Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative...
-
Upload
milo-parks -
Category
Documents
-
view
215 -
download
0
Transcript of Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative...
![Page 1: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/1.jpg)
1
Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast
Jordan GerthCooperative Institute for Meteorological Satellite StudiesUniversity of Wisconsin at Madison
Contributions byRobert AuneAdvanced Satellite Products BranchNational Environmental Satellite, Data, and Information ServiceMadison, Wisconsin
![Page 2: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/2.jpg)
2
Defining Sky Cover•The National Weather Service (NWS) web
site defines “sky cover” as “the expected amount of opaque clouds (in percent) covering the sky valid for the indicated hour.”
•No probabilistic component.•No definition of “opaque cloud” or “cloud”.•The implication is cloud coverage of the
celestial dome (all sky visible from a point observer).
![Page 3: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/3.jpg)
3
Cirro
stra
tus
How would the NWS classify sky cover for this scene?
Photo source: National Weather Service JetStream, http://www.srh.noaa.gov/jetstream/synoptic/h7.htm
![Page 4: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/4.jpg)
4
Defining Sky Cover
•Meteorologists have requested a new sky cover product because1. purely numerical weather prediction
(NWP) methods do not account for radiative transparency of clouds and
2. current satellite-based techniques and algorithms, used alone or in existing analysis tools, do not match the forecasters’ interpretation of the NWS-established definition.
![Page 5: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/5.jpg)
5
Questions to Answer•How are cloud scenes represented in a
NWP model? How does that compare to the National Digital Forecast Database (NDFD)?
•How are cloud scenes observed on satellite imagery and represented on satellite products?
•What is the predictive skill of NWP model cloud fields? How can that be improved?
![Page 6: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/6.jpg)
6
Defining Sky CoverHow are cloud scenes represented in a numerical weather prediction (NWP) model? How does that compare to the NDFD?
![Page 7: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/7.jpg)
7
Sky Cover Representation•Xu and Randall (1996) developed the semi-
empirical cloud fraction formulation found in the Weather Research and Forecast (WRF) model.▫Cloud fraction varies exponentially according
to the grid-averaged condensate mixing ratio.▫When the grid-averaged relative humidity is
100%, so is the cloud fraction.▫Total cloud fraction is produced through
averaging the maximum fractions from three primary layers (642 hPa, 350 hPa, 150 hPa).
![Page 8: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/8.jpg)
8
Sky Cover RepresentationNAM Analysis NDFD 1-hour Forecast
Both images valid at 00 UTC on 3 October 2012
Projected onto a 20 km Lambert Conformal grid
![Page 9: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/9.jpg)
9
Defining Sky CoverHow are cloud scenes observed on satellite imagery and represented on satellite products?
![Page 10: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/10.jpg)
10
Sky Cover RepresentationIR Window, Band 8 (11.0 µm)
Effective Cloud Amount
Both images valid at 00 UTC on 3 October 2012
A measure of the opacity (“radiative thickness”) of a cloud
GOES-14/15 (East/West) Sounder Composite
![Page 11: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/11.jpg)
11
Sky Cover RepresentationCRAS Analysis Effective Cloud Amount
Both images valid at 00 UTC on 3 October 2012
CIMSS Regional Assimilation System
![Page 12: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/12.jpg)
12
Defining Sky CoverWhat is the predictive skill of NWP model cloud fields?How can that be improved?
![Page 13: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/13.jpg)
13
Performance (3—24 Sept. 2012)
2012
-09-
03-0
0
2012
-09-
04-0
0
2012
-09-
05-0
0
2012
-09-
06-0
0
2012
-09-
07-0
0
2012
-09-
08-0
0
2012
-09-
09-0
0
2012
-09-
10-0
0
2012
-09-
11-0
0
2012
-09-
12-0
0
2012
-09-
13-0
0
2012
-09-
14-0
0
2012
-09-
15-0
0
2012
-09-
16-0
0
2012
-09-
17-0
0
2012
-09-
18-0
0
2012
-09-
19-0
0
2012
-09-
21-1
2
2012
-09-
22-1
2
2012
-09-
23-1
2
2012
-09-
24-1
20
10
20
30
40
50
60
70
RMSE for WRF Sky Cover 12-hr Forecasts from 3 through 24 September 2012
CurrentMax Frac
RM
SE
![Page 14: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/14.jpg)
14
Performance (3—24 Sept. 2012)Current WRF Output Maximum Fraction Output
Mean RMSE: 37.5% Mean RMSE: 32.4%
Both images valid at 12 UTC on 3 October 2012
![Page 15: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/15.jpg)
15
Defining Sky CoverWhat is the predictive skill of NWP model cloud fields?How can that be improved?
![Page 16: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/16.jpg)
16
Types of Numerical Tools
•Numerical Weather Prediction▫Mathematical models which predict the
future state of the atmosphere•Statistical Model
▫Relates random variables to other variables•Linear Programming/Optimization (LP)
▫Applies mathematical method to a linear model in an attempt to find the ideal solution Solves an objective function Requires a list of linear relationships
![Page 17: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/17.jpg)
17
Linear Model
• The LP objective is to minimize the absolute difference between the human-produced “truth” (the NDFD 1-hour forecast) and the NWP 12-hour forecast (no NWP analysis available).
![Page 18: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/18.jpg)
18
WRF Total Sky Cover Output
is a point on a two-dimensional grid of the 12-hour sky cover forecast, obtained from a WRF model.
min 𝑤=∑𝑖=1
𝑛
|𝑥 𝑗 𝑎𝑖𝑗+𝑧 𝑗−𝑏𝑖|
![Page 19: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/19.jpg)
19
NDFD Sky Cover 1-hr Forecast
is a point on a re-projected two-dimensional grid of the 1-hour sky cover forecast (analysis) obtained from the human-produced NDFD.
min 𝑤=∑𝑖=1
𝑛
|𝑥 𝑗 𝑎𝑖𝑗+𝑧 𝑗−𝑏𝑖|
![Page 20: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/20.jpg)
20
Linear Model
• is used as truth to calculate the coefficient and scalar adjustments (variables) for each cloud classification (six total at approximately 20% increments).
![Page 21: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/21.jpg)
21
Linear Model
•The value of the adjusted sky cover fraction must fall within physical meteorological values (between 0% and 100%). The value of the coefficient must be nonnegative.
![Page 22: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/22.jpg)
22
Apply Results: Procedure• The process employed in this project for staging and
using background data is as follows:▫ Initialize the WRF model at time t = -12 (hours).▫Obtain the 12-hour forecast from the model initialized
at t = -12.▫Obtain the analysis from t = 0.▫Run linear program to minimize the objective function
comparing the 12-hour forecast and analysis both valid at t = 0.
▫ Initialize the WRF model at time t = 0.▫Apply fixed variable values from the linear program (a
coefficient and scalar) to WRF model output at t = 12, t = 24, and t = 36. Calculate skill.
![Page 23: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/23.jpg)
23
Apply Results: Test Case
Fraction (%)
Coefficient (x)
Scalar(z)
0 ≤ ai < 20 4.4 12.0
20 ≤ ai < 40
0 93.0
40 ≤ ai < 60
1.0 0
60 ≤ ai < 80
1.0 0
80 ≤ ai < 95
1.0 0
95 ≤ ai ≤ 100
1.0 0Applied to forecast output from model initialized at 00 UTC on 3
October 2012
𝒙 𝒋 𝒂𝒊𝒋+𝒛 𝒋
![Page 24: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/24.jpg)
24
Test Case: Local WRF Model
RMSE: 45.0%RMSE: 30.5%
LP produces net RMSE improvement of 14.5%
![Page 25: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/25.jpg)
25
Performance (3—24 Sept. 2012)
2012
-09-
03-0
0
2012
-09-
04-0
0
2012
-09-
05-0
0
2012
-09-
06-0
0
2012
-09-
07-0
0
2012
-09-
08-0
0
2012
-09-
09-0
0
2012
-09-
10-0
0
2012
-09-
11-0
0
2012
-09-
12-0
0
2012
-09-
13-0
0
2012
-09-
14-0
0
2012
-09-
15-0
0
2012
-09-
16-0
0
2012
-09-
17-0
0
2012
-09-
18-0
0
2012
-09-
19-0
0
2012
-09-
21-1
2
2012
-09-
22-1
2
2012
-09-
23-1
2
2012
-09-
24-1
20
10
20
30
40
50
60
70
RMSE for WRF Sky Cover 12-hr Forecasts from 3 through 24 September 2012
CurrentMax FracLP Result
RM
SE
Mean RMSE: 21.7%
Mean RMSE net decrease of
15.8%
![Page 26: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/26.jpg)
26
Improving Results
•Decreased performance is a consequence of:▫Lacking sky cover congruity between
forecast areas on NDFD domain▫Assumption that relative horizontal
distribution of model-developed cloud fraction matches NDFD
▫No qualification for cloud forcing or air mass regimes within NWP model (parameterizations and performance)
▫No corroboration between other NWP model output parameters or variables
![Page 27: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/27.jpg)
27
Cloud Top Pressure (CTP)
Considering the best matches, for which 200 hPa deep layer does the effective cloud amount (ECA) analysis best correspond to the NDFD sky cover 1-hr forecast?
The CTP composite from the GOES Sounders provides a cloud height for each corresponding ECA value.
Image valid at 00 UTC on 3 October 2012
![Page 28: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/28.jpg)
28
Gridded CTP
This can be solved through implementing another linear model:
Objective function is to minimize absolute error between ECA and NDFD.
At least 250 ECA/NDFD matches must be used. This is approximately 2.5% of all points on the grid (but not all points are considered).
Clear points are excluded.The set of points contributing
to the objective function must be within 100 hPa of central pressure level.
Image valid at 00 UTC on 3 October 2012
![Page 29: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/29.jpg)
29
Improving Results• Examined ECA-NDFD
pairs producing the least amount of collective absolute error
• Trial period suggests lower MAE associated with higher cloud tops (thicker cloud?)
• Available CTPs can vary according to weather scenario and subject to Sounder bias
• Mean central pressure over trial period was 437 hPa (layer from 537 hPa to 337 hPa)
• Mean MAE was 1.7 (%)
2012-0
9-2
8-0
0
2012-0
9-2
8-1
2
2012-0
9-2
9-0
0
2012-0
9-2
9-1
2
2012-0
9-3
0-0
0
2012-0
9-3
0-1
2
2012-1
0-0
1-0
0
2012-1
0-0
1-1
2
2012-1
0-0
2-0
0
2012-1
0-0
2-1
2
2012-1
0-0
3-0
0
2012-1
0-0
3-1
2
2012-1
0-0
4-0
0
2012-1
0-0
4-1
2
100
200
300
400
500
600
700
800
900
1000 0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Central Pressure and Mean Absolute Error for 28 Sept. to 4 Oct. 2012
Pressure MAE
Pre
ssure
(hP
a)
MA
E
![Page 30: Linear Optimization as a Solution to Improve the Sky Cover Guess, Forecast Jordan Gerth Cooperative Institute for Meteorological Satellite Studies University.](https://reader035.fdocuments.in/reader035/viewer/2022062423/5697bf911a28abf838c8e8fb/html5/thumbnails/30.jpg)
30
Summary of Results•Sky cover definitions are inconsistent. Model
output needs major adjustment to match the NDFD result. The NDFD needs more congruity.
•Linear optimization of model output leads to a reduction in RMSE for subsequent forecasts.
•ECA skill is in question for verifying the NDFD sky cover grid in low-topped cloud regimes.
•Future work will focus on building a celestial dome and time-averaging new geostationary satellite cloud products over a one-hour window.
Author information: Jordan Gerth, [email protected]