Calibration and Optimization#

SYMFLUENCE provides comprehensive calibration capabilities for hydrological models through automated parameter estimation, multi-objective optimization, and robust evaluation frameworks.

Overview#

Model calibration in SYMFLUENCE follows a systematic approach:

  1. Parameter Selection — Define which parameters to calibrate

  2. Objective Functions — Choose appropriate metrics for optimization

  3. Algorithm Configuration — Select optimization strategy

  4. Evaluation — Assess calibrated model performance

Basic Calibration Setup#

Configure calibration in your project YAML:

# Calibration periods
CALIBRATION_PERIOD: "2018-01-01,2018-06-30"
EVALUATION_PERIOD: "2018-07-01,2018-12-31"

# Parameters to calibrate
PARAMS_TO_CALIBRATE: [k_snow, fcapil, newSnowDenMin]

# Optimization settings
OPTIMIZATION_ALGORITHM: DE
OPTIMIZATION_METRIC: KGE
POPULATION_SIZE: 48
NUMBER_OF_ITERATIONS: 30

Parameter Selection#

SUMMA Parameters

Common parameters for calibration:

SETTINGS_SUMMA_PARAMS_TO_CALIBRATE:
  - k_snow          # snow thermal conductivity
  - fcapil          # capillary fringe thickness
  - newSnowDenMin   # minimum new snow density
  - theta_sat       # soil porosity
  - theta_res       # residual soil moisture
  - vGn_alpha       # van Genuchten alpha parameter
  - vGn_n           # van Genuchten n parameter

FUSE Parameters

For FUSE model calibration:

SETTINGS_FUSE_PARAMS_TO_CALIBRATE:
  - alpha           # baseflow recession parameter
  - beta            # percolation parameter
  - k_storage       # storage coefficient
  - qbrate_2c       # baseflow rate
  - percfrac        # percolation fraction

NextGen Parameters

Noah-OWP parameters:

NGEN_NOAH_PARAMS_TO_CALIBRATE:
  - bexp            # pore size distribution
  - dksat           # saturated hydraulic conductivity
  - psisat          # saturated soil potential
  - refkdt          # reference infiltration parameter

Optimization Algorithms#

Differential Evolution (DE)

Robust global optimizer. Recommended for most applications.

OPTIMIZATION_ALGORITHM: DE
POPULATION_SIZE: 48
NUMBER_OF_ITERATIONS: 30
DE_STRATEGY: best1bin
DE_MUTATION_FACTOR: 0.5
DE_CROSSOVER_PROB: 0.7

Dynamically Dimensioned Search (DDS)

Efficient for high-dimensional problems.

OPTIMIZATION_ALGORITHM: DDS
NUMBER_OF_ITERATIONS: 1000
DDS_R: 0.2

Particle Swarm Optimization (PSO)

Good for continuous optimization problems.

OPTIMIZATION_ALGORITHM: PSO
POPULATION_SIZE: 30
NUMBER_OF_ITERATIONS: 50
PSO_INERTIA: 0.9
PSO_COGNITIVE: 2.0
PSO_SOCIAL: 2.0

Multi-Objective (NSGA-II)

For multiple competing objectives.

OPTIMIZATION_ALGORITHM: NSGA-II
OPTIMIZATION_METRICS: [KGE, NSE, PBIAS]
POPULATION_SIZE: 100
NUMBER_OF_ITERATIONS: 50

Objective Functions#

Single Objective

OPTIMIZATION_METRIC: KGE

Available metrics: - KGE — Kling-Gupta Efficiency (recommended) - NSE — Nash-Sutcliffe Efficiency - RMSE — Root Mean Square Error - PBIAS — Percent Bias - R2 — Coefficient of Determination

Multi-Objective

OPTIMIZATION_METRICS: [KGE, NSE, PBIAS]
MULTI_OBJECTIVE_WEIGHTS: [0.5, 0.3, 0.2]

Custom Objectives

Define custom objective functions:

CUSTOM_OBJECTIVE: peak_weighted_kge
PEAK_WEIGHT_THRESHOLD: 0.9  # 90th percentile
PEAK_WEIGHT_FACTOR: 2.0

Advanced Calibration#

Distributed Calibration

Calibrate spatially distributed parameters:

DISTRIBUTED_CALIBRATION: true
SPATIAL_AGGREGATION: hru  # or 'basin', 'elevation_bands'
PARAMETER_REGIONALIZATION:
  - elevation
  - slope
  - land_cover

Multi-Model Calibration

Calibrate multiple models simultaneously:

MODELS_TO_CALIBRATE: [SUMMA, FUSE]
ENSEMBLE_WEIGHTING: performance  # or 'equal', 'bayesian'

Temporal Calibration

Different parameters for different seasons:

TEMPORAL_CALIBRATION:
  winter: [k_snow, newSnowDenMin]
  summer: [theta_sat, vGn_alpha]
SEASON_DEFINITIONS:
  winter: "12-01,03-31"
  summer: "06-01,09-30"

Calibration Execution#

Command Line

# Run calibration step
symfluence workflow step calibrate_model --config my_project.yaml

# Run full workflow including calibration
symfluence workflow run --config my_project.yaml

# Check workflow status
symfluence workflow status --config my_project.yaml

Python API

from symfluence import SYMFLUENCE

# Initialize SYMFLUENCE with configuration
sf = SYMFLUENCE('my_project.yaml')

# Run calibration step
sf.run_calibration()

# Or run the optimization manager directly
from symfluence.optimization import OptimizationManager

opt_manager = OptimizationManager(config, logger)
results = opt_manager.run_optimization()

# Get best parameters from results
best_params = results.get('best_parameters', {})
best_score = results.get('best_score', None)

Results and Evaluation#

Output Files

Calibration produces:

  • calibration_results.csv — Parameter evolution

  • best_parameters.yaml — Optimal parameter set

  • objective_history.png — Convergence plot

  • parameter_sensitivity.csv — Sensitivity analysis

Performance Metrics

EVALUATION_METRICS:
  - KGE
  - NSE
  - RMSE
  - PBIAS
  - R2
  - peak_timing_error
  - volume_error

Validation

Always validate on independent period:

VALIDATION_PERIOD: "2019-01-01,2019-12-31"
SPLIT_SAMPLE_TEST: true

Best Practices#

  1. Parameter Bounds

    Set realistic parameter ranges:

    PARAMETER_BOUNDS:
  k_snow: [0.01, 1.0]
  theta_sat: [0.3, 0.6]

  2. Convergence Criteria

    CONVERGENCE_TOLERANCE: 1e-6
STAGNATION_GENERATIONS: 10

  3. Computational Efficiency

    PARALLEL_CALIBRATION: true
N_CORES: 16
BATCH_SIZE: 8

  4. Robustness Testing

    MONTE_CARLO_RUNS: 100
BOOTSTRAP_VALIDATION: true

Troubleshooting#

Common Issues

  • Slow convergence: Increase population size or iterations

  • Parameter bounds: Check realistic ranges for your domain

  • Memory issues: Reduce batch size or use distributed computing

  • Poor performance: Verify observation data quality

Debugging

DEBUG_CALIBRATION: true
SAVE_INTERMEDIATE_RESULTS: true
PLOT_PARAMETER_EVOLUTION: true

Example Workflows#

Basic Single-Objective

CALIBRATION_PERIOD: "2015-01-01,2017-12-31"
EVALUATION_PERIOD: "2018-01-01,2020-12-31"
PARAMS_TO_CALIBRATE: [k_snow, fcapil, theta_sat]
OPTIMIZATION_ALGORITHM: DE
OPTIMIZATION_METRIC: KGE
POPULATION_SIZE: 30
NUMBER_OF_ITERATIONS: 50

Multi-Objective with Validation

CALIBRATION_PERIOD: "2010-01-01,2015-12-31"
EVALUATION_PERIOD: "2016-01-01,2018-12-31"
VALIDATION_PERIOD: "2019-01-01,2021-12-31"
PARAMS_TO_CALIBRATE: [alpha, beta, k_storage, qbrate_2c]
OPTIMIZATION_ALGORITHM: NSGA-II
OPTIMIZATION_METRICS: [KGE, NSE, PBIAS]
POPULATION_SIZE: 100
NUMBER_OF_ITERATIONS: 100
SPLIT_SAMPLE_TEST: true

—

See Also