SkyPollen's grass pollen level is a modelled estimate, not an official count. It uses the Victorian Grass Pollen Emission Module (VGPEM, Emmerson et al. 2019), a published model from the Bureau of Meteorology, CSIRO and university partners, fed only with open data that may be republished. In Australia it is labelled not yet checked against local counts, because no openly licensed counts exist to check it with.
The model
Daily grass pollen = a season curve × the day's weather. The season curve is a Cauchy curve 19 days wide. Its peak day comes from when satellite greenness (EVI) over the pastures of south-west Victoria (37.3-38.3°S, 142.0-143.3°E) starts to fall, and its height from how green those pastures got in winter. Warm, dry, windy days raise the level; rain cuts it. In the original study the model reached a correlation of 0.69 with observed counts at eight Victorian sites in 2017, put 69 % of days in the right category and caught 69 % of High-or-above days. Its weather terms are published as curves; we read them off the paper's figure.
How the categories are set
The paper compares its model with counts only after scaling it to the observed seasonal mean. We do the same with two published figures: Melbourne's 2017 seasonal mean (31 grains per cubic metre) sets the size, and the share of High-or-above days in Melbourne (22 %, AusPollen 2016-2020) sets where "High" begins. Moderate and Extreme keep the Australian scale's proportions. Because the lines are calibrated, we show the category, not a count in grains.
How we checked it (2018-2025)
We ran the model for past seasons using only its inputs and compared it with published figures: the season starting between 20 October and 24 November (mean 6 November; Ong 1995), and High-or-above on about 22 % of days.
| Season | Modelled peak | First High day | High+ days, Oct-Dec |
|---|---|---|---|
| 2018 | 12 November | 19 October | 19 % |
| 2019 | 20 November | 23 October | 27 % |
| 2020 | 9 November | 3 November | 19 % |
| 2021 | 22 November | 2 November | 16 % |
| 2022 | 11 November | 5 November | 9 % |
| 2023 | 9 November | 18 October | 28 % |
| 2024 | 16 November | 22 October | 25 % |
| 2025 | 3 December | 24 November | 11 % |
First High day: mean 30 October, inside the published range in 6/8 seasons. High-or-above share in 2021-2025, which were not used to set the lines: 18 %. Past weather: NASA POWER daily data for Melbourne.
Limits
- The model's day-to-day swings are smaller than real counts: it will call a High day, but real peaks can be several times higher than the model suggests.
- Early in October it can show Moderate days a little before the real season starts.
- Grass is the main model. Cypress and she-oak are added as our own, unvalidated estimates: the typical week from a trap record (Brisbane 1994-99, Melbourne 2009-11) adjusted for the day's forecast weather. Other trees (plane tree, birch, privet, olive) are not yet included.
- Adelaide uses the same method with its own satellite greenness and has no local check at all.
- Brisbane, the Gold Coast and the Sunshine Coast use a different, subtropical grass model that is our own and unvalidated: no published model exists for those grasses. It follows grazing-land greenness about five weeks earlier, the rain of two to four weeks earlier and the day's weather, sized from published Brisbane statistics (details and past seasons).
- Thunderstorm asthma is not forecast here. Use the official Victorian forecast. City pages add a note when ECMWF storm energy (CAPE 500 J/kg or more) and rain meet high grass pollen (how the note works).
Wind direction check
The map's wind was compared with observed airport wind at Melbourne and Avalon: the median difference in direction was 15°.
Data and licences
Every source and its licence is listed at the foot of each page. Weather: ECMWF open data (CC BY 4.0). Mixing height: NOAA GFS. Greenness: NOAA VIIRS and NASA MODIS. Land use: ABARES (CC BY 4.0). Vegetation: NVIS (CC BY 4.0). Place names: GeoNames (CC BY 4.0). Basemap: OpenFreeMap / OpenStreetMap.