What It Solves
A clear local weather prediction system
This platform turns public weather observations into daily forecasts for Minneapolis-St. Paul, then shows whether the predictions were right as actual weather arrives.
Weather Intelligence Platform
Minneapolis-St. Paul
Predicting temperature and precipitation from public weather dataPrediction
Temperature and precipitation
The live system predicts seven daily high temperatures, estimates the chance of precipitation, and displays forecast precipitation amount as guidance.
Accuracy
Measured over time
Temperature forecasts are verified as actual days complete. Precipitation chance is evaluated separately because occurrence and amount behave differently.
System
AWS data pipeline
Lambda jobs collect weather data, S3 stores prediction-ready JSON, and the Amplify dashboard displays the latest model output.
Platform Snapshot
What visitors can inspect
Temperature7-day high-temperature ML forecast
Precipitation chance7-day forecast-correction model
Precipitation amountForecast QPF baseline
DataNOAA history + NWS observations and forecast guidance
DeploymentAWS Lambda, S3, API, and Amplify
StatusLive temperature plus precipitation guidance
Tomorrow High
--F
Loading prediction data
Data Freshness
--days
Waiting for source metadata
Model MAE
--F
Max temp validation score
7 Day Outlook
Next 7 predicted highs
Dots show the model forecast; shaded bands show the typical error range for each day.
Confidence
Forecast horizon
Verification
Day-ahead predicted vs actual highs
Waiting for verified actuals
Tomorrow Chance
--%
Loading precipitation chance data
Forecast Amount
--in
Loading forecast precipitation amount
Forecast Inputs
QPF
Forecast precipitation amount is used as the conservative amount baseline.
Model Shape
2stage
ML estimates precipitation chance first; amount correction is being evaluated separately.
Precipitation Outlook
Predicted chance and forecast amount
The line is Jim's ML model predicting the chance of measurable precipitation. The bars are the forecast precipitation amount for each day, not actual observed rainfall.
Training Upgrade
Precipitation model status
Input forecastForecast QPF
TargetActual daily precip
Chance modelActive in dashboard
Amount displayQPF baseline
Amount modelExperimental
Forecast horizons1-7 days
Pipeline Status
Precipitation data foundation
1Archive forecast snapshotsNWS grid data is stored in S3 for future training rows
2Flatten forecast featuresProbability, QPF, humidity, wind, and condition fields become model columns
3Join with actual precipForecast issue dates can be matched to observed daily precipitation
4Train 7-day chance modelChance predictions are now available for the precipitation view
Data Pipeline
What data powers the forecast?
Source
Latest Date
Why It Matters
Status
Recent weather summaries
--
Feeds the temperature model and recent-observation features
Current
Temperature prediction
--
Displays the next seven daily high temperatures
Ready
Precipitation prediction
--
Displays ML chance plus forecast QPF amount
Waiting
Prediction verification
--
Scores temperature forecasts after actuals arrive
Waiting
NWS forecast snapshots
Collecting
Provides precipitation forecast guidance and model features
Collecting
NOAA historical archive
--
Long-term model training data
Archived
How It Works
Current operational path
Waiting for live pipeline metadata
Coverage
Operational forecast window
Temperature rows used--
Latest observation date--
Incomplete dates--
Future days filled--
Last forecast date--
Latest precip issue--
Generated--
Temperature Model
High-temperature forecast performance
Tomorrow horizon
--
Selected MAE
--
Horizon +1 R2
--
Horizon +1 MAE
--
Horizon +1 RMSE
--
Horizons
7 days
Trained
--
Verified records
--
Precipitation Model
Chance forecast status
Chance source--
Day +1 Brier score--
Amount display--
Forecast issue--
Horizons7 days
Chance model trained--
Temperature Verification
MAE by forecast horizon
Waiting for verification records
Roadmap
Where the Weather Intelligence Platform goes next
1Historical training setNOAA archive processed into daily model features
2Live max-temp forecastsDaily Lambda writes 7-day prediction JSON
3Verification loopForecast errors are scored as actual days complete
4Precipitation chance modelForecast guidance now feeds 7-day precip chance predictions
5Precip amount correctionImprove beyond the QPF baseline once amount verification is stronger
Next Prediction Targets
Making precipitation more useful
Chance of precipitationActive
Precipitation amountQPF baseline
Amount correction modelExperimental
Precip verificationNext
Forecast confidenceImproving
More locationsFuture