1. Overview

The Weather Project is an automated forecasting and weather-analysis system. Most products begin with gridded model data, which are then subset to Australian regions, converted into consistent units, adjusted for local terrain where useful, and transformed into maps, meteograms, forecast tables and derived diagnostics.

The aim is not to replace official forecasts or warnings. The site is designed to make model guidance easier to interpret, compare and explore, particularly in situations where local elevation, snow level, frost, wind or rainfall gradients matter.

Important: The Weather Project is model guidance, not an official warning service. For severe weather warnings, fire weather warnings and other official advice, refer to the Bureau of Meteorology and relevant emergency agencies.

2. Forecast model data

GFS

The main forecast source is the NOAA Global Forecast System (GFS), using approximately 0.25° global model data. Forecast fields are downloaded for relevant Australian domains and processed through the site's own scripts.

ECMWF AIFS

Selected products also use ECMWF's Artificial Intelligence Forecasting System (AIFS). Where AIFS appears beside GFS, the purpose is model comparison rather than treating either model as automatically superior.

Common fields used

2 m air temperature and dew point
10 m wind components and wind gust
mean sea-level pressure
precipitation rate and accumulated precipitation
850, 700 and 500 hPa temperature, humidity, geopotential height and wind
instability fields such as CAPE and lifted index

Availability varies by model and product. Some diagnostics use only fields present in the underlying model archive.

3. Multi-run processing and forecast consistency

Many town products are built from several consecutive GFS runs rather than a single forecast cycle. This allows recent model guidance to be compared through time and reduces the visual impact of one unusually noisy run.

A typical multi-run product uses the latest seven model runs, spaced 12 hours apart. The forecasts are aligned by valid time and can be displayed individually or combined into low, middle and high groups.

Example 7-run set: latest run, then -12 h, -24 h, -36 h, -48 h, -60 h, -72 h

Where the site displays a "stitched" daily forecast table, the most recent portions of successive runs are combined into a continuous forecast series. The purpose is to retain recent guidance at short range while still providing a long forecast horizon.

4. Terrain adjustment

Global models cannot resolve every Australian mountain, valley, plateau or coastal slope. The Weather Project therefore applies terrain-aware post-processing to some surface fields using a higher-resolution digital elevation model.

Temperature

Model 2 m temperature can be adjusted according to the difference between model terrain height and higher-resolution terrain height. A standard environmental lapse rate is used as a first-order correction.

Temperature adjustment ≈ elevation difference × 0.0065 °C per metre

This is a practical downscaling approximation, not a full local atmospheric model. Temperature inversions, cold-air drainage and local land-surface effects may still differ substantially from the adjusted value.

Wind and rainfall

Selected terrain-aware products also modify wind and precipitation according to whether terrain is exposed to or sheltered from the model-flow direction. Windward slopes can receive an enhancement while lee slopes can be reduced. The adjustment is smoothed spatially to avoid unrealistic pixel-scale jumps.

5. Rainfall processing

Model precipitation rate is converted into rainfall depth over each forecast step. For a three-hour forecast interval:

3-hour rainfall (mm) = precipitation rate (kg m⁻² s⁻¹) × 10,800 seconds

Because 1 kg m⁻² of liquid water is approximately 1 mm of rainfall, the conversion produces rainfall depth directly.

Some rainfall products apply terrain redistribution to better represent broad windward enhancement and lee-side reduction. This should be interpreted as downscaled model guidance, not as a replacement for radar, gauges or a convection-resolving numerical model.

Daily rainfall tables

Town daily tables use a Bureau of Meteorology-style 9 am to 9 am local-day convention. Rainfall is allocated across the relevant local 9 am boundaries so that the forecast table is easier to compare with conventional Australian daily rainfall observations.

6. Snow methodology

Snow products are intentionally separated into several different concepts. A map showing cold-enough conditions is not the same thing as a map showing where snow is forecast to fall, and neither is identical to accumulated snow depth.

Product	What it represents
Snow settling probability	Probability that the thermal profile is sufficiently cold for snow to settle. It is not the probability that precipitation will occur.
Snow settling level	Estimated elevation above which conditions are cold enough for falling snow to settle.
Terrain snow cover	The settling level intersected with terrain elevation, showing the areas that are actually high enough for settling snow.
Cumulative snow depth	Modelled accumulated snow at a specified fixed elevation, derived from precipitation classified as snow.
Cold outbreak overlap	Areas where selected upper-air temperature and thickness criteria overlap, highlighting broad cold-air environments supportive of unusually low snow levels or wintry conditions.

Snow / rain phase

The snow-processing system uses a settling probability rather than switching phase from snow to rain at a single hard threshold. To reduce rapid flip-flopping between categories, hysteresis is applied:

Snow if probability ≥ 55% · Rain if probability ≤ 45% · 45–55% retains the previous phase

Snow accumulation is then based on the portion of model precipitation classified as snow. Fixed-elevation products are also masked where the surrounding terrain does not reach the nominated elevation.

7. Derived weather diagnostics

Several products are calculated from the raw model fields rather than downloaded as ready-made variables.

Wind speed

Wind speed = √(u² + v²)

Calculated from the model's east-west and north-south wind components.

Moisture and comfort

Relative humidity, wet-bulb temperature, apparent temperature and heat index products are calculated or post-processed from model temperature, moisture and wind fields.

Frost risk

Frost guidance combines forecast near-surface temperature and related moisture conditions into a probability-style risk field. It is a model diagnostic rather than a direct prediction of frost at every exposed surface.

Thunderstorm environment

Thunderstorm products combine instability and related atmospheric fields to summarise potential coverage, severity and composite storm-supportive conditions.

Cold outbreak maps

Cold outbreak maps identify overlap between selected 850 hPa temperature, 500 hPa temperature and 1000–500 hPa thickness thresholds. They are intended as a diagnostic of the broader cold-air mass, rather than a direct snowfall forecast.

8. Observations

Live observation pages use recent Bureau of Meteorology station observations. Common fields include air temperature, dew point, relative humidity, wind speed, wind gust, QNH pressure and cumulative rainfall trace.

The site derives recent rainfall increments from the cumulative rain trace while accounting for gauge resets. Observation pages can also generate short plain-English summaries of current conditions and recent trends.

Observation displays are intended to complement, not replace, the Bureau of Meteorology's official observations and quality-control systems.

9. Town forecasts and daily tables

Town forecasts are extracted from the gridded model at each site's latitude and longitude. Where processed gridded datasets are available, values can be interpolated to the exact town location rather than simply taking the nearest raw model grid point.

Daily forecast tables summarise the continuous forecast into BoM-style local days. Depending on the parameter, the script uses either the period from the current 9 am to the next 9 am, or the previous 9 am to the current 9 am, so minimum temperature and rainfall align more naturally with Australian daily climate conventions.

The tables can include:

minimum and maximum temperature
24-hour rainfall
maximum sustained wind and gust
maximum wet-bulb temperature
maximum heat index and apparent temperature
maximum frost risk
daily chill-hour accumulation

Text summaries shown on town pages are generated deterministically from these forecast values. They are designed to describe the data already produced by the model-processing system, rather than introduce a separate forecast source.

10. Update frequency

Most forecast products are generated from the 00Z and 12Z model cycles and are refreshed when the required model data become available. Observation products are updated more frequently as new station observations arrive.

Individual pages may retain the previous successful output if a source file is temporarily unavailable. The model run or valid time shown on each chart should therefore be used to confirm the age of the guidance.

11. Limitations and interpretation

Global models have finite resolution and cannot explicitly resolve every local terrain feature, sea breeze, valley inversion or thunderstorm.
Terrain adjustment improves spatial realism in many situations but remains an empirical post-processing method.
Rainfall can vary sharply over short distances, especially in convective weather and complex terrain.
Snow depth depends on precipitation phase, snow density, melting, compaction and local exposure. A simple model accumulation cannot capture all of these processes.
Forecast uncertainty generally increases with lead time. Agreement between several consecutive model runs is usually more informative than a single long-range run.
Observation data may contain missing values, delayed reports or quality-control issues originating in the source feed.

The Weather Project should therefore be used as an additional source of model guidance. Important weather-sensitive decisions should also consider official forecasts, warnings, radar, observations and local knowledge.

12. Data attribution

Forecast and observation products on The Weather Project are derived from publicly available meteorological datasets and automated post-processing. Key sources include NOAA/NCEP GFS data, ECMWF AIFS where shown, and Bureau of Meteorology observations.

Product descriptions on individual pages identify the type of guidance being shown. The Weather Project's terrain adjustment, forecast stitching, snow diagnostics, visualisation and derived calculations are additional processing applied by the site.