Challenges of Nowcasting by George A. Isaac 1, Monika Bailey 1, Faisal Boudala 1, Stewart Cober 1,...
-
Upload
rebecca-callahan -
Category
Documents
-
view
214 -
download
0
Transcript of Challenges of Nowcasting by George A. Isaac 1, Monika Bailey 1, Faisal Boudala 1, Stewart Cober 1,...
Challenges of Nowcasting
by
George A. Isaac1, Monika Bailey1, Faisal Boudala1, Stewart Cober1,
Robert Crawford1, Ivan Heckman1, Laura Huang1, Paul Joe1,
Jocelyn Mailhot2, Jason Milbrandt2, and Janti Reid1
1Cloud Physics and Severe Weather Research Section, Environment Canada
2Numerical Weather Prediction Research Section, Environment Canada
Workshop on Use of NWP for NowcastingBoulder, Colorado, 24-26 October 2011
Introduction
• Results from Canadian Airport Nowcasting (CAN-Now) and Science of Nowcasting Olympic Weather for Vancouver 2010 (SNOW-V10)
• Funding from Environment Canada, NAV CANADA, SAR-NIF, Transport Canada
Canadian Airport Nowcasting (CAN-Now)
• To improve short term forecasts (0-6 hour) or Nowcasts of airport severe weather.
• Develop a forecast system which will include routinely gathered information (radar, satellite, surface based data, pilot reports), numerical weather prediction model outputs, and a limited suite of specialized sensors placed at the airport.
• Forecast/Nowcast products are issued with 1-15 min resolution for most variables.
• Test this system, and its associated information delivery system, within an operational airport environment (e.g. Toronto and Vancouver International Airports ).
Isaac et al., 2011: The Canadian Airport Nowcasting System (CAN-Now, Submitted to Meteorological ApplicationsIsaac et al., 2011: Decision Making Regarding Aircraft De-Icing and In-Flight Icing Using the Canadian Airport Nowcasting System (CAN-Now), SAE 2011 International Conference on Aircraft and Engine Icing and Ground Deicing, 13-17 June 2011, Chicago, Illinois, USA
Main equipment at Pearson at the old Test and Evaluation site near the existing Met compound
CAN-Now Situation Chart
SNOW-V10Science of Nowcasting Olympic Weather – Vancouver 2010
• To improve our understanding and ability to forecast/nowcast low cloud, and visibility.
• To improve our understanding and ability to forecast precipitation amount and type.
• To improve forecasts of wind speed, gusts and direction.
• To develop better forecast system production systems.
• Assess and evaluate value to end users.
• To increase the capacity of WMO member states.
Olympics
Societal Benefits
The Winter Olympic ChallengeSteep topography, highly variable weather elements in space and time
5 km
Village Creekside
Model Name Organization Country Spatial Resolution
Temporal Resolution Available
Times of Day Run (UTC)
Length of Forecst (hours)
General Description
ABOMLAM1km Environment Canada Canada 1 Km 15 min Every 15 min Max 6 h
Adaptive Blending of Observation and Models
using GEM LAM1k
ABOMREG Environment Canada Canada 15 km 15 min Every 15 min Max 6 h
Adaptive Blending of Observation and Models
using GEM Regional
INTW Environment Canada Canada 1 and 15
km 15 min Every 15 min Max 6 hINTegrated Weighted
Model using LAM1k, GEM Regional and Observations
LAM1k Environment Canada Canada 1 km
30 s (Model), 15 min
(Tables)
11 and 20 UTC 19 h Limited-Area version of
GEM model
LAM2.5k Environment Canada Canada 2.5 km
1 min (Model), 15 min (Tables)
06 and 15 UTC 33 h Limited-Area version of
GEM model
REG Environment Canada Canada 15 km
7.5 min (Model), 15 min (Tables)
00, 06, 12, 18 UTC 48 h
Regional version of GEM (Global Environmental
Multiscale) model
Canadian Models Used in SNOW-V10
Model Name Organization Country Spatial Resolution
Temporal Resolution Available
Times of Day Run (UTC)
Length of Forecst (hours) General Description
CMAChinese
Meteorological Administration
China 15 km & 3 km 1 hour 00 and 12 UTC 48 h & 24 h CMA GRAPES-Meso NWP model
WDTUSL
Weather Decision
Technologies and
NanoWeather
USA pointwise or 100 m grid 02, 08, 14, 20 UTC 48 h
Surface layer model nested in NAM. Works particularly well in
quiescent condistions..
WSDDM
National Center for Atmospheric
Research (NCAR)
USA Radar Resolution
10 min (based on
radar update)
Every 10 min 2 hours
Nowcast based on storm tracking of radar echo using cross correlation and real-time calibration with surface
precipitation gauges.
ZAMGINCA
Central Institute for Meteorology
and Geodynamics
(ZAMG)
Austria 1 km 1 hour Every hour 18 hours
The Integrated Nowcasting Through Comprehensive Analysis (INCA) system uses downscaled ECMWF forecasts as a first guess and applies corrections according
to the latest observation.
ZAMGINCARR
Central Institute for Meteorology
and Geodynamics
(ZAMG)
Austria 1 km 15 min Every 15 min 18 hours
The precipitation module of INCA combines raingauge and radar
data, taking into account intensity-dependant elevation effects. The
forecasting mode is based on displacement by INCA motion
vectors, merging into the ECMWF model through prescribed
weighting.
Other Countries Models Used in SNOW-V10
Google Map of Outdoor Venues for Sochi 2014 Olympics. A box 7 km in size is drawn on map. The tops of the mountains are approximately 2200m and the valley 600 m. The area is about 35 km inland from Black Sea.
Products During Olympics• Each group produced a Table showing 24 hour forecast
of significant variables for main venue sites (hourly intervals and 10 to 15 min intervals in first two hours). Similar to what forecasters produce
• A Research Support Desk was run during Olympics and Paralympics (virtual and on-site) providing real time support to forecasters.
• A SNOW-V10 Web site was created with many of the products (time series for sites, remote sensing products, area displays, soundings (gondola and others), and a very successful Blog.
Equipment on Whistler mountain provided good data for forecasters and help in understanding
weather processes
Harvey’s Cloud
Adaptive Blending of Observations and Models (ABOM)
ABOM reduces to:• Observation persistence when r = s = 0, w = 1• Pure observation trend when r = 0, s = 1, w = 0 • Model + obs trend + persistence for all other r, s and w
Change predictedby model
Forecast atlead time p
CurrentObservation
Change predictedby obs trend
• Coefficients s and r are proportional to the performance over recent history, relative to obs persistence (w) and to
• Local tuning parameters (not yet optimized at all SNOWV10 sites)
• Updated using new obs data every 15 minutes
Bailey ME, Isaac GA, Driedger N, Reid J. 2009. Comparison of nowcasting methods in the context of high-impact weather events for the Canadian Airport Nowcasting Project. International Symposium on Nowcasting and Very Short Range Forecasting, Whistler, British Columbia, 30 August - 4 September 2009.
Temperature Mean Absolute Error: Averaged by Nowcast Lead Time to 6 Hours
0 1 2 3 4 5 60
0.5
1
1.5
2
2.5
3
3.5
4
Forecast Lead Time (Hours)
Me
an
Ab
solu
te E
rro
rs
VOC TEMPERATURE MAE
ABOM REGREGObs PersLAM 2.5ABOM LAM 2.5LAM 1ABOM LAM1
0 1 2 3 4 5 60.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
Forecast Lead Time (Hours)
Me
an
Ab
solu
te E
rro
rs
RND TEMPERATURE MAE
ABOM REGREGObs PersLAM 1ABOM LAM1LAM 2.5ABOM LAM2.5
1) Nowcast - Model crossover lead-time 2) Improvement over persistence
Background of INTW• INTW refers to integrated weighted modelINTW refers to integrated weighted model
• LAM 1k, LAM 2.5k, REG 15k and RUC (CAN-Now) were selected to generate INTW
• Major steps of INTW generationMajor steps of INTW generation– Data pre-checking - defining the available NWP models and
observations
– Extracting the available data for specific variable and location
– Calculating statistics from NWP model data, e.g. MAE, RMSE
– Deriving weights from model variables based on model performance
– Defining and performing dynamic and variational bias correction
– Generating Integrated Model forecasts (INTW)
Huang, L.X., G.A. Isaac and G. Sheng, 2011: Integrating NWP Forecasts and Observation Data to Improve Nowcasting Accuracy, Submitted to Weather and Forecasting
Aerials’ jump
Spectator grandstand
Preparation of field-of-play 2 hours before training…
Freestyle Skiing Ladies Aerials Final February 24, 2010
Notes by Mindy Brugman as posted on SNOW-V10 Blog
The event….Last night at the Cypress aerial site it began with drizzle and was quite foggy… We walked to the bright lights of the venue. There was no precipitation at the venue the entire competition. The competition started at 1930 PST. The first pictures (not shown here) are taken at the start of the competition and you can see that the fog was much worse at the start. I stopped taking pictures since I could not see anything. Then it began to clear just near the end of the competiton – and I clicked a few pictures. You can detect a flying bug above the lights. Thats really the gold medal winner, Lydia Lassila, of Australia!Based on the event last night – I suspect ladies aerials may qualify as a new paralympic sport for the visually impaired.
INTW Forecasts of Visibility at VOG
Visibility LAM 1Kincluding 24 hours before ladies aerials
09:00 15:00 21:00 03:00 09:00 15:00 21:00 03:00 09:00 15:0010
1
102
103
104
105
Time
min
vis_
m
VOG: 2010-02-23 09:00 to 2010-02-25 15:30
3600 sec7200 sec14400 sec21600 secObservationsLAM 1K
Ceiling LAM 2.5K including 24 hours before ladies aerials
09:00 15:00 21:00 03:00 09:00 15:00 21:00 03:00 09:00 15:0010
1
102
103
104
105
Time
min
ceil_
m
VOG: 2010-02-23 09:00 to 2010-02-25 15:30
3600 sec7200 sec14400 sec21600 secObservationsLAM 2.5K
Total precip rate REG 15K
including 24 hours before ladies aerials
09:00 15:00 21:00 03:00 09:00 15:00 21:00 03:00 09:00 15:000
0.5
1
1.5
2
2.5
3
3.5
Time
tlep
rate
_m
mp
h
VOG: 2010-02-23 09:00 to 2010-02-25 15:30
3600 sec7200 sec14400 sec21600 secObservationsREG 15K
NWP Model with Minimum MAE in CAN-Now for Winter Dec 1/09 – Mar 31/10 and
Summer June 1/10 to Aug 31/10 Periods
Based on First 6 Hours of Forecast
Winter period – Dec. 1, 2009 to Mar. 31, 2010
Summer period - June 1 to August 31, 2010
Variable LAM REG RUC INTW
CYYZ CYVR CYYZ CYVR CYYZ CYVR CYYZ CYVR
TEMP 6 3 4 3.5 4.5 5 2.5 0.5
RH 6 6 no 6 no no 3.5 3
WS 2.5 3.5 4.5 3.5 3 no 1 2.5
GUST no no no 5 3.5 no 1.5 1.5
Winter
Summer
Variable LAM REG RUC INTW
CYYZ CYVR CYYZ CYVR CYYZ CYVR CYYZ CYVR
TEMP 2.5 2.5 2.2 2.5 1.5 no 0.5 0.5
RH 3 3 3.2 4.5 3 no 1 1
WS 3 5 3.5 5 2.2 no 1.5 2.5
GUST no no 5.5 no 2.2 no 0.5 4
Time (h) for Model to Beat Persistence
6h RUC Used
Huang, L.X., G.A. Isaac and G. Sheng, 2011: Integrating NWP Forecasts and Observation Data to Improve Nowcasting Accuracy, Submitted to Weather and Forecasting
Observation – Model grid match
Temperature at CYVR
10
12
14
16
18
20
22
24
26
28
30
Jul 09 Jul 10 Jul 11 Jul 12 Jul 13 Jul 14 Jul 15 Jul 16 Jul 17 Jul 18 Jul 19
Date [UTC]
Te
mp
era
ture
[C
]
OBS
YVR NN
YVR BI
VAN 1Pt East
Mean Absolute Errors for RUC 6h
YVR NN YVR BI VAN 1PE
Temperature (ºC) 5.7 5.2 1.5
Wind Speed (m/s) 1.6 1.5 1.7
Wind Direction (º) 49.7 49.1 48.3
RUC temperature at CYVR July 2010
• Compared data from 3 model points
• Consider season when selecting the best point
• CAN-Now: usually nearest grid point (with no water mask)
J. Reid 2010
Heidke Skill Score: Multi-Categories
1 2 3 . . . . . K total
1 N(F1)
2 N(F2)
3 N(F3)
. . . .
K N(Fk)
total N(O1) N(O2) N(O3) N(Ok) N
Observed category
ForecastCategory
j
i Using:
Calculate:
Variable Category 1 Category 2 Category 3 Category 4 Category 5 Category 6 Category 7 Category 8Winds < 5 kts 5 ≤ w < 10
kts10 ≤ w < 15
kts15 ≤ w < 20
kts 20 ≤ w < 25
kts w ≥ 25 kts - -
Wind Direction
d ≥ 339 & d < 24º (N)
24 ≤ d < 69º (NE)
69 ≤ d < 114º (E)
114 ≤ d < 159º (SE)
159 ≤ d < 204º (S)
204 ≤ d < 249º (SW)
249 ≤ d < 294º (W)
294 ≤ d < 339º (NW)
Visibility v < 1/4 SM 1/4 ≤ v < 1/2 SM
1/2 ≤ v < 3 SM
3 ≤ v < 6 SM v ≥ 6 SM - - -
Ceiling c < 150 ft 150 ≤ c< 400 ft
400 ≤ c< 1000 ft
1000 ≤ c< 2500 ft
2500 ≤ c< 10000 ft
c ≥ 10000 ft - -
Precip Rate r = 0 mm/hr (None)
0 < r ≤ 0.2 mm/hr (Trace)
0.2 < r ≤ 2.5 mm/hr (Light)
2.5 < r ≤ 7.5 mm/hr
(Moderate)
r > 7.5 mm/hr
(Heavy)
- - -
Precip Type No Precip Liquid Freezing Frozen Mixed (w/Liquid)
Unknown - -
Table 2 (From Bailey's CAWW Talk)
Categories Being Used in CAN-Now Analysis
Summary of HSS/ACC results
Model Variable Original Overall HSS / ACC
Relaxed Timing (±60min)
HSS / ACC
REG CLD_BASE_HGT 0.45 / 0.62 0.46 / 0.61
PCP_RATE 0.30 / 0.70 0.26 / 0.62
VIS_FB 0.28 / 0.78 0.29 / 0.74
LAM PCP_RATE 0.29 / 0.73 0.27 / 0.67
VIS_FB 0.26 / 0.78 0.28 / 0.74
RUC6h CLD_BASE_HGT 0.28 / 0.51 0.32 / 0.52
PCP_RATE 0.40 / 0.84 0.40 / 0.81
VIS 0.22 / 0.66 0.19 / 0.60
CYYZ (Winter)
J. Reid 2010
•Scores dominated by long periods of non-events
•Relaxed timing: No clear changes, for better or worse
•Similar results at CYVR
Variable Category 1 Category 2 Category 3 Category 4 Category 5 Category 6 Category 7 Category 8 Category 9Temperature -25 C < -25≤ T<-20C -20≤ T<-4C -4≤ T<-2C -2≤ T< 0C 0≤ T< +2C +2 ≤ T< +4C ≥ +4C
RH < 30% 30≤ RH< 65% 65≤ RH< 90% 90≤ RH< 94% 94≤ RH< 98% ≥ 98% Winds < 3 m/s 3 ≤ w < 4
m/s4 ≤ w < 5
m/s5 ≤ w < 7
m/s7 ≤ w < 11
m/s11 ≤ w < 13
m/s13 ≤ w < 15
m/s15 ≤ w < 17
m/s≥ 17m/s
Wind Gust < 3 m/s 3 ≤ w < 4 m/s
4 ≤ w < 5 m/s
5 ≤ w < 7 m/s
7 ≤ w < 11 m/s
11 ≤ w < 13 m/s
13 ≤ w < 15 m/s
15 ≤ w < 17 m/s
≥ 17m/s
Wind Direction
d ≥ 339 & d < 24º (N)
24 ≤ d < 69º (NE)
69 ≤ d < 114º (E)
114 ≤ d < 159º (SE)
159 ≤ d < 204º (S)
204 ≤ d < 249º (SW)
249 ≤ d < 294º (W)
294 ≤ d < 339º (NW)
Visibility v < 30m 30 ≤ v < 50m
50 ≤ v < 200m
200 ≤ v < 300m
300 ≤ v < 500m
≥ 500m - -
Ceiling c < 50m 50 ≤ c< 120 m
120 ≤ c< 300 m
300 ≤ c< 750 m
750 ≤ c< 3000 m
c ≥ 3000 m - - -
Precip Rate r = 0 mm/hr (None)
0 < r ≤ 0.2 mm/hr (Trace)
0.2 < r ≤ 2.5 mm/hr (Light)
2.5 < r ≤ 7.5 mm/hr
(Moderate)
r > 7.5 mm/hr
(Heavy)
- - - -
Precip Type No Precip Liquid Freezing Frozen Mixed (w/Liquid)
Unknown - - -
Table 5 (2nd Revised Suggestion for SNOW-V10 Verification)
Categories for SNOW-V10 HSS Analysis
LAM 1km - 40LAM 2.5km - 23GEM-15km - 7
NWP Model with Highest HSS Score During SNOW-V10
Summary
• RH predictions are poor, barely beating climatology. (Impacts visibility forecasts)
• Visibility forecasts are poor from statistical point of view. (also require snow and rain rates)
• Cloud base forecasts, although showing some skill, could be improved with better model resolution in boundary layer.
• Wind direction either poorly forecast or measured. • There are many difficulties in measuring parameters,
especially precipitation amount and type. • Overall statistical scores do not show complete story.
Need emphasis on high impact events.• Selection of model point to best represent site is a critical
process.
Summary (continued)
• Weather changes rapidly, especially in complex terrain, and it is necessary to get good measurements at time resolutions of at least 1 -15 min. CAN-Now and SNOW-V10 attempted to get measurements at 1 min resolution where possible.
• Because of the rapidly changing nature of the weather, weather forecasts also must be given at high time resolution.
• Verification of mesoscale forecasts, and nowcasts, must be done with appropriate data (time and space). Data collected on hourly basis are not sufficient.
• Nowcast schemes which blend NWP models and observations at a site, outperform individual NWP models and persistence after 1-2 hours.
Questions?