Description
The Trans-African Hydro-Meteorological Observatory (TAHMO, www.tahmo.org) operates a network of meteorological and hydrological measurement stations in Sub-Saharan Africa. At this moment, there are 620 stations in twenty countries, making TAHMO the largest provider of scientific in situ data in the region. Over 90% of the weather stations have been placed at schools and universities. The stations are robust and do not have moving parts (see: https://www.metergroup.com/en/meter-environment/products/atmos-41-weather-station). Unfortunately, hosts cannot always take complete care of the station. TAHMO has engineers in the different countries who can travel to the stations for major maintenance issues. The travel costs tend to be a major operating expense so TAHMO tries to minimize station visits. We would very much like to be able to assess sensor performance from the data alone. The QA system consists of a hybrid machine/human system of over-conservative automated flagging and human assessment of flagged entries. When there is a clear failure, a ticket is generated and communication with the responsible engineer is started to resolve the issue.An important failure mechanism of any rain gauge with a funnel is clogging with dust, leaves, and “guano”. It is, however, very difficult to detect clogging quickly from the data stream. When it rains and the rain goes undetected (‘false negative”), there is no immediate measure to determine this. Rainfall in Africa is often spatially very variable. Of course, when a station does not report any rain over an extended part of the rainy season, there is probably something wrong, but how can one detect this as rapidly as possible? Currently, TAHMO uses Machine Learning methods, fed with information from neighboring stations and satellites, to determine if a station is likely to be clogged. If there are active hosts at the station sites, this can then often quickly be confirmed and remedied. Detection of false negatives remains, however, a wickedly complicated problem and our recent work will be presented.
| Period | 12 Dec 2022 |
|---|---|
| Event title | AGU Fall Meeting 2022 |
| Event type | Conference |
| Location | Chicago, United StatesShow on map |