How puhub computes the weather — methodology
puhub.ee shows Estonia's weather in real time, combining official observations with a numerical weather prediction background and its own 1 km analysis.
Observations and quality control
Inputs are the weather stations of the Estonian Environment Agency, the Finnish (FMI) and Swedish (SMHI) meteorological institutes, and Estonian road weather stations. The precipitation field comes from the Environment Agency's Harku and Sürgavere radar composite, with the European OPERA network as fallback.
Every observation passes spatial quality control before entering the analysis: the value is checked against physical limits, against neighbouring stations (buddy check) and by a spatial consistency test. The scheme follows the TITAN approach [1] used in Nordic operational services.
Analysis: optimal interpolation
The background field is MEPS, the Nordic convection-permitting model on a 2.5 km grid [2]; ICON-EU from the German weather service when MEPS is unavailable.
Observations are combined with the background by optimal interpolation (OI), a statistical interpolation method weighting the error variances of observation and background together with horizontal and vertical correlation scales [3]. Far from stations the model structure is retained; near a station the measured value dominates. The Norwegian meteorological institute uses the same method.
Precipitation: radar nowcasting
Precipitation is not interpolated between stations. The near-term forecast is obtained by semi-Lagrangian advection of radar echoes along a detected motion field — the Lagrangian persistence assumption, whose scale dependence of predictability has been quantified [4]: large structures stay predictable for hours, convective cells for minutes.
Later hours come from MEPS, statistically downscaled from the 2.5 km grid to the 1 km analysis grid; the relationship is calibrated against radar observations.
Surface effects on wind
The wind field is corrected at 100 m resolution. Forest height and cover give an aerodynamic roughness length z₀, from which the logarithmic wind profile yields the speed reduction — up to roughly 40% in dense tall forest. Terrain effects are not modelled separately.
Climate normals and uncertainty
Climate normals are computed from the ERA5 reanalysis over 1994–2023 [5].
Seven-day forecast uncertainty comes from ECMWF's 51 parallel model runs started from slightly different initial conditions. The p10–p90 range shown is their spread: a narrow range means the runs agree, a wide one that they do not. Candidate forecasts are scored continuously, per lead time, against measurements from 24 weather stations.
References
- Båserud, L., Lussana, C., Nipen, T. N., Seierstad, I. A., Oram, L., Aspelien, T. (2020). TITAN automatic spatial quality control of meteorological in-situ observations. Advances in Science and Research 17, 153–163. doi:10.5194/asr-17-153-2020
- Müller, M., Homleid, M., Ivarsson, K.-I., et al. (2017). AROME-MetCoOp: A Nordic Convective-Scale Operational Weather Prediction Model. Weather and Forecasting 32(2), 609–627. doi:10.1175/WAF-D-16-0099.1
- Nipen, T. N., Seierstad, I. A., Lussana, C., Kristiansen, J., Hov, Ø. (2020). Adopting Citizen Observations in Operational Weather Prediction. Bulletin of the American Meteorological Society 101(1). doi:10.1175/BAMS-D-18-0237.1
- Germann, U., Zawadzki, I. (2002). Scale-Dependence of the Predictability of Precipitation from Continental Radar Images. Part I: Description of the Methodology. Monthly Weather Review 130(12), 2859–2873.
- Hersbach, H., Bell, B., Berrisford, P., et al. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society 146(730), 1999–2049. doi:10.1002/qj.3803