Tree60 Weather Help & Documentation
Overview
Tree60 Weather displays weather forecasts from NOAA, ECMWF, and DWD. The site shows data from multiple weather models, real-time radar, and snow observation networks.
View the onboarding tutorial for a guided introduction to the site.
- Click the map anywhere to get a detailed forecast
- Use search (top left) to find locations by name
- Toggle layers (left panel) for radar, satellite, alerts, etc.
- Change settings (top right) to customize models and units
Understanding the Basics
What Are Weather Models?
Weather models are computer programs that use physics and mathematics to predict future atmospheric conditions. They work by:
- Observation - Collecting current weather data from satellites, weather stations, radar, aircraft, and balloons worldwide
- Initialization - Creating a mathematical representation of the current atmosphere
- Computation - Using supercomputers to solve equations that simulate how the atmosphere evolves over time
- Output - Producing forecasts for temperature, wind, precipitation, pressure, and other variables
Different models use different mathematical approaches and assumptions. No single model is perfect, which is why meteorologists compare multiple models. Where models agree, confidence is high. Where they disagree, the forecast is more uncertain.
- High-resolution models (HRRR, RAP) - Capture small features like thunderstorms, forecast 1-2 days
- Global models (GFS, ECMWF) - Cover entire Earth, forecast 10-16 days, but miss small details
- Ensemble models (GEFS) - Run many versions to estimate forecast uncertainty
Ensemble vs. Deterministic Models
Weather models come in two fundamental types: deterministic and ensemble. Understanding the difference helps you interpret forecast confidence.
Deterministic Models
A deterministic model runs once with a single best estimate of current conditions. It produces one forecast — a single temperature, one precipitation amount, etc. Examples include GFS, HRRR, and DWD ICON.
- Pros: Computationally efficient, provides specific values
- Cons: No indication of uncertainty — you don't know if the forecast is reliable or highly uncertain
Ensemble Models
An ensemble model runs the same forecast many times (called "members") with slightly different starting conditions. This reveals how sensitive the forecast is to small changes. Examples include GEFS (31 members) and ECMWF ENS (51 members).
- Pros: Shows uncertainty range, provides probability estimates
- Cons: Lower resolution than deterministic models (more runs = less detail per run)
| Aspect | Deterministic | Ensemble |
|---|---|---|
| Output | Single forecast | Multiple forecasts (mean, min, max, spread) |
| Uncertainty | Not shown | Shown via spread between members |
| Resolution | Higher (more detail) | Lower (trade-off for multiple runs) |
| Best for | Short-range (1-3 days), precise timing | Medium/long-range (5-35 days), trend confidence |
| Examples | GFS, HRRR, DWD ICON | GEFS, ECMWF ENS |
- Mean: Average of all ensemble members — best single estimate
- Min/Max: Range showing possible outcomes — wider = more uncertain
- Spread: Difference between min and max — narrow spread = high confidence
- Probability: Percentage of members predicting an event (e.g., 70% chance of precipitation means 70% of ensemble members show precipitation)
Model Runs & Initialization Times
Most models don't run continuously—they run at specific times called model runs or cycles.
| Term | Explanation | Example |
|---|---|---|
| Initialization Time | When the model starts calculations using current observations | 12Z = 12:00 UTC (noon Greenwich time) |
| Forecast Hour | Hours into the future from initialization | F+06 = 6 hours after model start |
| Valid Time | The actual time being forecasted | 12Z init + 6 hours = 18Z valid time |
| UTC / Z Time | Coordinated Universal Time (Greenwich) | 00Z = midnight UTC |
Common model run schedules:
- GFS, ECMWF, DWD ICON: 00Z, 06Z, 12Z, 18Z (every 6 hours)
- HRRR: Every hour (CONUS), every 3 hours (Alaska)
- GEFS 35-Day: 00Z only (once daily) — note: NOAA runs GEFS 4x daily but only 00Z extends to 35 days
- ECMWF ENS: 00Z only (once daily)
- NWS NDFD: Updates hourly with latest data
Grid Resolution
Weather models divide Earth into a 3D grid. Each grid cell gets one forecast value. Resolution is the spacing between grid points - smaller numbers mean higher detail.
| Resolution Type | Size | Models | Best For |
|---|---|---|---|
| High Resolution | 2-3km | HRRR, NOAA NWS NDFD | Valleys, coastlines, individual storms |
| Medium Resolution | 10-30km | DWD ICON, RAP | Balance of detail and coverage |
| Coarse Resolution | 25-50km | GFS, GEFS, ECMWF | Global coverage, long-range forecasts |
Effects of Grid Resolution
- Mountains: Valley temperatures can differ by 10-20°F from ridgetops within one grid cell
- Coastlines: Coarse models blend land and water temperatures
- Thunderstorms: Individual cells require 1-3km resolution to predict accurately
- Cities: Urban heat island effects not captured by most models
Weather Models
Quick Comparison
| Model | Resolution | Coverage | Range | Updates | Best For |
|---|---|---|---|---|---|
| NOAA NWS | ~2.5km | US | 7 days | Hourly | Most accurate US forecasts |
| HRRR | 3km | CONUS + AK | 18-48hr | Hourly | Storm timing |
| RAP | 32km | N. America | 51hr | Hourly | Canada, Mexico |
| GFS | 25km | Global | 16 days | 4x daily | Medium-range worldwide |
| GEFS | 25-50km | Global | 35 days | 1x daily (00Z) | Long-range ensemble |
| ECMWF ENS | 25km | Global | 15 days | 1x daily (00Z) | Medium-range ensemble |
| WeatherNext 2 | 28km | Global | 15 days | 4x daily | AI ensemble (DeepMind) |
| ECMWF IFS | 40km | Global | 10 days | 4x daily | Wind, Europe |
| DWD ICON | 13km | Global | 7-10 days | 4x daily | Alternative global |
| NBM | 2.5km | CONUS | 7 days | Hourly | Multi-model blend |
US Models (NOAA)
NOAA NWS (National Weather Service) — RECOMMENDED
NOAA meteorologists review and edit forecasts from the National Digital Forecast Database (NDFD), combining multiple models with human judgment. Covers CONUS, Alaska, Hawaii, Puerto Rico, and US territories. Most reliable forecast source for US locations.
HRRR (High-Resolution Rapid Refresh)
NOAA's highest-resolution operational model (3km grid spacing). Runs every hour for CONUS and every 3 hours for Alaska. Can resolve individual thunderstorms. Generates both weather forecasts (48 hours) and simulated radar forecasts (18 hours). Used for precise precipitation timing.
RAP (Rapid Refresh)
Parent model of HRRR with 32km resolution covering all of North America. Runs hourly. Extends rapid-refresh forecasting to Canada, Mexico, and offshore areas not covered by HRRR.
GEFS (Global Ensemble Forecast System)
NOAA's 31-member ensemble system. Runs once daily at 00Z for the full 35-day forecast (shorter runs available 4x daily). Resolution is 0.25° (Days 1-10) and 0.5° (Days 10-35). Wide spread between min/max indicates low confidence; narrow spread indicates high confidence. Best for long-range trends and uncertainty estimation.
ECMWF ENS (European Centre Ensemble System)
ECMWF's 51-member ensemble system (1 control + 50 perturbed members). Runs once daily at 00Z with 0.25° resolution (~20km). Forecasts extend to 15 days (360 hours) with 3-hourly resolution (0-144h) then 6-hourly (144-360h). Considered the world's leading medium-range ensemble system, particularly strong for European weather and storm tracking.
AI Weather Models
WeatherNext 2 (Google DeepMind) — NEW
What is WeatherNext 2? WeatherNext 2 is Google DeepMind and Google Research's most advanced AI weather forecasting system. It uses a Functional Network Generative model to produce 64-member ensemble forecasts, providing both accurate predictions and uncertainty estimates for medium-range weather.
How it Works: Unlike traditional numerical weather prediction (NWP) models that solve complex physical equations, WeatherNext 2 uses machine learning trained on decades of reanalysis data. The AI learns atmospheric dynamics directly from historical weather patterns, running 8x faster than previous AI weather models.
Technical Specifications:
- Resolution: 0.25° × 0.25° (~28 km / 27,830 m), 721 × 1440 global grid
- Ensemble Members: 64 probabilistic samples for uncertainty quantification
- Forecast Range: 15 days (360 hours) at 6-hourly intervals
- Update Frequency: 4x daily (00Z, 06Z, 12Z, 18Z UTC)
- Data Availability: ~6-7 hours after initialization time
- Variables: 60+ including 2m temperature, 10m wind (U/V), mean sea level pressure, precipitation, specific humidity
- Historic Data: Available from 2022-present
Key Advantages:
- Speed: 8x faster than previous AI weather models
- Accuracy: Outperforms ECMWF ENS on 97.4% of 1,320 evaluated targets
- Extreme Events: Better prediction of tropical cyclones, severe weather, wind power
- Probabilistic: 64-member ensemble provides reliable uncertainty estimates
Limitations: This is experimental data for modelling purposes only, not validated for operational use. Best used in combination with traditional models for complete weather analysis. Cloud cover and humidity must be estimated from other variables.
Data Access: Available via Google Cloud Storage (gs://weathernext), Earth Engine, BigQuery, and Vertex AI.
Citation: © 2025 DeepMind Technologies Limited. Historic data licensed under CC BY 4.0.
WeatherNext 2 Info | Earth Engine Catalog | Developer Documentation
Global Models
GFS 16-Day (Global Forecast System)
NOAA's primary global model with 25km resolution. Runs four times daily. Provides worldwide coverage out to 16 days.
ECMWF (European Centre for Medium-Range Weather Forecasts)
European global model with 40km resolution (open data version). Runs four times daily. Often outperforms other models for 3-10 day forecasts, with strong accuracy for European weather. Available on Tree60 for wind visualization.
DWD ICON (German Weather Service)
German global model with 13km resolution using an icosahedral grid (triangular cells instead of latitude-longitude squares). Runs four times daily. An alternative to GFS and ECMWF with good performance for Europe.
NBM (National Blend of Models) — EXPERIMENTAL
NOAA's statistically calibrated blend of multiple models including GFS, NAM, HRRR, RAP, and Canadian models. Uses machine learning to combine forecasts. Represents NOAA's next-generation forecast approach.
Radar Systems
Weather radar detects precipitation by sending radio waves and measuring reflections. The site displays both real-time radar observations and model-generated radar forecasts.
Real-Time Radar
MRMS (Multi-Radar Multi-Sensor) — RECOMMENDED
Seamless composite of all NEXRAD radars with automatic quality control. Removes ground clutter and fills gaps between individual radar sites. 1km resolution covering CONUS and Alaska. Updates every 2-10 minutes.
NEXRAD Single-Site Radar
Individual WSR-88D Doppler radar sites. Provides 0.25-1km resolution near the radar site. Updates every 4-6 minutes. Most detailed view for areas close to the radar.
Radar Forecasts
| Feature | Real-Time Radar | Radar Forecast |
|---|---|---|
| Data Source | NEXRAD observations | HRRR/RAP model predictions |
| Updates | 2-10 minutes | Hourly |
| Time Range | Current + 18hr history | Next 18 hours future |
| Accuracy | Highly accurate (actual observations) | Estimates (may differ from reality) |
| Best For | Current conditions, tracking storms | Planning, storm arrival timing |
Coverage & Limitations
Radar beams travel in straight lines, but Earth is curved. This creates challenges:
- Beam height increases with distance: 100 miles from radar, beam samples 10,000+ feet above ground
- Terrain blocking: Mountains create "radar shadows"
- Range limits: Most radars effectively see precipitation within 150 miles
- Low-level gaps: Between radar sites, low-level precipitation may be missed
Radar Color Scale
- Light blue/green (15-30 dBZ): Light rain, drizzle, or light snow
- Yellow (30-40 dBZ): Moderate rain
- Orange (40-50 dBZ): Heavy rain
- Red (50-60 dBZ): Very heavy rain or small hail
- Pink/Purple (60+ dBZ): Extreme precipitation, large hail, severe weather
Snow Tracking
Snow data comes from NOHRSC (National Operational Hydrologic Remote Sensing Center), which collects observations from hundreds of stations across North America.
Measurement Types
| Type | What It Measures | Use |
|---|---|---|
| Snowfall (24hr) | Depth of new snow in last 24 hours | Storm intensity, recent accumulation |
| Snow Depth | Total depth of snow on ground | Snowpack conditions |
| SWE (Snow Water Equivalent) | Water content if snow melted | Water supply forecasting, flood prediction |
Snow Density
The relationship between snow depth and water content varies:
- Heavy, wet snow: 5:1 to 8:1 ratio
- Average snow: 10:1 ratio (10 inches snow = 1 inch water)
- Light, fluffy snow: 15:1 to 20:1 ratio
- Very cold powder: 30:1 or higher (Rocky Mountains, Alaska)
Snow-to-Liquid Ratio Algorithms
When forecasting snowfall amounts, the system must convert liquid precipitation (from weather models) into snow depth. This conversion uses a snow-to-liquid ratio (SLR). Tree60 Weather offers 7 different algorithms you can select in Settings > Snowfall Calculation.
Snow density varies dramatically based on temperature, humidity, wind, and crystal type. A single fixed ratio (like 10:1) works on average but can be wrong by 50-300% for individual storms. Temperature-based algorithms provide more accurate snowfall predictions by accounting for these physical processes.
Available Algorithms
1. Standard 10:1 (Default)
Type: Fixed ratio
Formula: 10 inches of snow = 1 inch of liquid
Best for: General purpose, matches NOAA default assumptions, consistent with historical climatology.
When to use: When you want simple, traditional snowfall estimates or don't know specific conditions.
2. Kuchera (NWS Temperature-Based)
Type: Temperature-dependent algorithm
Used by: National Weather Service operationally
How it works: Uses maximum temperature in atmospheric column below 500mb pressure level. Warmer temperatures = wetter/heavier snow with lower ratios. Colder temperatures = lighter/fluffier snow with higher ratios.
Ratio range:
- ≥34°F: 5:1 (very wet snow, near rain-snow line)
- 28-33°F: 10:1 (wet snow)
- 15-27°F: 15:1 (average snow)
- 5-14°F: 20:1 (light snow)
- <5°F: 25:1 (very light/fluffy snow)
Best for: Variable temperature conditions, coastal storms transitioning from rain to snow, general-purpose accuracy improvement over fixed 10:1.
Research: Developed through linear regression on thousands of snow depth and liquid equivalent observations.
3. Cobb-Waldstreicher (Dendritic Growth Zone)
Type: Physics-based algorithm using Gaussian distribution
Used by: NOAA MDL for operational SLR calculations
How it works: Accounts for the dendritic growth zone where snowflakes form large, branching crystals. Maximum snow production occurs at -12°C to -18°C (10°F to 0°F) where dendrites grow most efficiently. Uses Gaussian function centered at optimal temperature (-15°C / 5°F).
Ratio range: 5:1 to 25:1 based on temperature deviation from dendritic zone
Peak efficiency: Approximately 20:1 at -15°C (5°F)
Best for: Scientific accuracy, mountain/high-elevation forecasts, situations where atmospheric temperature profile is well-known.
Research: Cobb and Waldstreicher (2005) applied Gaussian relationship between SLR and temperature in regions of inferred snow growth.
4. Byun (Temperature + Precipitation Rate)
Type: Dual-parameter physics-based algorithm
Developed by: Byun et al. (2008) for numerical snowfall prediction
How it works: Considers BOTH temperature AND precipitation rate. Key insight: heavier precipitation rates produce denser snow due to riming (ice coating on crystals) and faster crystal growth under supersaturated conditions. Formula uses exponential temperature decay plus precipitation rate adjustment.
Key physics:
- Temperature effect: Colder temperatures = lighter snow (exponential relationship)
- Precipitation rate effect: Higher rates = denser snow (crystals grow faster, more riming)
- Light snow (0.5 mm/hr): Produces higher ratios (fluffier)
- Heavy snow (5+ mm/hr): Produces lower ratios (denser, wetter)
Ratio range: 4:1 to 30:1 based on combined temperature and precipitation rate
Best for: Variable intensity snowfall, convective snow showers, situations where precipitation rate is known to vary significantly (e.g., lake effect snow, orographic enhancement).
Advantage over temperature-only: Accounts for microphysical processes that temperature alone cannot capture. More physically complete than Kuchera.
Research: Byun et al. (2008) "A Snow-Ratio Equation and Its Application to Numerical Snowfall Prediction"
Algorithm Selection Guide
| Situation | Recommended Algorithm | Why? |
|---|---|---|
| General use / Don't know conditions | Standard 10:1 | Safe default, matches NOAA climatology |
| Variable temperature (warm to cold) | Kuchera | Most accurate for temperature-driven density changes |
| Very cold conditions (0-20°F) | Cobb-Waldstreicher | Accounts for dendritic growth zone physics |
| Variable intensity snowfall | Byun | Accounts for both temperature AND precipitation rate |
| Heavy snowfall rates (>5 mm/hr) | Byun | Captures riming effects that make snow denser |
| Lake effect snow | Byun | Highly variable intensity needs rate-dependent algorithm |
| Mountain snow / orographic | Cobb-Waldstreicher or Byun | Physics-based for complex terrain conditions |
| Coastal/transition events | Kuchera | Handles warm/cold temperature transitions well |
Why Not Fixed Ratios?
Earlier versions of this system included fixed ratios (5:1 heavy, 12:1 average, 15:1 light, 20:1 powder). These were removed because:
- No physics: Fixed ratios don't account for actual atmospheric conditions
- User uncertainty: Difficult to know which ratio to select without meteorological expertise
- Better alternatives exist: Temperature-based (Kuchera) and dual-parameter (Byun) algorithms automatically select appropriate ratios based on conditions
- Misleading precision: Choosing "15:1" implies you know snow will be exactly that density, when reality varies
Additional Algorithms (Not Currently Implemented)
Research literature describes several other SLR algorithms that were considered but not implemented:
- Roebber Neural Network: Uses artificial neural network trained on eastern US data. Research shows high bias and large prediction errors (MAE=9.45) in western mountains. Not implemented due to poor performance.
- MaxTaloft: Operational method used by some NWS offices. Studies indicate it overpredicts SLR with high bias (MAE=6.51). Not implemented due to systematic overprediction.
- Thickness Method: Based on 1000-500mb or 1000-850mb atmospheric thickness. Requires sounding data not readily available in forecast models. Not implemented due to data requirements.
Note: Recent research (Veals et al. 2025) compared these methods and found that simpler physics-based approaches (like Kuchera, Cobb-Waldstreicher, and Byun) often outperform more complex algorithms, especially in mountainous terrain.
How to Change Your Snow Ratio Algorithm
- Open Settings (gear icon, top right)
- Scroll to Snowfall Calculation section
- Select your preferred algorithm from dropdown
- Forecast text will automatically update to use new ratio
- Selection is saved and persists across visits
Data Sources
- SNOTEL stations: 800+ automated sites in western mountains
- Cooperative observers: 10,000+ trained volunteers
- Automated weather stations: State DOTs, utilities, ski areas
- NWS offices: Airports and forecast offices
Data Fusion Algorithm
When multiple reports exist for a single station, our system intelligently merges them:
Some stations report to multiple networks. The fusion algorithm prevents duplicate markers on the map.
Data Accuracy & Limitations
- Station density: High in mountains, sparse in valleys/plains
- Observation timing: Stations report at different times (not all real-time)
- Wind effects: Drifting can make readings unrepresentative
- Settling: Snow compacts naturally, reducing depth without melting
- Coverage gaps: Remote areas may have no nearby stations
How-To Guides
- Click a location on the map to view forecast
- Open Settings (top right)
- Under "Weather Source," select a different model (GEFS, GFS, HRRR)
- Compare temperature curves, precipitation timing, other elements
- Agreement between models = higher confidence
- Disagreement = lower confidence, uncertain forecast
- Open Settings (top right)
- Enable "Show Radar Forecast Timeline"
- In layers panel (left), enable HRRR or RAP radar
- Timeline slider appears at bottom of map
- Drag slider or click play (▶) to see simulated future radar
- Watch precipitation evolve over next 18 hours
Remember: Radar forecast is a model prediction, not guarantee. Reality often differs by 1-2 hours in timing.
- Select GEFS as weather source in Settings
- Click location to view forecast chart
- Look at temperature chart: mean line with min/max shading
- Narrow spread (min/max close to mean) = high confidence
- Wide spread (min/max far from mean) = low confidence
- Use mean for planning, prepare for anywhere in min/max range
- Enable "Snow Reports" layer in left panel
- Click snow station marker to see observations
- Note observation time (e.g., "7:00 AM local")
- Bookmark coordinates of stations to track
- Return daily at same time for updated values
- Compare snowfall (24hr), depth, and SWE trends
Note: Snow depth can decrease from settling even when SWE stays constant. SWE is true measure of water content.
- Open Settings (top right)
- Enable "Show Forecast Grid"
- Click anywhere on map to view forecast
- Green box shows model's grid cell
- All locations within box receive identical values
- Switch weather sources to see how grid size changes
Frequently Asked Questions
For storm timing: HRRR model or radar forecast
For medium-range (5-15 days): ECMWF ENS ensemble (51 members, highest skill)
For long-range (15-35 days): GEFS ensemble (35 days, trend guidance only)
For international: GFS 16-Day, ECMWF ENS, or DWD ICON
For comparing: Toggle between models to gauge confidence
• Light blue/green (15-30): Light rain, drizzle, snow
• Yellow (30-40): Moderate rain
• Orange (40-50): Heavy rain
• Red (50-60): Very heavy rain or small hail
• Pink/Purple (60+): Extreme precipitation, large hail, severe weather
HRRR: Hourly (CONUS), every 3 hours (Alaska)
GFS, DWD, ECMWF: 4x daily (00Z, 06Z, 12Z, 18Z)
GEFS 35-Day: Once daily (00Z only for full 35-day range)
ECMWF ENS: Once daily (00Z)
NWS: Hourly with major updates 4x daily