WarnMOS – A MOS-based weather warning system
Sebastian Trepte
Deutscher Wetterdienst (DWD)Offenbach am Main, Germany
11th EMS Annual Meeting10th European Conference on Applications of Meteoro logy (ECAM)12 – 16 September 2011Berlin, Germany
11th EMS & 10th ECAM, 12-16 September, Berlin
Overview
Goal of the application
Model Output Statistics and some specific characteristics
Features of WarnMOS
Weather warning elements
Mapping on the 1x1km²-grid
Forecast examples and comparisons
Outlook and ongoing work
11th EMS & 10th ECAM, 12-16 September, Berlin
Goal of WarnMOS
• Generation of an automatic weather warning guidance for Germanybased on the DWD warning criteria
• Derived from numerical model data, SYNOP station data, DWD radar network, and lightning observation data
• All forecast elements should be available hourly on the nowcast and the very short-term scale
11th EMS & 10th ECAM, 12-16 September, Berlin
Model Output Statistics (MOS)
• Develops relationship equations between observed and modelforecast weather elements (using multiple linear regression)
• Combination of different independent data sources
• Use of extrapolation/persistence and model forecasts (NWP) as predictors
• Takes advantages of predictor variables in NWP output notaccessible as observations
• Bias correction
Kind of forecasts: deterministic, probabilistic and derived elements
11th EMS & 10th ECAM, 12-16 September, Berlin
MOS specifics (1): Classification of the SYNOP station s
WarnMOS domain
0 – 200Coast
> 1000F
800 – 1000E
600 – 800D
400 – 600C
200 – 400B
0 – 200A south
0 – 200A middle
0 – 200A north
Elevation (m)Elevation Class
11th EMS & 10th ECAM, 12-16 September, Berlin
MOS specifics (2): Multi-Station Equations
▪ ▪
▪
▪ ▪
▪ ▪
“region“
▪ SYNOP stationin elevation class
MOS regression equation(Multi-Station)
WarnMOS domain
county
●
grid point
●
11th EMS & 10th ECAM, 12-16 September, Berlin
MOS (3): Estimation of the observation predictorOperational mode
▪ ▪
▪ ▪ ▪
x▪
▪
▪
■ station
x county centre / grid point
5 nearest stations
weighted meandepending on horizontal distance and difference of the elevations
11th EMS & 10th ECAM, 12-16 September, Berlin
Features of WarnMOS
• MOS code by Knüpffer & Haalman (2007), modified by DWD• Input: GME and IFS model forecasts (two runs per day)
260 SYNOP observations (hourly)radar reflectivity, lightning observations (last 15 min)
• 6 years of historical data for the MOS development• Advection of SYNOP, radar and lightning observations (trajectories)• Running every 15 min• 74h, 24h, and 6h forecasts in 1h and 3h intervals• Raw output: deterministic and probabilistic predictands• Transformation of point probabilities to area probabilities (counties)
depending on area size and spatial autocorrelation function of the predictand (Taubenheim, 1969)
• Condensed output: warning elements (probabilities of exceedance)
11th EMS & 10th ECAM, 12-16 September, Berlin
Condensing of the forecasts
There are 177 WarnMOS predictands!Predictands are valid for different reference periods: 1, 3, 6, 12, 24, 48 hours
→ Reduction of the number of predictands to 27 warning elementsAggregation according to the DWD warning criteriaWeighting of the reference periods depending on forecast valid time
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecasted warning elements
Wind gust speed greater than 27, 33, 47, 53, 63, 75 kts
Thunderstorm 3 categories
Heavy rainfall 2 categoriesContinuous rainfall 3 categories
Snowfall light, moderate, heavySnowdrift 2 categoriesThaw heavy
SlicknessBlack ice 2 categories
Frost 2 categories
Fog Visibility below 150 m
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecasts for German counties
Example:Probability ofthunderstorm+12h forecast
The orangecolour indicatesthe 50% threshold
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecasts on the fine mesh
Goal: Provide warnings on a 1x1 km² grid for combination with otherhigh-resolution grid-based forecast products (→ DWD AutoWARN)
Grid configuration corresponds to the radar product grid at the DWD→ adequate grid for verification
Grid is independent from a county reconfiguration
Forecasts on a 2D grid can be visualized easier than having the 3th dimension of the elevation classes
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecasts on the fine mesh
Best way: direct calculation of MOS forecasts on the high-resolution grid
Problem: computing time much longer than forecast run interval
Solution:(1) Forecast calculation stepcalculation of forecasts on a coarse 20 x 20 km² grid and all elevation classes occurring within a grid cell
(2) Mapping on the fine mesh- the elevation of the point in the fine mesh appoints the elevation class- the nearest neighbour point in the coarse grid having the same
elevation class defines the forecast value on the fine mesh
Target grid covers all elevation classes but is only two-dimensional
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecast example on counties
Probability ofrainfall > 25mm/12h+4h forecast
The orangecolour indicatesthe 50% threshold
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecast example on high-resolution grid
Probability ofrainfall > 25mm/12h+4h forecast
The light greencolour indicatesthe 50% threshold
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecast example on counties
Probability ofmax wind gust > 27 kts+5h
The orangecolour indicatesthe 50% threshold
11th EMS & 10th ECAM, 12-16 September, Berlin
Forecast example on high-resolution grid
Probability ofmax wind gust > 27 kts+5h
The light greencolour indicatesthe 50% threshold
11th EMS & 10th ECAM, 12-16 September, Berlin
Outlook and ongoing work
• Evaluation of WarnMOS on the fine mesh
• Spatial interpolation of the MOS forecasts on the fine mesh
• Integration into the DWD AutoWARN-ModelMIX concept
• Talk by C. Primo (AM5): “On the choice of thresholds to give warnings”
11th EMS & 10th ECAM, 12-16 September, Berlin
Thank you for your attention!