thoughts on assessing decadal precipitation variations as surrogate forecasts
DESCRIPTION
THOUGHTS ON ASSESSING DECADAL PRECIPITATION VARIATIONS AS SURROGATE FORECASTS. Jeanne M. Schneider USDA Agricultural Research Service Grazinglands Research Laboratory El Reno, OK. - PowerPoint PPT PresentationTRANSCRIPT
THOUGHTS ON ASSESSING DECADAL PRECIPITATION
VARIATIONS AS SURROGATE FORECASTS
THOUGHTS ON ASSESSING DECADAL PRECIPITATION
VARIATIONS AS SURROGATE FORECASTS
Jeanne M. SchneiderUSDA Agricultural Research Service
Grazinglands Research Laboratory
El Reno, OK
Jeanne M. SchneiderUSDA Agricultural Research Service
Grazinglands Research Laboratory
El Reno, OK
But first, a quick review/tutorial on probability of exceedance functions….which illustrates why someone might
want to use them.
But first, a quick review/tutorial on probability of exceedance functions….which illustrates why someone might
want to use them.
0
2
4
6
8
10
12
1970 1975 1980 1985 1990 1995 2000
May Precipitation at Kingfisher OK 1971-2000
Year
30 years of data
0
2
4
6
8
10
12
1970 1975 1980 1985 1990 1995 2000
May Precipitation at Kingfisher OK 1971-2000
Year
30 years of data
average = 4.91"
0
2
4
6
8
10
12
Ranked May Precipitation at Kingfisher OK 1971-2000
"Below"Tercile
"Normal"Tercile
"Above"Tercile
33% 33% 33%Odds to be in tercile:
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10 12
May Precipitation at Kingfisher OK 1971-2000
Probability of Exceedance (%)
Precipitation (inches)
Probability of Exceedance = 1 - (cumulative probability density function)
"a posteriori"
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10 12
May Precipitation at Kingfisher OK 1971-2000
Precipitation (inches)
Probability of E
xceedance (%
)
67%
3.2”
“Two in three chance for more than 3.2 inches.”
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10 12
May Precipitation at Kingfisher OK 1971-2000
Precipitation (inches)
Probability of E
xceedance (%
)
67%
3.2” 4.2”
50%
“Two in three chance for more than 3.2 inches.”
“Fifty-fifty chance for more than 4.2 inches.”
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10 12
May Precipitation at Kingfisher OK 1971-2000
Precipitation (inches)
Probability of E
xceedance (%
)
67%
33%
3.2” 6.1”4.2”
50%
“Two in three chance for more than 3.2 inches.”
“Fifty-fifty chance for more than 4.2 inches.”
“One in three chance for more than 6.1 inches.”
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10 12
May Precipitation at Kingfisher OK 1971-2000
Precipitation (inches)
Probability of E
xceedance (%
)
67%
33%
3.2” 6.1”4.2”
50%
“Two in three chance for more than 3.2 inches.”
“Fifty-fifty chance for more than 4.2 inches.”
“One in three chance for more than 6.1 inches.”
If you can associate a potential financial loss with each of these outcomes, then you have a definition of “risk”.
So, what do we plan to do when the NOAA/CPC forecasts offer nothing beyond
the 30-year climatology ("EC")?
Or when the forecast skill is so low as to preclude practical use?
Do we have any other options for climate-conditioned decision support for agriculture in
areas where ENSO impacts are marginal?
So, what do we plan to do when the NOAA/CPC forecasts offer nothing beyond
the 30-year climatology ("EC")?
Or when the forecast skill is so low as to preclude practical use?
Do we have any other options for climate-conditioned decision support for agriculture in
areas where ENSO impacts are marginal?
Continuing where we left off last year….Continuing where we left off last year….
Annual Precipitation in Central Oklahoma
Year
Precipitation [in] Annual Precipitation
5-yr weighted average
Dry PeriodsWet Periods
CD3405; 1895-200415
25
35
45
55
1895 1915 1935 1955 1975 1995
0
0.2
0.4
0.6
0.8
1
0 1 2 3 4 5 6 7 8 9 10 11
Dry periods1900-19041909-19171936-19391952-19561963-19721976-1980
Wet periods1905-19081941-19451957-19611982-2000
Probability of Exceedance of September PrecipitationClimate Division 3405; Central Oklahoma; 1895-2005
Pro
b.
of
Exc
eed
ance
Precipitation [in/mo]
Dry PeriodsMean = 3.2 [in]
38 Years
Wet PeriodsMean = 4.6 [in]
33 Years
1895-2005Mean = 3.7 [in]
111 Years
Preliminary data, subject to revision
USDA-ARS-GRL
Assume that the earth/ocean/atmosphere
system behaves as a chaotic system with a
very large number of degrees of freedom, but
that the net expression of that chaos on the
precipitation processes at any location can be
usefully described as a small collection of
states (dry, wet, perhaps transition), where
each state persists for several years.
Assume that the earth/ocean/atmosphere
system behaves as a chaotic system with a
very large number of degrees of freedom, but
that the net expression of that chaos on the
precipitation processes at any location can be
usefully described as a small collection of
states (dry, wet, perhaps transition), where
each state persists for several years.
A faint flavor of chaotic dynamics:A faint flavor of chaotic dynamics:
Climate changes:
To what degree does the past predict the future?
Data is incomplete and imperfect:
if a predictive signal exists in the phenomena,
can we discern it in the record?
Where, when, and how do we search for the signal?
And on and on, ad nauseum….
Climate changes:
To what degree does the past predict the future?
Data is incomplete and imperfect:
if a predictive signal exists in the phenomena,
can we discern it in the record?
Where, when, and how do we search for the signal?
And on and on, ad nauseum….
But there are problems with statistical approaches to climate forecasts:
But there are problems with statistical approaches to climate forecasts:
However, as has been noted by several presenters already:
"….decisions have to be made".
However, as has been noted by several presenters already:
"….decisions have to be made".
We need monthly, location specific guidance relative to precipitation for producers,
agricultural extension agents, and others involved with small to medium scale
agriculture.
To be useful in the near term, that guidance needs to be built using existing data and tools.
We need monthly, location specific guidance relative to precipitation for producers,
agricultural extension agents, and others involved with small to medium scale
agriculture.
To be useful in the near term, that guidance needs to be built using existing data and tools.
The good news is that we have a couple of simplifying constraints:The good news is that we have a couple of simplifying constraints:
Can we, or can we not, produce probabilistic
monthly precipitation guidance for specific
locations that is more reliable than the
standard 30-year climatology,
given data and tools currently in hand?
Can we, or can we not, produce probabilistic
monthly precipitation guidance for specific
locations that is more reliable than the
standard 30-year climatology,
given data and tools currently in hand?
The Acid Test:The Acid Test:
Klaus Wolter'sExperimental Climate Divisions
Klaus Wolter'sExperimental Climate Divisions
http://www.cdc.noaa.gov/people/klaus.wolter/ClimateDivisions/http://www.cdc.noaa.gov/people/klaus.wolter/ClimateDivisions/
Klaus Wolter'sExperimental Climate Divisions
Klaus Wolter'sExperimental Climate Divisions
http://www.cdc.noaa.gov/people/klaus.wolter/ClimateDivisions/http://www.cdc.noaa.gov/people/klaus.wolter/ClimateDivisions/
Because these are based on precipitation variability, I will use these experimental climate divisions to define
"location specific".
Century-scale monthly data from PRISMCentury-scale monthly data from PRISM
http://www.prism.oregonstate.edu/http://www.prism.oregonstate.edu/
Select a grid point near the center of each experimental forecast division, and use the 103-year long PRISM
time series data to define the decadal variations in precipitation.
I will test different lengths of base record, de-trending options, and
definitions of state and persistence.
Use individual station data to test the reliability
of both the decadal and 30-year climatology
as probabilistic forecasts over the last decade.
Given the short period, use as many stations
as possible within each experimental forecast
division for the assessment.
Use individual station data to test the reliability
of both the decadal and 30-year climatology
as probabilistic forecasts over the last decade.
Given the short period, use as many stations
as possible within each experimental forecast
division for the assessment.
NOAA/NCDC Coop Station DataNOAA/NCDC Coop Station Data
Conceptually, this is so simple it can be done graphically for each month and
experimental forecast division.But realistically,
we will build assessment PoEsfrom the station data, and adapt or
develop measures of relative reliability.
Conceptually, this is so simple it can be done graphically for each month and
experimental forecast division.But realistically,
we will build assessment PoEsfrom the station data, and adapt or
develop measures of relative reliability.
More next year….