Drought 2018 - How predictable?


From August 2018, WSL was able to provide a hydrological monthly forecast on drought.ch for the first time in real time. In the aftermath of the drought, the following questions now arise:

  1. at what point in time and in which parameters (runoff, soil moisture, groundwater) were the first signs of a possible coming dry period visible?
  2. at what point were the predictions reliable and had added value?
  3. spatial variability of reliability - in which regions of Switzerland was drought more predictable and for which parameters?
  4. is there a correlation between general weather conditions (for medium-term forecasts, Richardson, et al., 2017), teleconnections (for long-term forecasts, Butler, et al., 2019) and spatial and temporal variability in the quality of the forecast?
  5. possibilities for improvement through pre- and postprocessing of the predictions (Bogner, et al., 2016; Monhart, et al., 2018)?
  6. comparison with predictions since 2012.

Completed analysis

  • monthly forecasts have been recalculated for the years 2012 - 2108
  • Implementation of forecast improvement methods (pre- and post-processing) 
    • Monhart, et al. (2019) developed post-processing methods for monthly forecasts, which have been applied for the year 2018 in Switzerland
  • Statistical Analyses – Verification
    • The skill of the monthly forecasts of hydrological relevant variables (e.g. surface runoff and soil moisture) has been analysed according to the methodologies applied to temperature and precipitation forecasts at Meteoswiss.   Therefor the probabilities of the variables being in the week of question below, above or in the medium range of long term averages will be shown.


In the figure below an example of such a forecast (so called tercile forecast) is shown for the surface runoff for the first week in July 2018 in comparison to the results of the simulations driven by observed meteorological data (i.e. reference).

First results of the verifications show that the application of post-processing methods improves the forecast skill for one to two weeks depending on the variable and the region. 

In general the results indicate that the predictability of the drought period 2018 was quite high caused by a long-lasting stable atmospheric conditions.

Planned analysis

  • Role of large-scale weather situations: a) Are there relationships to patterns of large-scale weather situations, atmospheric conditions and their predictability within Switzerland? b) Application of machine learning methods for pattern identification (Ambühl, 2010; Shen, 2018)
  • Collaboration with Prof. Daniela Domeisen (SNF-EHT Professorship) regarding Teleconnections
  • Further analysis of tercile forecasts for hydrological quantities - How can the information from temporal and spatial forecasts best be summarized in warning levels?