MS Ozone Forecast Dashboard

--

Tomorrow's RF Forecast

--

Tomorrow's AQM Bias-Corr

--

Tomorrow's Forecasted High Temp

Forecast Comparison (Today)

Forecast Comparison (Tomorrow)

7-Day Ozone Trend

Model Diagnostics — Variable Importance

Model Control

Updating the model uses the latest downloaded historical records.

Comparing RF against AQM: most RF rows in the history log were reconstructed after the fact using the observed next-day weather, while the NOAA AQM values are genuine forecasts. Perfect-prognosis rows score much better than the model can achieve live, so the RF column flatters itself in any pooled comparison. The Type column in the Forecast History Log marks which rows are operational (real forecasts) versus hindcast; weigh the operational rows when judging live skill.

Performance

Performance Analysis

Forecast History Log

Score which forecasts?

Overall Performance

Moderate+ Days (≥ 55ppb)

USG+ Days (≥ 71ppb)

Reading the threshold tables: RMSE, Bias, NMB, NME and R² are computed only on days where the observed value cleared the threshold. Conditioning on the observation narrows the remaining spread, so R² collapses toward 0 and bias turns negative even for a skillful model — a known artifact of conditional verification, not proof the model failed. Judge event performance on POD / FAR / CSI (computed on the full record) and use Overall for correlation.

NMB / NME are the EPA photochemical benchmarks (ozone: NMB within ±15%, NME under 25%).

Raw Data Preview (Last 5 Rows)

Data Pipeline

Data is synced via the R pipeline (GitHub Actions or manual Rscript run). The export script generates the JSON files this dashboard reads.

To refresh data:

Run Ozone_Master_Sync.R (updates data + models + forecasts)
Run web-dashboard/export_json.R (exports JSON for this dashboard)
Push or deploy