff parameters
percfrac:     [0.01, 0.99]    # percolation fraction
fprimqb:      [0.001, 10]     # baseflow exponent
qbrate_2a:    [0.001, 0.25]   # baseflow coefficient

# Snow parameters (if applicable)
snofall_t:    [-2, 2]         # °C
tempf:        [0.5, 10]       # mm/°C/day

Structure Ensemble Calibration#

When using structure ensembles:

  1. Calibrate each structure independently:

    # Each structure gets its own parameter set
OPTIMIZATION_MAX_ITERATIONS: 1000

  2. Compare performance:

    • Evaluate on calibration period

    • Test on validation period

    • Check for overfitting

  3. Select or weight structures:

    • Use best structure

    • Or create weighted ensemble based on performance

Known Limitations#

  1. Primarily Daily Timestep:

    • FUSE designed for daily modeling

    • Sub-daily possible but less tested

    • Flash flood applications limited

  2. Limited Process Representation:

    • No explicit glacier modeling

    • Simple snow module (degree-day)

    • No detailed energy balance

    • Limited groundwater dynamics

  3. Routing:

    • No internal routing in lumped mode

    • Requires mizuRoute for distributed applications

    • Unit hydrograph approach in some structures

  4. Parameter Identifiability:

    • Some structures have >20 parameters

    • High-dimensional calibration challenging

    • Parameter equifinality common

  5. Computational Considerations:

    • Structure ensembles multiply compute time

    • 1,248 structures = extensive testing required

    • Distributed mode with many bands can be slow

Troubleshooting#

Common Issues#

Error: “FUSE executable not found”

# Solution: Verify installation
FUSE_INSTALL_PATH: /absolute/path/to/fuse/bin
FUSE_EXE: fuse.exe

Error: “Invalid model structure ID”

# Solution: Use valid structure (60-1248)
FUSE_DECISION_OPTIONS:
  smodl:
    model_id: 60  # Valid structure

Error: “Dimension mismatch in forcing data”

Check forcing file dimensions:

ncdump -h forcing_input.nc

# For distributed mode, verify:
# - hru dimension matches n_elevation_bands
# - gru dimension matches n_subcatchments

Error: “FUSE simulation crashed - NaN values”

Common causes:

  1. Extreme parameters → Adjust calibration bounds

  2. Missing forcing data → Check for gaps in precipitation/PET

  3. Very small timestep → Use daily timestep

  4. Invalid structure combination → Use pre-defined structure ID

Elevation bands not created

# Solution: Ensure distributed mode + discretization
FUSE_SPATIAL_MODE: distributed
FUSE_N_ELEVATION_BANDS: 5

DISCRETIZATION:
  elevation_bands:
    n_bands: 5

Routing output missing

# Solution: Enable routing integration
ROUTING_MODEL: mizuRoute
FUSE_ROUTING_INTEGRATION: mizuRoute

Performance Tips#

Speed up calibration:

  1. Use lumped mode during initial testing

  2. Reduce number of elevation bands (3-5 sufficient for most basins)

  3. Start with single structure, not ensemble

  4. Use shorter calibration period (3-5 years)

Improve accuracy:

  1. Use distributed mode for heterogeneous basins

  2. Add elevation bands for snow-dominated regions

  3. Test multiple structures to find best fit

  4. Calibrate on long period (8-10 years) if data available

Debug slowly:

# Run FUSE manually to see detailed output
cd <project_dir>/settings/FUSE/
../../installs/fuse/bin/fuse.exe fm_catch.txt

Additional Resources#

FUSE Documentation:

SYMFLUENCE-specific:

Publications:

  • Clark et al. (2008): “Framework for Understanding Structural Errors (FUSE): A modular framework to diagnose differences between hydrological models”

  • Henn et al. (2015): “Spatial dependence of FUSE model parameters across basins”

  • Knoben et al. (2020): “Modular Assessment of Rainfall-Runoff Models Toolbox (MARRMoT)”

Example Workflow:

# View distributed FUSE example
symfluence examples list | grep FUSE

# Copy example config
symfluence examples copy bow_river_fuse ./my_fuse_project