How well has the Mongolian dzud risk map, whose share of 'very high risk' area in each province triggers Red Cross and FAO anticipatory cash and fodder before killer winters, actually separated winters with heavy livestock losses from mild ones, and what trigger rule built from the same public inputs would have done better on years it was not tuned on?
- Good it would do
- 3 / 5
- Can AI help
- 4 / 5
- How neglected
- 4 / 5
- Can we check it
- 3 / 5
Why it's stuck
The people who run the trigger say plainly that it has never been scored. The Mongolian Red Cross early action protocol (last updated November 2020) lists the dzud risk map's 'Exact skill score' as 'unknown' and says 'It is unknown if there has been a quantitative assessment of skill'; it adds that with so few years of the map's output, 'the false alarm rate and so forth' were difficult to assess, and that an earlier version of the trigger 'would trigger too frequently'.
What we'd try
Build a year-by-province table of observed adult livestock losses from Mongolia's National Statistics Office (public table DT_NSO_1001_136V1, 'Losses of adult animals' by species and province, 1991 to 2025, which this run opened through the office's open API) and score the trigger against it: hits, misses and false alarms for each winter the map has existed (first produced for 2015/16), using archived maps from Red Cross, FAO and UN reports. Because ten winters is too few, also rebuild the map back to about 2000 from its published recipe (11 weighted layers listed in the Red Cross protocol: snow depth, snow cover, drought index, biomass, carrying capacity, livestock numbers, temperature and precipitation anomalies and forecasts) using public satellite (MODIS) and reanalysis data, check the rebuilt maps against the real ones for the overlapping years, then compare the 20 percent province rule with alternative rules and a simple weather-only baseline using leave-one-winter-out testing.
What could stop us
The risk map only exists since the 2015/16 winter, so about ten winters can be scored, and the checker could not confirm that archived maps or the province loss table can be downloaded as data.
Do you know the answer?
One reply from someone in the field can settle what our AI can't.
Members can read the full record for this problem, check its sources and comment. Open the member view → Not a member? Here is how to join.