Quick Start Guide

Document Purpose: This guide is designed for end users who want to quickly start using hydromodel for hydrological modeling. For developers who need detailed understanding of the code architecture, please refer to Usage Guide.

Get started with hydromodel in 5 minutes! This guide walks you through a complete workflow using command-line scripts - no complex coding required.

What You'll Learn

✅ Install hydromodel and prepare data
✅ Calibrate a hydrological model
✅ Evaluate model performance
✅ Run simulations with custom parameters
✅ Visualize results

Prerequisites

Python 3.9 or higher
Basic command-line knowledge
Understanding of hydrological modeling concepts

Step 1: Installation (2 minutes)

Install hydromodel with data support:

pip install hydromodel hydrodataset

Or using uv (faster):

uv pip install hydromodel hydrodataset

Verify installation:

python -c "import hydromodel; print(hydromodel.__version__)"

Step 2: Get the Scripts (1 minute)

Clone the repository to access example scripts and configs:

git clone https://github.com/OuyangWenyu/hydromodel.git
cd hydromodel

The key files you'll use:

hydromodel/
├── scripts/                         # Command-line scripts
│   ├── run_xaj_calibration.py      # Calibrate models
│   ├── run_xaj_evaluate.py         # Evaluate performance
│   ├── run_xaj_simulate.py         # Run simulations
│   └── visualize.py                 # Visualize results
└── configs/                         # Configuration files
    ├── example_config.yaml          # Example calibration config
    ├── example_simulate_config.yaml # Example simulation config
    └── example_xaj_params.yaml      # Example parameters

Step 3: Prepare Data (5-30 minutes)

Option A: Use CAMELS Public Data (Recommended for First Try)

The data downloads automatically on first use:

# Check available basins
from hydrodataset.camels_us import CamelsUs
from hydrodataset import SETTING

data_path = SETTING["local_data_path"]["datasets-origin"]
ds = CamelsUs(data_path, download=True)
basin_ids = ds.read_object_ids()

print(f"Available basins: {len(basin_ids)}")
print(f"Example IDs: {basin_ids[:5]}")

First download may take 30-120 minutes for the complete CAMELS dataset (~70GB). See Data Guide for details.

Option B: Use Your Own Data

See Data Guide - Custom Data Section for preparing your own basin data.

Step 4: Configure Your Experiment (2 minutes)

Edit configs/example_config.yaml:

# Configuration for model calibration and evaluation
data_cfgs:
  data_source_type: "camels_us"                 # Dataset type
  basin_ids: ["01013500"]                       # Basin(s) to calibrate
  train_period: ["1990-10-01", "2000-09-30"]   # Calibration period
  test_period: ["2000-10-01", "2010-09-30"]    # Evaluation period
  warmup_length: 365                            # Warmup days
  variables: ["precipitation", "potential_evapotranspiration", "streamflow"]

model_cfgs:
  model_name: "xaj_mz"                          # XAJ model with Muskingum routing
  model_params:
    source_type: "sources"
    source_book: "HF"

training_cfgs:
  algorithm: "SCE_UA"                           # Algorithm: SCE_UA, GA, or scipy

  # SCE-UA (Shuffled Complex Evolution) - Recommended for global optimization
  SCE_UA:
    rep: 1000                                   # Iterations (5000+ for production)
    ngs: 1000                                   # Number of complexes
    kstop: 500                                  # Stop if no improvement
    peps: 0.1                                   # Parameter convergence
    pcento: 0.1                                 # Percentage change allowed
    random_seed: 1234

  # GA (Genetic Algorithm) - Flexible and customizable
  GA:
    pop_size: 80                                # Population size
    n_generations: 50                           # Generations (100+ for production)
    cx_prob: 0.7                                # Crossover probability
    mut_prob: 0.2                               # Mutation probability
    random_seed: 1234

  # scipy - Fast gradient-based optimization
  scipy:
    method: "SLSQP"                             # L-BFGS-B, SLSQP, TNC, etc.
    max_iterations: 500                         # Maximum iterations

  loss: "RMSE"                                  # Loss function: RMSE, NSE, KGE
  output_dir: "results"
  experiment_name: "quickstart_exp"

evaluation_cfgs:
  metrics: ["NSE", "KGE", "RMSE", "PBIAS"]

Quick Tips:

Start with one basin for first试验
Choose an algorithm:
- SCE_UA: Most robust, good for complex problems (slower)
- GA: Flexible, good balance of speed and accuracy
- scipy: Fastest, good for smooth objective functions
For quick testing: reduce iterations
- SCE_UA: rep: 1000, ngs: 100
- GA: pop_size: 50, n_generations: 30
- scipy: max_iterations: 200
For production: use higher values
- SCE_UA: rep: 10000, ngs: 1000
- GA: pop_size: 100, n_generations: 200
- scipy: max_iterations: 1000

Step 5: Run Calibration (5-30 minutes)

python scripts/run_xaj_calibration.py --config configs/example_config.yaml

You'll see output like:

================================================================================
XAJ Model Calibration using UnifiedCalibrateor
================================================================================

[1/3] Loading configuration...
  Dataset: camels_us
  Basins: ['01013500']
  Model: xaj_mz
  Algorithm: SCE_UA
  Training period: 1990-10-01 to 2000-09-30

[2/3] Loading data...
  Input shape: (3653, 1, 2)
  Qobs shape: (3653, 1, 1)

[3/3] Starting calibration...
  Basin 01013500: SCE-UA optimization
  Progress: [====================] 100% Complete
  Best RMSE: 0.2534

✓ Calibration complete! Results saved to: results/quickstart_exp/

Results location:

results/quickstart_exp/
├── calibration_results.json         # ⭐ Best parameters (unified format, used by evaluation)
├── 01013500_sceua.csv              # SCE-UA iteration history (if using SCE_UA)
├── 01013500_ga.csv                 # GA generation history (if using GA)
├── 01013500_scipy.csv              # scipy iteration history (if using scipy)
├── calibration_config.yaml          # Config used (for reproducibility)
└── param_range.yaml                 # Parameter ranges (optional, for denormalization)

Understanding the results:

calibration_results.json: Contains best parameters for all basins, works with all algorithms
Algorithm-specific CSV: Detailed iteration/generation history with all parameter values
param_range.yaml: Defines parameter physical ranges (not calibration results)

Step 6: Evaluate Model Performance (1 minute)

Evaluate on the test period:

python scripts/run_xaj_evaluate.py \
    --calibration-dir results/quickstart_exp \
    --eval-period test

Output:

================================================================================
XAJ Model Evaluation
================================================================================

[1/3] Loading calibrated parameters...
  Found parameters for 1 basin(s)

[2/3] Running evaluation on test period...
  Basin 01013500: Simulating...
  ✓ Simulation complete

[3/3] Calculating metrics...

================================================================================
Evaluation Results (Test Period: 2000-10-01 to 2010-09-30)
================================================================================
Basin: 01013500
  NSE:   0.756
  KGE:   0.721
  RMSE:  0.312
  PBIAS: -5.23 %

✓ Results saved to: results/quickstart_exp/evaluation_test/

Results include:

results/quickstart_exp/evaluation_test/
├── basins_metrics.csv                  # Performance metrics
├── basins_denorm_params.csv            # Calibrated parameters
└── xaj_mz_evaluation_results.nc        # Full simulation results (NetCDF)

Step 7: Run Custom Simulation (1 minute)

Important: Simulation does NOT require calibration! You can run simulations with any parameters.

Option A: Use Calibrated Parameters

python scripts/run_xaj_simulate.py \
    --param-file results/quickstart_exp/01013500_sceua.csv \
    --plot

Option B: Use Custom Parameters

Edit configs/example_xaj_params.yaml:

# XAJ model parameters
K: 0.75
B: 0.25
IM: 0.06
UM: 18.0
LM: 80.0
DM: 95.0
C: 0.18
SM: 120.0
EX: 1.5
KI: 0.35
KG: 0.45
A: 0.85
THETA: 0.012
CI: 0.85
CG: 0.95

Then run:

python scripts/run_xaj_simulate.py \
    --config configs/example_simulate_config.yaml \
    --param-file configs/example_xaj_params.yaml \
    --output simulation_results.csv \
    --plot

Output:

================================================================================
XAJ Model Simulation using UnifiedSimulator
================================================================================

[1/4] Loading configuration from: configs/example_simulate_config.yaml
  Model: xaj_mz
  Basin: 01013500 (index 0)
  Period: ['2000-10-01', '2010-09-30']

[2/4] Loading parameters from: configs/example_xaj_params.yaml
  Parameters:
    K        = 0.750000
    B        = 0.250000
    IM       = 0.060000
    ...

[3/4] Loading data and initializing simulator
  Input shape: (3653, 1, 2)
  Qobs shape: (3653, 1, 1)
  ✓ UnifiedSimulator initialized

[4/4] Running simulation (warmup=365 days)
  ✓ Simulation completed (3288 time steps)

================================================================================
Simulation Results
================================================================================
Basin: 01013500
Time steps: 3288

Performance Metrics:
  NSE        =   0.7234
  KGE        =   0.6912
  RMSE       =   0.3456

✓ Results saved to: simulation_results.csv

Step 8: Visualize Results (1 minute)

python scripts/visualize.py --eval-dir results/quickstart_exp/evaluation_test

This creates plots showing:

Observed vs. simulated streamflow
Flow duration curves
Monthly aggregated comparison
Residual analysis

Output files:

results/quickstart_exp/evaluation_test/
├── 01013500_timeseries.png          # Time series plot
├── 01013500_flow_duration.png       # FDC plot
└── 01013500_monthly.png             # Monthly comparison

Common Workflows

Workflow 1: Calibration → Evaluation → Visualization

# 1. Calibrate
python scripts/run_xaj_calibration.py --config configs/example_config.yaml

# 2. Evaluate
python scripts/run_xaj_evaluate.py \
    --calibration-dir results/quickstart_exp \
    --eval-period test

# 3. Visualize
python scripts/visualize.py --eval-dir results/quickstart_exp/evaluation_test

Workflow 2: Simulation Only (No Calibration)

# Use custom parameters directly
python scripts/run_xaj_simulate.py \
    --config configs/example_simulate_config.yaml \
    --param-file configs/example_xaj_params.yaml \
    --plot

Workflow 3: Multi-Basin Calibration

Edit config to include multiple basins:

data_cfgs:
  basin_ids: ["01013500", "01022500", "01030500"]  # Multiple basins

Then run:

python scripts/run_xaj_calibration.py --config configs/multi_basin_config.yaml

Tips for Success

1. Start Small

Use one basin initially
Use shorter periods for testing
Use fewer iterations (rep: 1000, ngs: 20)

2. Check Data Quality

# Verify your data before calibration
from hydrodataset.camels_us import CamelsUs

ds = CamelsUs(data_path)
data = ds.read_timeseries(
    gage_id_lst=["01013500"],
    t_range=["1990-10-01", "2000-09-30"],
    var_lst=["precipitation", "streamflow"]
)

print(f"Data shape: {data.shape}")
print(f"Missing values: {data.isna().sum()}")

3. Monitor Progress

Calibration saves checkpoints every N iterations
Check partial results in results/ directory
Use --verbose flag for detailed output

4. Production Settings

For real research, use longer calibration:

training_cfgs:
  algorithm_params:
    rep: 10000        # More iterations
    ngs: 100          # More complexes
    kstop: 50         # Stricter convergence

Troubleshooting

Issue 1: "Data not found" Error

Solution: Check data path and basin IDs

from hydrodataset import SETTING
from hydrodataset.camels_us import CamelsUs

# Check data path
print(SETTING["local_data_path"]["datasets-origin"])

# Check basin IDs
ds = CamelsUs(SETTING["local_data_path"]["datasets-origin"])
basin_ids = ds.read_object_ids()
print(f"Available basins: {len(basin_ids)}")

Issue 2: Slow Calibration

Solutions:

Use fewer basins initially
Reduce rep and ngs for testing
Use xaj_mz (faster than full XAJ)
Consider using GA or scipy algorithms

Issue 3: Poor Model Performance

Check:

Data quality (missing values, outliers)
Training period length (need ≥5 years)
Warmup period (use 365 days minimum)
Parameter ranges (default ranges may not suit all basins)

Issue 4: Memory Issues

Solutions:

Process fewer basins at once
Reduce time period length
Use data caching (see Data Guide)

What's Next?

1. Explore Advanced Features

Different Datasets: Try CAMELS-GB, CAMELS-AUS, etc.

data_cfgs:
  data_source_type: "camels_gb"
  basin_ids: ["28015"]

Different Algorithms: Try GA or scipy
```
training_cfgs:
  algorithm_name: "GA"
```

Custom Periods: Evaluate on different periods

python scripts/run_xaj_evaluate.py \
    --calibration-dir results/quickstart_exp \
    --eval-period custom \
    --custom-period 2010-01-01 2015-12-31

2. Use Your Own Data

Follow Data Guide to prepare custom basin data.

3. Compare Algorithms

Try different algorithms to see which works best for your case:

# Method 1: Edit algorithm in config
# In configs/example_config.yaml, change:
#   algorithm: "SCE_UA"  # to "GA" or "scipy"

# Method 2: Run comparisons
python scripts/run_xaj_calibration.py --config configs/example_config_sceua.yaml
python scripts/run_xaj_calibration.py --config configs/example_config_ga.yaml
python scripts/run_xaj_calibration.py --config configs/example_config_scipy.yaml

Algorithm comparison:

Algorithm	Speed	Robustness	Memory	Best For
SCE-UA	Slow	⭐⭐⭐⭐⭐	High	Complex landscapes, global optimum
GA	Medium	⭐⭐⭐⭐	Medium	Flexible, good balance
scipy	⭐⭐⭐⭐⭐	⭐⭐⭐	Low	Smooth objectives, quick tests

Convergence analysis:

import pandas as pd
import matplotlib.pyplot as plt

# Load iteration history
df_ga = pd.read_csv("results/xaj_GA/01013500_ga.csv")
df_scipy = pd.read_csv("results/xaj_scipy/01013500_scipy.csv")

# Plot convergence
plt.figure(figsize=(10, 6))
plt.plot(df_ga["generation"], df_ga["objective_value"], label="GA")
plt.plot(df_scipy["iteration"], df_scipy["objective_value"], label="scipy")
plt.xlabel("Iteration/Generation")
plt.ylabel("Objective Value (RMSE)")
plt.legend()
plt.title("Convergence Comparison")
plt.show()

4. Learn Code Architecture

For deeper understanding, read Usage Guide - the developer documentation.

5. API Usage

Transition from scripts to Python API for more flexibility:

from hydromodel.trainers.unified_calibrate import calibrate
from hydromodel.trainers.unified_evaluate import evaluate

# Load configuration
config = {...}

# Run calibration
results = calibrate(config)

# Run evaluation
eval_results = evaluate(config, param_dir="results/exp", eval_period="test")

See Usage Guide for API details.

Summary

Congratulations! You've learned how to:

✅ Install hydromodel and prepare data
✅ Configure experiments using YAML files
✅ Calibrate models using command-line scripts
✅ Evaluate model performance on test periods
✅ Run simulations with custom parameters
✅ Visualize results

Key Commands:

# Calibration
python scripts/run_xaj_calibration.py --config configs/example_config.yaml

# Evaluation
python scripts/run_xaj_evaluate.py --calibration-dir results/exp --eval-period test

# Simulation
python scripts/run_xaj_simulate.py --param-file results/exp/basin_sceua.csv --plot

# Visualization
python scripts/visualize.py --eval-dir results/exp/evaluation_test

Additional Resources

Usage Guide: usage.md - Developer documentation with code architecture details
Data Guide: data_guide.md - Comprehensive data preparation guide
FAQ: faq.md - Common questions and solutions
GitHub: https://github.com/OuyangWenyu/hydromodel
Issues: https://github.com/OuyangWenyu/hydromodel/issues

Happy modeling! 🌊

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Quick Start Guide

What You'll Learn

Prerequisites

Step 1: Installation (2 minutes)

Step 2: Get the Scripts (1 minute)

Step 3: Prepare Data (5-30 minutes)

Option A: Use CAMELS Public Data (Recommended for First Try)

Option B: Use Your Own Data

Step 4: Configure Your Experiment (2 minutes)

Step 5: Run Calibration (5-30 minutes)

Step 6: Evaluate Model Performance (1 minute)

Step 7: Run Custom Simulation (1 minute)

Option A: Use Calibrated Parameters

Option B: Use Custom Parameters

Step 8: Visualize Results (1 minute)

Common Workflows

Workflow 1: Calibration → Evaluation → Visualization

Workflow 2: Simulation Only (No Calibration)

Workflow 3: Multi-Basin Calibration

Tips for Success

1. Start Small

2. Check Data Quality

3. Monitor Progress

4. Production Settings

Troubleshooting

Issue 1: "Data not found" Error

Issue 2: Slow Calibration

Issue 3: Poor Model Performance

Issue 4: Memory Issues

What's Next?

1. Explore Advanced Features

2. Use Your Own Data

3. Compare Algorithms

4. Learn Code Architecture

5. API Usage

Summary

Additional Resources

FilesExpand file tree

quickstart.md

Latest commit

History

quickstart.md

File metadata and controls

Quick Start Guide

What You'll Learn

Prerequisites

Step 1: Installation (2 minutes)

Step 2: Get the Scripts (1 minute)

Step 3: Prepare Data (5-30 minutes)

Option A: Use CAMELS Public Data (Recommended for First Try)

Option B: Use Your Own Data

Step 4: Configure Your Experiment (2 minutes)

Step 5: Run Calibration (5-30 minutes)

Step 6: Evaluate Model Performance (1 minute)

Step 7: Run Custom Simulation (1 minute)

Option A: Use Calibrated Parameters

Option B: Use Custom Parameters

Step 8: Visualize Results (1 minute)

Common Workflows

Workflow 1: Calibration → Evaluation → Visualization

Workflow 2: Simulation Only (No Calibration)

Workflow 3: Multi-Basin Calibration

Tips for Success

1. Start Small

2. Check Data Quality

3. Monitor Progress

4. Production Settings

Troubleshooting

Issue 1: "Data not found" Error

Issue 2: Slow Calibration

Issue 3: Poor Model Performance

Issue 4: Memory Issues

What's Next?

1. Explore Advanced Features

2. Use Your Own Data

3. Compare Algorithms

4. Learn Code Architecture

5. API Usage

Summary

Additional Resources