Results

TWO-YEAR ASSESSMENT OF NOWCASTING
PERFORMANCE IN THE CASA SYSTEM
Evan Ruzanski1, V. Chandrasekar2, and Delbert Willie2
1Vaisala,
Inc., Louisville, Colorado, USA
2Colorado State University, Fort Collins, Colorado, USA
July 27, 2011
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Introduction
• The Collaborative and Adaptive Sensing of the
Atmosphere radar network uses nowcasting in a
distributed closed-loop system
– For emergency decision support (1–20 min lead time)
– For radar scanning adaptation (1–5 min lead time)
• Can nowcasting be done operationally in a
geographically distributed processing
environment?
– Fast radar data update cycles (1 min)
– An efficient algorithm is needed for motion estimation
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Introduction
Data Fusion / Algorithm
Meteorological
Command and Control
steers radars to scan
when and where user
needs are greatest
MC&C
Radar Network
Users
NWS forecasters,
emergency managers,
researchers
Overarching CASA - Distributed Collaborative Adaptive Sensing (DCAS) Concept
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Introduction
NWS forecasters
evaluate CASA
data
E. Bass and B. Philips, CASA researchers
CASA data and products being used at the NWS Norman, Oklahoma, forecast
office
Intro DARTS AlgorithmSystem Architecture
Data
Results Conclusions
DARTS Algorithm



F ( x, y, t )  U ( x, y ) F ( x, y, t )  V ( x, y) F ( x, y, t )  S ( x, y, t )
t
x
y
kt FDFT (k x , k y , kt ) 

Ny
 1  Nx
 U DFT (k 'x , k ' y ) 

   
  k x  k 'x  FDFT (k x  k 'x , k y  k ' y , kt )


N
N
T
/
T
 x y  k 'x  N x k ' y  N y 
x
t



Discretize the
general continuity
equation using FFT
Ny
 1  Nx
 VDFT (k 'x , k ' y ) 

   
 k y  k ' y FDFT (k x  k 'x , k y  k ' y , kt )


Ty / Tt
 N x N y  k 'x  N x k ' y  N y 


 i

 2

 Tt SDFT (k x , k y , kt ) 



y  Hx
x  H y
E. Ruzanski, V. Chandrasekar, and Y. Wang, “The CASA
Nowcasting System,” J. Atmos. Oceanic Technol., vol.
28, no. 5, pp. 640–655, 2011.
Formulate linear
system
Solve linear
system; recover
motion estimates
using IFFT
Intro DARTS Algorithm System Architecture Data
Results Conclusions
System Architecture
Nowcasting
system operation
Data processing/nowcasting software
development and evaluation
Real-time data
transfer via LDM
The CASA radar network
(KSAO, KCYR, KLWE, KRSP)
System Operations Control Center
(SOCC) and Meteorological
Command and Control (MC&C)
Intro DARTS Algorithm System Architecture Data
Results Conclusions
System Architecture
SOCC
Merge and Grid
MC&C
Radial reflectivity cuts from
each CASA radar node
LDM
Nowcasting
(DARTS)
Internet Display
Ingest
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Data
• Verification was done using reflectivity data from
24 weather events collected during the CASA IP1
experiments Feb. 2009–May 2010
– Avg. duration of each event was ~3 hrs. (total ~95 hrs.)
– Data set includes a wide range of precipitation types
(super-cellular, quasi-linear, multi-cellular events)
• Ground clutter filtering and attenuation
correction were applied at each radar node
• Data were gridded to 1-km AGL CAPPIs covering
a +/- 70 km area with avg. resolution of 0.5 km/1
min. using a 20 dBZ threshold
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Results
Example CASA observation and corresponding 10-min. prediction (web display)
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Results
Example CASA observation and corresponding 10-min. prediction sequences
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Results
(a) CSI, (b) POD, (c) FAR, (d) MAE scores for 2009–2010 events
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Results
• Forecaster feedback on adaptive scanning was
positive
– 1 min update rates is important
• Steering using the latest observation vs
nowcasting has drawbacks
– Sector scans can be too narrow
– Important areas of the storm are missed
• Forecaster feedback suggested steering using
nowcasting eliminated sector scanning issues
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Results
Leading edge
missed without
nowcasting
Storm motion
Leading edge
observed with
nowcasting
support
Storm motion
MC&C observation 20090517-0246005 showing steering using previous
observations (left) vs steering using previous observations and 5-min.
DARTS nowcasts (right). The leading edge of the storm cut-off on the left.
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Conclusions
• Nowcasting has been successfully demonstrated
in the CASA system
– Nowcasting 0–20 min is beneficial for emergency
decision-making support
– Nowcasting 1–5 min is used to set up the radar
network scanning strategy
• Computational efficiency is a key concern given
the high resolution of the data and distributed
nature of the system
– The DARTS algorithm estimates storm motion using
LLSE in the Fourier domain
Intro DARTS Algorithm System Architecture
Data
Results Conclusions
Conclusions
• Approximately 95 h (5700 frames) of data from
Feb. 2009–May 2010 were used for evaluation
• Quantitative and qualitative scores were
favorable
– CSI, POD, FAR and MAE scores showed nowcasting
consistently outperformed a persistence forecast
– Forecaster surveys suggested steering using
nowcasting eliminated sector scanning issues