1. Mike Halpert NOAA-NWS-Climate Prediction Center October, 2011
Climate Prediction Center Outlooks (ERF/LRF): Basis, Tools, Verification
Mike Halpert
NOAA-NWS-Climate Prediction Center
October, 2011

2. Climate is Constructed from Weather
Wildly oscillating curve = daily “weather”
Smooth curve = 30 year mean (climatology)
Weather and climate are different parts of the same thing and are dependent upon one another.

3. Objectives Understand the science behind CPC extended-range
and long-range forecasts
Understand the proper interpretation and limitations
of CPC’s operational forecasts
Obtain the ability to conduct basic interviews about
CPC’s forecasts, including interpretation, science
behind, and verification (skill)
Weather and climate are different parts of the same thing and are dependent upon one another.

4. Outline Part I – Long-Range Forecasts
Sources of LRF predictability
Meaning (interpretation) of the forecasts
Operational forecast tools and example

5. Sources of S-I Predictability
ENSO – Walker and Bliss (1932), Bjerknes (1969), Rasmussen and Carpenter (1982)…….
Trend (Huang (1996), van den Dool (2006)…..
Ocean-Atmosphere-Land (van den Dool (2006)…..
Statistically-derived signals of unknown origin (Barnston, 1994)
Dynamical model-derived signals (Saha et al, 2006)

6. Outline Part I – Long-Range Forecasts
Sources of LRF predictability
Meaning (interpretation) of the forecasts
Operational forecast tools and example

7. U. S. Seasonal Outlooks March 2011 - May 2011
Temperature
Precipitation
What is the interpretation of these forecasts? What is the probability of above-average temperatures in central Texas; near-average?; below-average?
a) 50,30,20
b) 50,33,17
c) 50,25,25
d) 50,50, 0

8. U. S. Seasonal Outlooks March 2011 - May 2011
Temperature
Precipitation
N. Georgia
Below: 29%
Near: 33%
Above: 38%
Minnesota
Below: 33%
Near: 33%
Above: 33%
What is the interpretation of these forecasts? What is the probability of above-average temperatures in central Texas; near-average?; below-average?
a) 50,30,20
b) 50,33,17
c) 50,25,25
d) 50,50, 0

9. Outline Part I – Long-Range Forecasts
Sources of LRF predictability
Meaning (interpretation) of the forecasts
Operational forecast tools and example (DJF )

10. Seasonal Forecasts
Which of the following factors influence the seasonal forecast (select all that apply):
A) Trends – 91%
B) Soil Moisture – 59%
C) El Niño/Southern Oscillation – 95%
D) Atmospheric Noise – 18%

11. FACTORS INFLUENCING A CLIMATE FORECAST
Climate Change - trends
Natural Climate Variability – “organizes” weather
El Niño-Southern Oscillation (ENSO)
Mid-latitude Oscillation modes (NAO, AO, PNA, …)
Land Surface Processes (Soil moisture, Snow cover, …)
Atmospheric Noise - unpredictable “climate” signals produced by chance through cumulative effects of weather. This is large in middle latitudes, small in the Tropics. Major cause of “uncertainty” in forecasts.

12. Optimal Climate Normal (OCN)
OCN, as it is used as a tool at CPC is, quite simply, a measure of the trend. For a given station and season, the OCN forecast is the difference between the seasonal mean (median) temperature (precipitation) during the last 10 (15) years and the 30 year climatology.
Much of CPC’s skill has historically been derived from OCN.

13. 30 year WMO normals: ; ; etc OCN = Optimal Climate Normals: Last K year average. All seasons/locations pooled: K=10 is optimal (for US T). Forecast for Jan = (Jan02+Jan Jan11)/10. – WMO-normal plus a skill evaluation for some 50+ years. Why does OCN work? 1) climate is not constant (K would be infinity for constant climate) 2) recent averages are better 3) somewhat shorter averages are better (for T) see Huang et al J.Climate. 9,

14. G H C N - A M S F 2 8

15. OCN DJF
Data through 2010
Data through 2011

16. Canonical Correlation Analysis (CCA)
CCA is a statistical technique relating tropical Pacific Ocean sea-surface temperatures (SSTs), 700 hPa heights, (the predictors) and U.S. surface temperatures (T) and precipitation (P) (the predictands)
When CCA is developed, relationships are found between observed U.S. T and P for a given season, say, January-February-March (JFM) and the predictors for the prior four non-overlapping seasons, in this case, OND, JAS, AMJ and JFM of the prior year.

17. CCA DJF
Temperature
Precipitation

18. CFS Niño 3.4 Forecast
PDF-corrected CFSv1
CFS21

19. CPC Official SST Forecast

20. CFS skill

21. Pacific Niño 3.4 SST Outlook
An increasing number of ENSO models predict the continuation of La Niña into the Northern Hemisphere winter (Niño-3.4 SST anomalies less than -0.5°C).
Figure provided by the International Research Institute (IRI) for Climate and Society (updated 19 October 2011).

22. CFS DJF Outlook °C
mm/day
Climate Forecast System version 2 – Ensemble average of 40 members from October Base period for climo is Forecast skill in gray areas is less than 0.3

23. NMME DJF Outlook
National Multi-Model Ensemble – Average of 7 models (CFSv1, CFSv2, NCAR, GFDL, NASA, IRI and (ECHAMA, ECHAMF) from October 2011.

24. Soil Moisture Largest impacts during summer season
Largest potential when extremes exist (flooding/drought).
Dry conditions have positive impact on T; negative impact on P
Wet Conditions have negative impact on T; positive impact on T

25. SMLR
CCA
OCN
LAN
CFSV1
LFQ
ECP
IRI
ECA
CON
La Nina Composites:

26. Seasonal Forecasts
Which of the following factors influence the seasonal forecast (select all that apply):
A) Trends
B) Soil Moisture
C) El Niño/Southern Oscillation
D) Atmospheric Noise

27. Questions

28. Outline Part II – Extended-Range Forecasts
Meaning of the forecasts
Scientific basis of ERF forecasts
Forecast Tools / Operational forecast example

29. Extended Range Forecasts
E. Montana
Below: 32%
Near: 36%
Above: 32%
E. Nebraska
Below: 42%
Near: 33%
Above: 25%
E. Alaska
Below: 4%
Near: 33%
Above: 63%

30. Outline Part II – Extended-Range Forecasts
Meaning of the forecasts
Scientific basis of ERF forecasts
Forecast Tools / Operational forecast example
Verification/Skill

31. Basis for Forecasts 31

32. Forecast Process Schematic
Recent observations
Dynamical model forecasts/multi- model ensembles
Historical observations.
Verifications/Statistical tools Downscaling, Analogs, Composites
Subjective weighted average 500-hPa height and anomaly forecast (BLEND).
Downscale to create surface temperature and precipitation tools using BLEND input.
Subjective formulation of the probability of temperature and precipitation tercile categories.
Write the forecast bulletin, FEUS40.
Generic forecast process. Basic elements of ANY forecast system.
Dissemination to public between 3-4 PM Eastern Time

33. Outline Part II – Extended-Range Forecasts
Meaning of the forecasts
Scientific basis of ERF forecasts
Forecast Tools / Operational forecast example

34. Question
Which of the following tools is not used in the preparation of the extended-range forecasts?
Analogs – 5%
Trends – 5%
Regression – 14%
Calibration – 72%

35. Forecast tools DYNAMICAL MODELS STATISTICAL TOOLS (Downscaling)
Global Forecast System (GFS) and ensembles
European Centre for Medium-range Weather Forecasts (ECMWF) ensembles
Canadian ensembles
STATISTICAL TOOLS (Downscaling)
Klein T – screening regression
ESRL calibrated T, P – calibrates recent model frequencies with atmos.
NAEFS – Bias-corrected ensemble forecasts – T, P
GFS P, T – Dynamical model output– calibrated P, T
Analog composites – Average T, P for the 10 best 500-hPa analogs
Tele-connections – Simultaneous, significant temporal correlations for two or more widely separated locations.

36. 500-hPa Heights
Forecast made: 1/31
Valid: 2/6-2/10

37. 500-hPa Height Anomalies
Forecast made: 1/31
Valid: 2/6-2/10

38. Blended 500-hpa Height/Anomalies
ECMWF ENS MEAN – 10%
Canadian ENS MEAN – 10%
GFS Superensemble – 40%
0Z GFS ENS MEAN – 10%
6Z GFS ENS MEAN – 10%
0Z Operational – 10%
6Z Operational – 10%
Forecast made: 1/31
Valid: 2/6-2/10

39. Downscaled Temp/Prec Probabilities
Analogs (T/P)
Klein Equations (T)
NAEFS
Calibrated Model Output (T/P)
ESRL (CDC) Reforecasts (T/P)

40. 500-hPa Analog to the GFS Z500 Superensemble Mean
500-hPa Height/Anomaly centered on 02/27/2009
0.87
500-hPa Analog to the GFS Z500 Superensemble Mean
Forecast made: 1/31
Valid: 2/6-2/10

41. Temperature Analog based on previous analog
Forecast made: 1/31
Valid: 2/6-2/10

42. Precipitation Analog based on previous analog
Forecast made: 1/31
Valid: 2/6-2/10

43. Stepwise Forward Screening Regression
Given a set of inputs, xj, and an output, y,
Start with all coefficients, bj= 0
Find the predictor, xi (700-hPa height) most correlated with y (Surface temperature). Include this predictor in the model. Find residuals
Continue in this manner, at each stage, adding the predictor most correlated with r to the model.
Stop when a threshold minimum correlation with r is reached (typically 4 or less terms).

44. Klein T 0Z GFS Superensemble Mean
Forecast made: 1/31
Valid: 2/6-2/10

45. Klein Ensemble Percentages
GFS
ECMWF
Forecast made: 1/31
Valid: 2/6-2/10

46. 2m Temperature – GFS ENS Mean
Uncalibrated
Calibrated
Forecast made: 1/31
Valid: 2/6-2/10

47. Precipitation - GFS Calibrated ENS Mean Percentages
Forecast made: 1/31
Valid: 2/6-2/10
Forecast made: 2/19
Valid: 2/25-3/1

48. North American Ensemble
Forecast System
Weather modeling system run by NWS and CMC
Multi-model ensemble that combines the global forecast
model ensemble and the Canadian global forecast model
ensemble.
Bias corrected using forecasts and observations over the
Past 120 days using a decaying average mean error.

49. NAEFS Forecast (T/P) Probabilities
Forecast made: 1/31
Valid: 2/6-2/10

50. Question
Which of the following tools is not used in the preparation of the extended-range forecasts?
Analogs
Trends
Regression
Calibration

51. Automated Forecast
Forecast made: 1/31
Valid: 2/6-2/10

52. Official Forecast
Forecast made: 1/31
Valid: 2/6-2/10

53. Observed Temp/Precipitation
Official Auto – 61.5
Official Auto – 25.8
Forecast made: 1/31
Valid: 2/6-2/10

54. Questions

55. Outline Part III – Verification
1. Modified Heidke Skill Score
2. Long Lead Forecasts
3. Extended Range Forecasts
4. Comparison

56. Verification
Verification of temperature/precipitation outlooks done on 2x2 grid for CONUS.
This encompasses 232 valid grid squares.
Main statistic used is the modified Heidke Skill Score, although other statistics are also calculated (RPSS, .

57. Modified Heidke Skill Score: % Improvement over Random Forecasts
c = # correct forecasts
t = # total forecasts
e = # correct randomly (climatology)

58. 2° x 2° Grid
20° - 56°N, 130° - 60°W, 232 valid points

59. Outline Part III – Verification
1. Modified Heidke Skill Score
2. Long Lead Forecasts
3. Extended Range Forecasts
4. Comparison

60. Temperature Skill Scores Long term actual
Mean = 22.3, Coverage = 50.9%

61. After the fact…..
Source Peitao Peng

62. Precipitation Skill Scores Long term actual
Mean = 10.9, Coverage = 31.4%

63. Regional Verification DJF/MAM
DJF - Temp
MAM - Temp
DJF - Prec
MAM - Prec

64. Regional Verification JJA/SON
JJA - Temp
SON - Temp
JJA - Prec
SON - Prec

65. Extended Range Temperature Verification
Mean = High = Low = -31.3
Mean = High = Low = -37.9

66. Extended Range Precipitation Verification
Mean = High = Low = -27.2
Mean = High = Low = -31.1

67. Extended Range Regional Verification - Temperature

68. Extended Range Regional Verification - Precipitation

69. Question
Which response ranks the skill of CPC temperature forecasts from highest to lowest?
Seasonal, 6-10, 8-14, 0.5 monthly – 0%
6-10, 8-14, 0.5 monthly, seasonal – 90%
Seasonal, 0.5 monthly, 6-10, 8-14 – 0%
6-10, seasonal, 8-14, 0.5 monthly – 10%

70. FY07-FY10 Mean Heidke Skill Scores
Outlook Period
Temperature
Precipitation
6-10 Day
29.2 (22.3)
13.8 (13.3)
8-14 Day
20.9
8.1
30 Day (0.5 Mo Lead)
14.8 (non-EC), 06.7 (All)
09.3 (non-EC), 02.7 (All)
30 Day (0.0 Mo Lead)
30.7 (non-EC), 13.9 (All)
24.1 (non-EC), 09.0 (All)
90 Day
22.0 (non-EC), 13.2 (All)
13.2 (non-EC), 04.3 (All)

71. Percent of “Successful” CPC Forecasts

72. Questions

73. Seasonal Predictions Lab
Mike Halpert
NOAA/NWS Climate Prediction Center

74. Scenario
You are called by noted New York Times science writer Andrew Revkin, who asks you if he can do a brief interview with you about climate forecasting on time scales from weeks to seasons. He says he wants to focus specifically on the techniques, meaning, and verification of the Climate Prediction Center’s extended range and long range forecasts. He says he wants to focus and CPC’s seasonal forecast for winter and the extended range forecasts that were issued late in December He has provided you the attached 6 questions: