Atlas 14 Spatial Variance Analysis

User-invocable skill for assessing precipitation spatial variability in HEC-RAS models

Primary Sources (Read These First)

Complete Workflows (lines 540-629):

• ras_commander/precip/CLAUDE.md
  • Lines 118-197: Module documentation (Atlas14Grid, Atlas14Variance)
  • Lines 540-629: Complete Atlas 14 Grid Workflow (4 steps)
  • Lines 522-538: Performance metrics

API Reference:

• ras_commander/precip/Atlas14Grid.py (404 lines)

  • get_pfe_from_project() - Main entry point for HEC-RAS integration
  • get_pfe_for_bounds() - Direct bounding box query
  • get_point_pfe() - Single point lookup

• ras_commander/precip/Atlas14Variance.py (322 lines)

  • analyze() - Full variance analysis
  • analyze_quick() - Rapid assessment (100-yr, 24-hr)
  • is_uniform_rainfall_appropriate() - Decision support
  • generate_report() - Export with plots

Working Example:

• examples/725_atlas14_spatial_variance.ipynb
  • Direct bounds queries
  • Point lookups
  • HEC-RAS project integration
  • Visualization

Quick Reference:

• .claude/rules/hec-ras/precipitation.md (lines 104-126)

---

Quick Reference

Typical Workflow (3 Steps)

from ras_commander.precip import Atlas14Variance

# Step 1: Quick check for representative event
stats = Atlas14Variance.analyze_quick("MyProject.g01.hdf")

# Step 2: Interpret results
if stats['range_pct'] > 10:
    # High variance - run full analysis
    results = Atlas14Variance.analyze(
        geom_hdf="MyProject.g01.hdf",
        durations=[6, 12, 24],
        return_periods=[10, 25, 50, 100]
    )

    # Step 3: Generate report
    Atlas14Variance.generate_report(
        results,
        output_dir="Atlas14_Report",
        project_name="My Project"
    )
else:
    # Low variance - uniform rainfall OK
    print("✓ Uniform rainfall appropriate")

Direct Grid Access

from ras_commander.precip import Atlas14Grid

# Get PFE for project extent
pfe = Atlas14Grid.get_pfe_from_project(
    geom_hdf="MyProject.g01.hdf",
    extent_source="2d_flow_area",  # or "project_extent"
    durations=[6, 12, 24],
    return_periods=[10, 50, 100],
    buffer_percent=10.0
)

# Access data
print(f"Grid size: {pfe['lat'].shape[0]} x {pfe['lon'].shape[0]}")
print(f"100-yr 24-hr max: {pfe['pfe_24hr'][:,:,5].max():.2f} inches")

Point Query (No Project)

# Quick lookup for single location
df = Atlas14Grid.get_point_pfe(
    lat=29.76,
    lon=-95.37,
    durations=[6, 12, 24],
    return_periods=[10, 50, 100, 500]
)
print(df)

---

When to Use This Skill

Use Atlas14Variance When:

1. Before rain-on-grid modeling - Assess if uniform rainfall is valid
2. Large model domains (>100 sq mi) - Spatial variance more likely
3. Multi-event analysis - Compare variance across return periods
4. Engineering review - Generate reports for professional documentation

Decision Criteria

Range %	Interpretation	Recommendation
< 5%	Very low variance	Uniform rainfall appropriate
5-10%	Low variance	Uniform likely OK, monitor
10-20%	Moderate variance	Consider spatially variable
> 20%	High variance	Use spatially variable rainfall

Representative Event: 100-year, 24-hour is typical for quick assessment

---

Common Workflows

Workflow 1: Quick Assessment

Purpose: Rapid check if variance analysis is needed

See: ras_commander/precip/CLAUDE.md lines 544-564

from ras_commander.precip import Atlas14Variance

stats = Atlas14Variance.analyze_quick("MyProject.g01.hdf")

if stats['range_pct'] > 10:
    print("⚠️ Run full analysis - high variance detected")

Workflow 2: Full Analysis

Purpose: Comprehensive variance across multiple events

See: ras_commander/precip/CLAUDE.md lines 566-585

results = Atlas14Variance.analyze(
    geom_hdf="MyProject.g01.hdf",
    durations=[6, 12, 24, 48],
    return_periods=[10, 25, 50, 100, 500],
    extent_source="2d_flow_area",
    variance_denominator='min',  # or 'max', 'mean'
    output_dir="Atlas14_Variance_Report"
)

Workflow 3: Report Generation

Purpose: Engineering documentation

See: ras_commander/precip/CLAUDE.md lines 587-605

report_dir = Atlas14Variance.generate_report(
    results_df=results,
    output_dir="reports/",
    project_name="My Project",
    include_plots=True
)

# Files created:
# - variance_statistics.csv
# - variance_summary.csv
# - variance_by_duration.png
# - variance_heatmap.png

Workflow 4: Custom Grid Analysis

Purpose: Export for HEC-RAS import or custom processing

See: ras_commander/precip/CLAUDE.md lines 607-629

from ras_commander.precip import Atlas14Grid

pfe = Atlas14Grid.get_pfe_from_project(
    geom_hdf="MyProject.g01.hdf",
    extent_source="2d_flow_area",
    durations=[24],
    return_periods=[100]
)

# Access raw arrays for custom analysis
lat = pfe['lat']
lon = pfe['lon']
data_100yr_24hr = pfe['pfe_24hr'][:, :, 5]

---

Key Parameters

extent_source

Controls which geometry is used for extent extraction:

• "2d_flow_area" (default, recommended)

  • Uses union of 2D flow area perimeters
  • Most relevant for rain-on-grid models
  • Can filter to specific areas via mesh_area_names

• "project_extent"

  • Uses full project bounding extent
  • Includes 1D and storage areas
  • Larger extent = more data downloaded

use_huc12_boundary

Controls whether to use HUC12 watershed instead of 2D flow area:

• False (default)

  • Uses 2D flow area perimeters or project extent
  • Fastest analysis (smaller area)

• True

  • Finds HUC12 watershed containing center of 2D flow area
  • Downloads HUC12 boundary from NHDPlus
  • Analyzes full contributing watershed
  • Typically higher variance (larger extent)
  • Requires pygeohydro package

Example:

# Analyze using HUC12 watershed
results = Atlas14Variance.analyze(
    geom_hdf="MyProject.g01.hdf",
    use_huc12_boundary=True
)

variance_denominator

Controls how range percentage is calculated:

• 'min' (default) - range_pct = (max - min) / min × 100

  • Shows variance relative to minimum value
  • Matches HEC-Commander approach
  • More sensitive to variance

• 'max' - range_pct = (max - min) / max × 100

  • Shows variance relative to maximum value
  • More conservative metric

• 'mean' - range_pct = (max - min) / mean × 100

  • Engineering standard
  • Balanced perspective

---

Technical Details

NOAA CONUS NetCDF Structure

URL: https://hdsc.nws.noaa.gov/pub/hdsc/data/tx/NOAA_Atlas_14_CONUS.nc

Property	Value
Coverage	CONUS (24°N-50°N, -125°W to -66°W)
Resolution	0.0083° (830m at 30°N)
Format	HDF5-based NetCDF-4
Size	320 MB (chunked for efficient access)
Chunking	(49, 111, 1) - optimized for spatial access
HTTP Support	Accept-Ranges: bytes ✓

Available Data:

• Durations: 1, 2, 3, 6, 12, 24, 48, 72, 96, 168 hours
• Return Periods: 2, 5, 10, 25, 50, 100, 200, 500, 1000 years
• Scale Factor: 0.01 (raw values × 0.01 = inches)

Data Transfer Efficiency

Project Size	Grid Cells	Data Transfer	Full Grid	Reduction
Small (0.5° × 0.5°)	~3,600	~60 KB	379 MB	99.98%
Medium (1° × 1°)	~14,400	~250 KB	379 MB	99.93%
Large (2° × 2°)	~57,600	~1 MB	379 MB	99.74%

Comparison: Traditional approach downloads 50-100 MB per state as ZIP files.

---

Critical Warnings

CONUS Coverage Only

The NOAA CONUS NetCDF covers Continental US only:

Covered: Lower 48 states (24°N-50°N, -125°W to -66°W)

Not Covered:

• Hawaii (use StormGenerator point API instead)
• Alaska (use StormGenerator point API instead)
• Puerto Rico (use StormGenerator point API instead)
• Offshore areas (no data)

Internet Required

Atlas14Grid requires internet access to NOAA servers:

• No offline mode currently implemented
• Cache coordinates in memory (cleared with Atlas14Grid.clear_cache())
• Future: Local disk caching planned

Return Period Mapping

The ari dimension uses return periods, not ARI indices:

ari = [2, 5, 10, 25, 50, 100, 200, 500, 1000]

# 100-year event is index 5
pfe_100yr = pfe['pfe_24hr'][:, :, 5]  # Correct

# NOT index 100
pfe_100yr = pfe['pfe_24hr'][:, :, 100]  # Wrong!

---

Dependencies

Required (already in ras-commander):

• h5py>=3.0.0 - HDF5/NetCDF access
• numpy - Array operations
• pandas - DataFrames
• geopandas>=0.12.0 - Spatial operations
• fsspec>=2023.0.0 - Remote file systems (HTTP)

Optional:

• matplotlib - Plotting (for generate_report())
• rioxarray - Enhanced raster operations (future)

Installation:

pip install ras-commander  # All required deps included

---

Navigation Map

For complete workflows: Read ras_commander/precip/CLAUDE.md lines 540-629

For API details: Read docstrings in:

• ras_commander/precip/Atlas14Grid.py
• ras_commander/precip/Atlas14Variance.py

For working code: Run examples/725_atlas14_spatial_variance.ipynb

For quick reference: See .claude/rules/hec-ras/precipitation.md lines 104-126

For research background: See .claude/outputs/atlas14-variance-research-summary.md

---

Common Questions

Q: When should I use this instead of StormGenerator?

Use Atlas14Grid/Variance when:

• You need spatial precipitation grids (not just point values)
• You want to assess uniform rainfall validity
• You have a large model domain where variance matters

Use StormGenerator when:

• You need point precipitation for a single location
• You want hyetograph generation (temporal distribution)
• You're doing design storm analysis (not variance assessment)

Q: How accurate is the spatial subsetting?

The HTTP range request approach downloads exactly the grid cells within the specified extent. Validation shows:

• ✓ Matches NOAA PFDS web interface values
• ✓ Houston, TX 100-yr 24-hr: 17.00 inches (correct)
• ✓ No data loss from subsetting
• ✓ Scale factor properly applied (0.01)

Q: What if my project spans multiple states?

The CONUS NetCDF covers the entire Continental US in a single file, so:

• ✓ Multi-state projects work automatically
• ✓ No manual merging required
• ✓ No need to specify states

This is a major advantage over HEC-Commander's approach, which requires downloading separate state datasets and manually merging them.

Q: Can I export the grid data to HEC-RAS?

Current: Atlas14Grid returns numpy arrays - export to GeoTIFF/NetCDF not yet implemented

Workaround: Use the data for variance analysis, then use StormGenerator or Atlas14Storm to create HEC-RAS precipitation input files

Future: Planned enhancement for direct GeoTIFF/NetCDF export

---

Skill Invocation

When user asks to:

• "Analyze Atlas 14 spatial variance for my HEC-RAS project"
• "Check if uniform rainfall is appropriate"
• "Get precipitation frequency grids for my 2D flow areas"
• "Assess precipitation spatial variability"

Respond with:

1. Quick Check First:

   from ras_commander.precip import Atlas14Variance
   
   stats = Atlas14Variance.analyze_quick("project.g01.hdf")
   print(f"Range: {stats['range_pct']:.1f}%")
   

2. Interpret Results:

   • Range < 10%: "✓ Uniform rainfall appropriate"
   • Range > 10%: "⚠️ Consider full analysis"

3. Full Analysis if Needed:

   results = Atlas14Variance.analyze("project.g01.hdf")
   ok, msg = Atlas14Variance.is_uniform_rainfall_appropriate(results)
   print(msg)
   

4. Generate Report:

   Atlas14Variance.generate_report(
       results,
       output_dir="Atlas14_Report",
       include_plots=True
   )
   

---

Related Skills and Agents

Related Skills:

• /analyzing-aorc-precipitation - Historic gridded precipitation
• (Future) /atlas14-design-storms - Hyetograph generation

Relevant Agents:

• precipitation-specialist - General precipitation workflows
• hdf-analyst - Extract HEC-RAS geometry for extent

Complementary Classes:

• StormGenerator - Point-based design storms (see ras_commander/precip/CLAUDE.md lines 44-77)
• Atlas14Storm - HMS-equivalent hyetographs (lines 79-116)
• PrecipAorc - Historic AORC data (lines 13-42)

---

Examples by Use Case

Use Case 1: Pre-Modeling Assessment

Scenario: Determine if uniform rainfall is valid before running rain-on-grid model

from ras_commander.precip import Atlas14Variance

# Quick check
stats = Atlas14Variance.analyze_quick("MyProject.g01.hdf")

if stats['range_pct'] <= 10:
    print("Proceed with uniform rainfall")
    # Run HEC-RAS with single precipitation source
else:
    print("Consider spatially variable rainfall")
    # Run full analysis to quantify

Use Case 2: Multi-Event Comparison

Scenario: Compare variance across multiple design storms

results = Atlas14Variance.analyze(
    geom_hdf="MyProject.g01.hdf",
    durations=[6, 12, 24, 48],
    return_periods=[10, 25, 50, 100, 200, 500]
)

# Find highest variance event
worst = results.loc[results['range_pct'].idxmax()]
print(f"Highest variance: {worst['duration_hr']}-hr, {worst['return_period_yr']}-yr")
print(f"Range: {worst['range_pct']:.1f}%")

Use Case 3: Engineering Report

Scenario: Generate documentation for design report

# Full analysis
results = Atlas14Variance.analyze("MyProject.g01.hdf")

# Generate report with plots
report_dir = Atlas14Variance.generate_report(
    results,
    output_dir="Engineering_Report/Atlas14_Variance",
    project_name="Smith Creek Dam",
    include_plots=True
)

# Files suitable for inclusion in engineering documentation

Use Case 4: Specific 2D Flow Areas

Scenario: Analyze variance for specific mesh areas only

pfe = Atlas14Grid.get_pfe_from_project(
    geom_hdf="MyProject.g01.hdf",
    extent_source="2d_flow_area",
    mesh_area_names=["Floodplain_Upper", "Floodplain_Lower"],
    durations=[24],
    return_periods=[100]
)

# Only analyzes specified mesh areas

---

Troubleshooting

"ImportError: fsspec not installed"

Cause: Missing fsspec dependency

Fix:

pip install fsspec>=2023.0.0
# or
pip install --upgrade ras-commander

"ValueError: No data within bounds"

Cause: Project extent outside CONUS coverage

Check:

from ras_commander.hdf import HdfProject

bounds = HdfProject.get_project_bounds_latlon("project.g01.hdf")
print(f"Project bounds: {bounds}")

# Valid CONUS: lon=-125 to -66, lat=24 to 50

Solution: Project must be within Continental US

"High variance but uniform rainfall expected"

Possible Causes:

1. Orographic effects (mountains causing local gradients)
2. Large model domain (>10° extent)
3. Edge of Atlas 14 data coverage

Validation:

• Compare with NOAA PFDS maps visually
• Check if variance concentrated at edges
• Consider physical geography (elevation changes)

"IOError: Cannot access NOAA Atlas 14 CONUS NetCDF"

Causes:

1. Network connectivity issues
2. NOAA server temporarily down
3. Firewall blocking HTTPS

Check:

# Test server availability
if Atlas14Grid.is_available():
    print("NOAA server reachable")
else:
    print("Cannot reach NOAA server - check network")

---

Performance Tips

1. Use Quick Check First

Don't run full analysis unless needed:

# Quick check is fast (~5 seconds)
stats = Atlas14Variance.analyze_quick("project.g01.hdf")

# Only run full analysis if variance is high
if stats['range_pct'] > 10:
    results = Atlas14Variance.analyze("project.g01.hdf")

2. Cache Coordinates

Coordinates are cached after first access:

# First call downloads coordinates (~11 seconds)
pfe1 = Atlas14Grid.get_pfe_for_bounds(...)

# Subsequent calls use cache (much faster)
pfe2 = Atlas14Grid.get_pfe_for_bounds(...)

# Clear cache when done to free memory
Atlas14Grid.clear_cache()

3. Minimize Extent

Smaller extent = less data transfer:

# Use 2D flow areas (smaller) instead of project extent (larger)
pfe = Atlas14Grid.get_pfe_from_project(
    geom_hdf="project.g01.hdf",
    extent_source="2d_flow_area",  # Smaller
    buffer_percent=5.0  # Minimal buffer
)

---

See Also

Primary Documentation:

• ras_commander/precip/CLAUDE.md - Complete precipitation workflows
• .claude/rules/hec-ras/precipitation.md - Quick reference guide

Related Features:

• StormGenerator - Point-based Atlas 14 queries and hyetograph generation
• Atlas14Storm - HMS-equivalent temporal distributions
• PrecipAorc - Historic gridded precipitation data

Example Notebooks:

• examples/725_atlas14_spatial_variance.ipynb - This workflow
• examples/720_atlas14_aep_events.ipynb - Point-based design storms
• examples/900_aorc_precipitation.ipynb - Historic precipitation

Research:

• .claude/outputs/atlas14-variance-research-summary.md - Discovery of CONUS NetCDF approach
• .claude/outputs/atlas14-variance-implementation-summary.md - Implementation details