–
Temperature — the model's 9 km grid sees the islands as sea, so its 2 m temperature barely follows the day; the correction puts the island heating back
Rain (3 h, bars) and chance of rain (line)
Wind speed (m/s)
Relative humidity (%)
Why: the corrections
Diurnal: the learner's day/night-cycle part of the temperature correction (sunshine × clear sky, calm, clear nights). Climate: the slow learner's MJO / ENSO / IOD / season / outlook adjustment. Cloud: satellite at first, then the model's humidity and rain.
Drivers at this island
All steps
JSONBy time of day (Maldives time) — mean error
By time of day — mean absolute error
Learning over time — mean absolute error per day, kept since the service started
Blend weights — re-learned daily from out-of-sample errors
raw: no correction · approved: forecasters' correction (validation) · kf: Kalman-filtered bias per station and time of day · ml: the learner · ml_kf: the learner with its own recent errors tracked · qm: rain quantile mapping
Rain: yes/no scores and probability
Daily maximum and minimum temperature
The raw daily max comes from the 3-hourly steps, which also miss the afternoon peak. The corrected max blends the corrected steps with a learner trained on the observed daily extremes.
By station
What the learner makes of each feature
Change in the correction for one standard deviation more of the feature, with all else equal. Refitted for every issue, with recent days and days of similar weather counting most.
Climate drivers (slow learner)
Set aside by quality control
Stations whose data disagree with all the others far more than siting explains (per time of day), sensors reading outside the believable range, or rain gauges that stay dry while others get rain. Their islands use the all-station correction until they recover.
Station effects — siting and island size, on top of the all-station correction
Settings
Now
MJO — local index (positive: convection enhanced over the Maldives) and amplitude
ENSO and IOD — Niño 3.4 anomaly and Dipole Mode Index (°C)
Regional seasonal outlooks
An outlook's rain and temperature tilt (above minus below) is one of the slow learner's drivers while it is in force. Sign in to add one.