Forecast against Google's own levels
GloFAS reanalysis, this season
Station tables
About this reading
- What is shown. Google Flood Hub's forecast at each station as a share of that station's Google level for the window, counted with the trigger's own rule: a number of stations over their level on one forecast day. The monitoring page reads the Deyr windows on GloFAS; this page applies the same rule to Google.
- Levels at 1-in-6. In Deyr the Google levels are set at a 1-in-6 return period on both rivers, not the trigger's 1-in-4 and 1-in-5. On the 1999 to 2023 record that makes Google's Shabelle Deyr activation years the same four as the GloFAS v5 design: 2006, 2014, 2019 and 2023. On the Juba the same setting gives 2006, 2008, 2019 and 2023.
- Why it is informational. The endorsed framework runs Deyr on GloFAS. In the trigger design, Google on Deyr catches the same four severe Shabelle years as GloFAS v5 and adds 2008, a flood year; on the Juba it activates in three years outside the gauge benchmark (2013, 2015, 2019) and its peaks run 8 to 10 days ahead of the gauges. The GloFAS version page describes the gap in the version 4 record on the Shabelle.
- Lead days. The live Google feed returns about five to six days ahead of the issue, so the 1 to 7 day band is read as far as the feed goes. A readiness band (8 to 12 days) is not available on Google.
- Reanalysis block. GloFAS also publishes a near-real-time reanalysis, a day or two behind. The block logs the season so far, from 1 September for Deyr and 1 February for Gu, at each station as a share of that version's own level for the window, at the trigger's return period. It is the model's account of what the river has been doing, not a forecast. Version 5's reanalysis is a fixed release to 2025 with no near-real-time stream yet; it joins the block the day EWDS extends it. The versions differ: on the 1999 to 2023 record version 4 does not reach its Shabelle level in Deyr 2006 or 2023, while version 5 does.
- Record. Levels are fitted on Google's own retrospective record, 1999 to 2023, with the same estimator as the trigger levels. Dollow reads the Juba main-stem gauge, the only one Google serves there.
Contact
Reach out to OCHA Data Science at ocha-datascience@un.org with any questions.

