New York City KLGA
Temperature & daily precipitation · America/New_York
Collection details
Automatic selection keeps the newest complete mean or deterministic forecast while a newer run fills. Temperature and precipitation may use different cycles, shown below. If no complete run covers this day, the best available partial run is shown. Individual member gaps remain unavailable; runs are never spliced together.
Current ensemble members: solid lines. Lagged members (−6h / −12h): dashed. Means remain thicker; NWS remains dashed gold. Five-minute readings depend on station publication; verification continues to use routine METAR and climate reports.
Temperature spaghetti KLGA · °F
White: hourly / SPECI · Blue: five-minute observationsLoading saved comparison…
Daily precipitation inches
KNYC · 1969-12-31 · 12:00 AM to 12:00 AM local · fixed standard-time day. The first interval is 12:00 AM–1:00 AM, plotted at 1:00 AM. The value ending at 12:00 AM belongs to the previous day. Accumulation starts at zero at the beginning of the first available interval. Partial subtotals sum available forecast hours only; they are not full-day predictions and do not include observed rain. Remaining totals start at the next full hour, excluding the partly elapsed hour. The chart covers this standard-time day only; accumulation resets for this day and starts at the first available interval. Internal gaps stop accumulation; missing hours are not treated as zero. The cards below remain totals for 1969-12-31, including in Max view.
Loading saved precipitation…
Member coverage & line controls
Thick solid lines are official mean products, drawn above thin member lines. Mean = arithmetic ensemble mean; PMM = probability-matched mean; LPMM = localized PMM; AVRG = NOAA mean/PMM combination. These are not performance-trained weights. Dashed member lines are older contributing runs, with their actual initialization shown above. REFS mean variants come from the corresponding NOAA products. PMM, LPMM and AVRG are precipitation-only. Their displayed daily totals sum the hourly products; they are not NOAA’s separately computed 24-hour PMM/LPMM/AVRG fields. Missing member hours stay unavailable; they are not filled with the mean.
Hourly scores include today so far; daily scores use completed days. Lead bins are [0,12), [12,24), [24,48), and 48+ hours, using exact initialization times. Each forecast case has equal weight.
| Model / member | Forecast cases | Valid hours | MAE (°F) ↓ | Bias (°F) | RMSE (°F) ↓ |
|---|
No verified cases in this selection yet. Scores appear after forecasts saved in advance have verified; full-day rain also waits for a completed climate report.
Uses the measure, period, lead range and sorting selected above. Created in your browser from saved verification statistics; no model downloads. Changing filters clears the preview.
MAE = mean absolute forecast error; Bias = mean(forecast − observed); RMSE = square root of mean squared error. Bias near zero does not necessarily imply small errors. Hourly temperature: routine METAR within 15 minutes of valid time (preliminary observations, subject to corrections). Daily high/low: maximum/minimum of 24 forecast hourly temperatures versus the same station’s completed CLI climate report; all forecast hours must be saved before the day starts. Hourly forecast extrema may miss between-hour extremes. No nearby station is substituted when CLI data are unavailable. Precipitation: 24 complete forecast hours saved before the standard-time day starts, verified against the station’s completed NWS climate report. Mean products are bold and scored separately from members; use Means first to group them ahead of the selected ranking. PMM/LPMM/AVRG daily scores verify the sum of hourly products. Positive departure means forecast too warm/wet; positive timing departure means late. Timing compares first/last wet hour (≥0.01″) in the standard-time day using 24 complete routine METAR hourly precipitation reports; only days with forecast AND observed rain qualify. This is an hourly approximation, not exact onset/cessation. NWS timing is excluded because its QPF may span multiple hours. Lead uses each member’s original initialization. Sample counts can differ; these are available-case scores, not a matched-case ranking. Rain occurrence threshold: 0.01″. Verified cases and daily statistics are archived from setup onward. Both CSV downloads are cumulative for the station shown above, with readable UTC/local timestamps, and refresh every 6 hours. Daily summaries add completed, eligible days as reports arrive; corrections replace existing cases. Blank summary cells mean unavailable, not zero. Forecasts must have been captured before verification; collection gaps cannot be reconstructed retrospectively.
Individual departures °F · forecast − observed
Uses the measure selected above. Latest 100 matching cases from the latest 200 saved cases for this measure (up to 7 days); CSV contains more. Daily timestamps identify the end of the verified day. Positive departure = forecast too warm or too wet.
| Valid time / day ended | Model / member | Forecast (°F) | Observed (°F) | Departure (°F) |
|---|
No qualifying verified cases yet. Daily metrics require forecasts saved before the whole day and a completed station climate report. Missing observations are never treated as zero.
Collection notices & methodology
NWS grid forecasts are fetched automatically and archived with their actual grid update time. The dashed gold NWS line appears alongside means. NWS multi-hour QPF is distributed uniformly into hourly bins; it does not identify exact rain timing. Historical cycle selection uses the newest saved NWS issue no later than that cycle. NOAA official REFS means, perturbations, control and HRRR; deterministic HRRR and RRFS use native NOAA initialization-pinned data on this page. Initial setup collects the current cycle and required lagged inputs. Subsequent cycles are collected automatically. Temperature and precipitation records preserve receipt time; past-valid downloads are never scored as forecasts. Ensemble members are correlated and their spread is not a calibrated probability. Hourly gaps are not estimated.