Or open a solved case
- 2021 Taiwan spring droughtTaiwan · Feb – May 2021Soil moisture, monthly anomaly ± 12 monthsUnusually low · event mean −1.7σ, p = 0.02PrecipitationSoil moistureTerrestrial water storageVegetation greennessOpen the analysis →
- 2022 Pakistan floodsIndus basin, Pakistan · Jul – Sep 2022Precipitation, monthly anomaly ± 12 monthsUnusually high · event mean +6.3σ, p < 0.001PrecipitationRunoffSoil moistureTerrestrial water storageOpen the analysis →
- 2019–20 Australian Black SummerSouth-east Australia · Sep 2019 – Jan 2020Soil moisture, monthly anomaly ± 12 monthsUnusually low · event mean −2.3σ, p = 0.01PrecipitationSoil moistureAir temperatureVegetation greennessOpen the analysis →
How an investigation runs
Name the event
What happened, in your own words: “2022 Pakistan floods”. Years and months in the text become a suggested period.
Set place and time
Pick one of 1,313 river sub-basins on the globe or by name, and confirm the event months.
Choose variables
A local AI model suggests which of 17 categories relate to the event (a keyword table steps in when it is offline). Tick or untick any.
Read the lanes
Each variable’s anomaly is a lane on one shared time axis. Drag an interval before, during or after the event; every lane’s statistics follow.
What counts as unusual
The full method- Baseline
- every month outside the event ± 12 months. The seasonal cycle and linear trend are fitted on it, so the event cannot hide itself.
- Anomaly, z
- value − trend − seasonal cycle; z divides it by σ, the typical size of a baseline anomaly.
- p value
- how often ordinary noise, with month-to-month memory, gives an interval mean this far from zero. A second test against every same-length baseline window must agree before we say “unusually low/high”.
- Rank
- where the interval sits among the same months of every year (“1st lowest of 26”). A description, not a p value.
What it cannot tell you
Timing match, not proof of cause.
An unusual interval lines up with the event. Why it happened needs physics and other evidence.
Monthly basin averages.
A week-long flood or one hard-hit valley is diluted: most basins cover 7,000 to 240,000 km².
Models, reanalyses, satellites.
Most layers are model or reanalysis output; others are satellite retrievals. Every layer names its product.