Using a Neural Network Calibration of the SREF PQPF to Increase ...

trainerhungarianΤεχνίτη Νοημοσύνη και Ρομποτική

20 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

83 εμφανίσεις

Using a Neural Network Calibration of the SREF PQPF to

Increase Situational Awareness

By

Jeff Davis, Senior Forecaster

WFO Tucson, Arizona

1.
Goals of this project:


a. Increase awareness of


potential high impact precipitation


events.



b. Begin integrating PQPF concepts


into WFO operations


c. Downscale calibrated SREF PQPF


to a 5km NDFD grid

2. Process used to create calibrated SREF PQPF values

AWIPS SREF Ensemble


Data

JOONE

Java Object Oriented Neural Engine API


-

Open source neural network API used to process SREF


data and determine relationships between the key


parameters.

-

The neural engine can determine these


relationships with a couple of years of data.

-

These relationships are divided into a winter


(Oct 1 to Mar 31) and summer (Apr 1 to Sept 30) season.

-

Ground truth is determined from the RTMA.

Visualization for Algorithm Development


Java API


-

VisAD

Java API used to extract SREF values from an


AWIPS
netCDF

file

-

Neural Network 3 hourly SREF Input

-

Mean
precipitable

water

-

5km gridded elevation

-

Probability of CAPE > 500

-

Forecast hour

-

PQPF for each threshold


(PQPF> .01, >.25,>.50, > 1.0)

-

Separate neural network for each


model cycle

3. Output

February 2, 2010

January 19, 2010

Threat key:

None


PQPF > 0.01 with
Prob

< 10% in any one time
period

Low


PQPF > 0.01 with
Prob

> 10% in more than 4 periods

Moderate


PQPF > 0.25 with
Prob

> 40% in 2 or more periods

High


PQPF > 0.25 with
Prob

> 40% in 4 or more periods


or


PQPF > 0.50 with
Prob

> 20% in 2 or more periods