The benchmark suite

The bar shows the latest release vs the previous one across 34 cases — 25 energy-balance (5 fluxes × full + 4 seasons, within 0.05 W m⁻²) and 9 air-temperature (3 RSL heights × full + 2 windows, within 0.05 K): better / same / worse. Basel, Berlin and IMU are sites esmae has presets for — not yet benchmarked here.

• Each SUEWS release is run on the same site configuration, meteorological forcing and observations, over a fixed evaluation period — only the model version changes, so any difference is attributable to the model.
• Each release runs a configuration valid in its own schema; performance is the mean absolute error (MAE) and mean bias (MBE) of the modelled energy-balance fluxes (K_up, L_up, Q_N, Q_H, Q_E) and the RSL air-temperature profile against the tower observations.
• A release is compared with the previous one to flag regressions; a site is flagged if error grows materially.
• Every run is reproducible: re-running a release yields a byte-identical statistics fingerprint.
• Observations are access-restricted and fetched at run time from a protected archive; only the derived statistics shown here are published.