Folie 1 - Copernicus

Download Report

Transcript Folie 1 - Copernicus

Deutscher Wetterdienst
Titelfoto auf dem Titelmaster einfügen
NowCastMIX
Automatic integrated warnings from continuously monitored nowcasting systems
using spatially clustered fuzzy-logic assessments of storm attributes
Dr. Paul James, German Weather Service, ECAM/EMS Conference, Reading, 9. Sept. 2013
NowCastMIX in AutoWARN
 The AutoWARN process at the DWD monitors several systems automatically for potential
warning situations on different time scales.
 Warning polygons are sent to duty meteorologists for possible modification before
products are generated.
 NowCastMIX integrates all data from nowcasting systems
 Consistent, optimised and intelligent warning solution
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX in AutoWARN
 NowCastMIX monitors several nowcasting systems on a 5 minutes update cycle
 Radar products, lightning, surface obs., NWP model outputs for background
 Data mapped onto a 1 x 1 km grid with assessment of cell motion vector field
 Storm severity assessments using a fuzzy-logic method
 Spatial and temporal optimisation using clustering techniques
 Warning polygons covering the next 60 minutes produced and sent on to AutoWARN
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX - Overview
Warning events: Thunderstorms / Rain
 Different warning events are given an
code number (ii) (e.g. „31“)
 10 thunderstorm and 3 torrential rain
event types need to be monitored in
AutoWARN
 The thunderstorm severity depends
on the presence (and magnitude) of
these attributes:
Severe Gusts
Torrential Rain
Hail
The 10 Thunderstorm and 3 Torrential rain events
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Cell motion vector field
 Background cell motion vector
field (CVF) is constructed
using the following sources:
 Pattern-Matching of radar
echoes in consecutive
images -> motion vectors
 Explicit cell tracking
vectors* (KONRAD and
CellMOS systems)
(*mapped with a Gaussian distribution)
 Form weighted mean -> CVF
 Loop over all input tracking
vectors to remove possible
erroneous outliers
 Successive improvement
of CVF
13:15 UTC, 22.06.2011
Cell motion vector field
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Construction of warning cones
 Warning cones are created, opening up in the direction of cell motion
 3 possible triggers for creating a cone:
 KONRAD Cell ( Radar echoes > 46 dBZ )
 CellMOS Cell ( Radar echoes > 37 dBZ + Lightning Strike )
 Lightning Strike with weaker radar echoes
 Fuzzy Logic applied at the cell centre to estimate storm severity
 Attribute strength (Gusts, Hail, Rain) as a function of input data
 Weighting function in cases where two or more cones overlap – higher severity preferred
 Total cone length = 60 min (as function of cell speed)
 Initial radius = 10km, Expansion angle = 7,5 degrees
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Fuzzy Logic Sets
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Integration of station reports
 In some cases the automatically determined thunderstorm severity may not be able to
capture the characteristics of the storm sufficiently
 To reduce the risk of missing very severe weather, real-time synoptic station reports are
routinely monitored by NowCastMIX
 These report e.g. measured gust speeds, hail occurrence, recent rainfall totals etc.
 The likely current location of the severe weather event is estimated using the background
cell motion vector field
 Note that a certain time has typically elapsed since the reported weather started
 A certain temporal and spatial uncertainty typically exists, depending on the type of
severe weather being reported
Cell motion vector
Station
Current region of relevance
 The severity levels of warnings in this region can be raised if necessary
P. James / DWD - ESSL 2013
NowCastMIX
Optimization of Warnings
 Optimal warnings need to find a balance between:
Precision / Accuracy
Workability / Usefulness
Realistic, strictly correct
Smoother in space and time. Easier
to process for the duty meteorologists,
easier for customers to assimilate.
Complex, over detailed and rapidly
changing warnings. Hard for duty
meteorologists to assess and process.
Hard for customers to understand and
assimilate
Locally less precise, some details
may be left out. Greater danger of
systematic biases (e.g. over warning)
 In nowcasting you cannot have both at the same time!
 NowCastMIX would tend by nature to be over-precise
 An optimal balance is approached via a clustering method
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Clustering
 Clusters (cell groups) are formed with the DBSCAN algorithm
 Examining all cells in a time window covering the last 20 minutes
 The highest severity value in a cluster is mapped onto all cells in that cluster
 High severity levels are thus held up for at least 20 minutes
 Results in a certain temporal smoothing
 Problem: The clusters themselves sometimes change significantly from one run to the next,
5 minutes later, resulting in a new source of temporal noise on larger scales
 A cluster ensemble (up to ~600) is created, using differing randomly perturbed cell
positions (up to 6km shifts)
 To improve computational efficiency the directions of the random perturbations are
successively biased towards more useful sectors as the ensemble members are
generated (Adaptive Clustering Ensemble)
 The clustering member which is finally used is that which is most similar to the clustering
which was used 5 minutes earlier
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX Core Technology
5. Clustering example
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX – From analysis
to warning polygons
14:05 to 15:05 UTC, 22.06.2011
Analysis* + Warning areas
Clusters + Warning polygons
Warning situation for the next hour
* 20 minute time window (13:45-14:05)
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Outlook
 Further refinement of clustering techniques to optimise warning proposals
 Inclusion of further data sources (e.g. rapid-scan based satellite products)
Can early warnings be given before the first lightning or radar signals are occurring?
 Feedback from verification results and from assessments by duty meteorologists
Parameter optimisation, further tuning of fuzzy logic sets
 Expansion of the warning domain to cover wider area of W. Europe (e.g. FAB EC)
 Winter nowcasting (snow, freezing rain)
Radar-based detection (snow, precip.) combined with surface freezing indicators
NWP-based snow forecasts and temperature profiles
Fuzzy logic combinations of such information
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013
NowCastMIX
Thanks for your attention !
Questions?
Dr. P. James (Deutscher Wetterdienst, Offenbach, Germany) – ECAM/EMS Conference, Reading, UK, 9. Sept. 2013