Sign inRegisterCheck connection

Forecast Versus Station Observation

A forecast is time-bounded model guidance, not a backyard reading. Why personal weather sites showed both, and how to keep the two families apart.

Back to Historical weather observations and forecasts

A weather forecast is a statement of expected conditions for a named place and a named future interval. It is produced by a forecast office or a numerical model. It is not the last sample from a backyard thermometer. Personal weather sites routinely published both on one dashboard. That pairing is useful only when each family keeps its own source, valid time, and authority.

This URL was a forecast slot on the historical TNET Weather station site. The original feed is not rehosted. The page now explains the product class that slot belonged to, so a recovered icon row is not mistaken for a live Mesa observation.

What a forecast is

The World Meteorological Organization treats a forecast as expected meteorological conditions for a specified time or period in the future. That definition is operational. It requires three fields before the adjectives matter.

Place. A zone, a grid cell, a city, an aerodrome, or a point. The forecast is valid for that geography, not for every sensor inside it.

Valid interval. Today, tonight, a 12-hour period, an hour, a lead time of T+6 hours. A high temperature in a “Tuesday” period is a claim about Tuesday, not about the moment the page loaded.

Producer and cycle. A National Weather Service Weather Forecast Office issues public text from a digital forecast database on a stated cycle. A global model has a run time. A local statistical model has an issue time. Without the producer and the cycle, the numbers are orphaned.

NWS public forecasts are created at each Weather Forecast Office on that office’s grid, at a resolution of about 2.5 km. The NWS API exposes period forecasts, hourly forecasts, and grid data as separate resources from station observations and from alerts. Forecast, observation, and warning remain different products even when they share a website. Republishing does not convert model guidance into a measurement.

What a station observation is

An observation is a value measured at a stated site at a stated time: temperature, humidity, pressure, rain, wind, solar radiation. WMO instrument guidance treats metadata as part of the measurement: identity, exposure, instrument, and time. A personal weather station is not a WMO observing site, but the same rule applies. A number without a clock is decoration.

The historical current-conditions page was the observation family on this site. It reported what the logger had already sampled. It could not say what Tuesday night would do. A disagreement between the latest temperature and an NWS afternoon high is expected: one is a sample, the other a period maximum for a grid or zone.

Dew point, heat index, and wind chill calculated from observations are not extra sensors. They inherit the observation’s time. They do not become forecasts because a template printed them under a sun icon.

Why a personal weather site showed both

Station software in the 2000s made it easy to publish a “now” block and a “forecast” block from the same template. Visitors arrived for the backyard. The operator wanted a one-page answer: what is it doing, and what is expected later.

That pairing is legitimate only as two evidence types. The forecast block was almost never computed from the roof sensor. Typical sources were an NWS zone or point forecast, a downloaded model text, or a local model such as WXSIM. Mixing them without labels taught readers that “the station” issued tomorrow’s weather.

A personal weather site is not an NWS office. TNET Weather is not an NWS office. Official U.S. public forecasts and watches, warnings, and advisories are issued by the National Weather Service. For live hazards, use weather.gov and local emergency authorities. TNET is not a warning service.

The pairing still has a scientific use. A labeled forecast and a labeled observation let a reader see whether the atmosphere is evolving as the last official cycle expected. That comparison is a verification habit. It is not a license to overwrite the forecast with the sensor, or the sensor with the icon.

Clocks that must not be mixed

Four times appear on a mixed forecast page. They are independent.

Observation time is when the sensor produced the sample.

Issuance time is when the forecast office or model released the product.

Valid time is the period the forecast describes.

Retrieval time is when the station script fetched a copy.

A page that prints only “updated 3:11 p.m.” does not say which of those four it means. If the script fetched an NWS product at 3:11 p.m., that is retrieval time. The NWS issuance may be earlier. Stale forecast text is still a forecast. The valid-time line has to show that.

Lead time is the gap between issuance and the start of the valid interval. A seven-day zone forecast is not a nowcast. Treating a Day-5 icon as if it had the authority of a current observation is a category error.

Forecast, derived, scenario, historical

Keep the families named.

Observed. A documented measurement at a site and time.

Forecast. A model or office value valid for a future interval.

Derived. A calculated summary, index, or score built from other inputs. It is not a thermometer and not an official forecast.

Scenario. A physics or planning calculation used to explore sensitivity, not a measured user outcome and not an NWS product.

Historical. A past snapshot. An undated copy of an old Mesa forecast is historical. It is not current Phoenix weather.

This page is evergreen. It does not reprint recovered period temperatures. Those numbers, if they existed on the old URL, were a snapshot of some issuance that is no longer in force.

How to read a mixed dashboard

Work the screen in a fixed order.

  1. Find the observation block. Record station, units, and observation time.
  2. Find the forecast block. Record producer, place, issuance, and valid periods.
  3. If a watch, warning, or advisory appears, leave the station site and open the official NWS product. Do not take the icon color as the warning.
  4. If the two blocks disagree, keep both. Note the valid interval of the forecast and the age of the observation. Do not average them.
  5. If the forecast source is missing, treat the icons as unlabeled model output. They are not the backyard.

A practical test: if you cannot say whether a printed “72°” is the sensor at 14:03 or the NWS period high, the page has failed. The same numeral can belong to either family.

Related station-site treatments of the same problem sit on the NWS Phoenix metro forecast product article, the republished NWS forecast article, and the station-site forecast composite article. Those pages cover product mechanics, copying, and mixed sources. This page is the definition.

What this page is not

It is not a live NWS forecast. For current U.S. public forecasts, use weather.gov or, for this geography, NWS Phoenix. It is not a Mesa nowcast. It is not TNET Connection Outlook, which is a derived estimate that interprets atmospheric and other evidence for a selected city, date, and interval—not a weather-office product and not a promised download speed. Local observation can disagree with a regional forecast; that disagreement is information only when both sides are labeled.

Modern relevance

Dashboards still collapse families: one card, one color, no source. The failure mode did not end with PHP templates. A phone widget that prints “partly cloudy, 84°” without saying whether that is observed now or forecast this afternoon repeats the 2000s error.

TNET’s public account of observed, modelled, and derived information is how the service works. That page is methodology for a connection-intelligence product. It does not issue weather forecasts and does not replace weather.gov. Historical forecast URLs on this host are collected under historical observations.

The useful outcome of /forecast is a habit: name the product class before you use the number.

Sources