Code: EDI-2018-2-UBIMET

Domain: Environment

Summary

Detect incorrect values in weather observation data in Europe.

Proposed by

 

Ubimet GmbH. The international weather service, with headquarters in Vienna, Austria, is the competence centre for meteorology and severe weather alerts and weather analytics.

Description

Station based weather observations serve as a crucial input to different analysis and subsequent forecast models. In-situ measurements commonly include standard parameters like temperature, humidity, pressure, wind speed, wind direction, gusts, precipitation and many others. In order to use these data in a reasonably way, erroneous data have to be discarded in a first step. The challenge proposed seeks to derive an algorithm capable to detect incorrect values while retaining extreme, but indeed realistic data. In doing so the following three components have to be taken into account:

 

  1. spatial consistency of observations (among stations),
  2. temporal consistency of observations, i.e. the generation of homogeneous time series, and
  3. consistency among a set of parameters (e. g. a drastic change in temperature goes, in general, along with a change in humidity and vice versa)

Data

The challenge has the following sample datasets available for download

Expected outcomes

  • Methodology to either cancel, flag or correct wrong values. Continuous, stringent documentation and process management of algorithms and methods used. New or novel algorithms for detection of false observational data outperforming UBIMET’s own quality control method.
  • Measurable result: Detection of at least 80% (and less than 200%) of values detected by own quality control method.
  • Detection of low quality stations (questionable quality in at least one observed parameter/time step) and flagging of those, according to quality indices mentioned before.
  • Reduction of cost of quality control by process automation, resulting in higher margins and longer lead times for time critical warnings by 20% (to be verified in operational use).
  • Detection of outliers and detection of low quality measurements of single parameters and detection of erroneous or suspicious patterns of parameters, according to quality indexes mentioned before.

How do we apply?

RSS
Follow by Email
Facebook
YouTube